I’ve seen a lot of people call themselves AI and GTM operators, but they’re just using Claude.

But Jerry O’Shea is one of the few true operators who build GTM engines for orgs using AI.

He’s the founder of Not Your Average BDR, and he builds systems that capture how a company sells, package that knowledge into a Company OS, and turn buyer signals into outbound that sounds like it came from someone who knows the account.

This week, Jerry walked me through an AI SDR workflow that starts the moment someone visits his website and ends with a ready-to-send email and LinkedIn sequence.

This week’s Stack

• 1 video: Build an AI SDR that writes on-brand outbound

• 1 prompt: Create the ICP brain behind better AI workflows

• 1 tool: Revisit Manus for multi-step AI execution

• 3 resources: Smarter Claude setups, SEO skills, and AI ad workflows

• 3 jobs: AI marketing and GTM engineering roles worth a look

Workflow Walkthrough: The AI SDR That Writes On-Brand Outbound

Most AI SDR tools or GTM-engineered workflows send a lot of bad email fast. They don’t know how you sell, so they pattern-match to the same generic opener everyone else is running. Prospects can smell it in half a second.

This week’s guest, Jerry O’Shea (founder of Not Your Average BDR), went the other way. He built a “Company OS,” then wired it into a Clay workflow that fires the moment a prospect hits his site. It researches the account, writes a sequence that adapts to what’s true about that company, and loads a ready-to-send email plus LinkedIn play into his sequencer via Lemlist.

This is true AI and GTM Engineering that companies aspire to have.

Here’s what you’ll learn:

• How to capture your ICP, product, and positioning once so every message sounds like you instead of a template.

• Why a single website visitor is the trigger that kicks off the whole workflow automatically.

• The “loose prompt” approach that lets each message match the account’s real signals, so you stop blasting one stale theme at everyone.

• How to run email and LinkedIn from one sequence with conditional logic (LinkedIn invite first, email fallback if they don’t accept after nine days).

• Where n8n fits, and why routing through it beats trying to force everything into a single tool.

If you’re getting started with Clay and need extra credits to experiment, sign up with my referral link and you’ll get 3,000 free credits.

Here’s Jerry’s Company OS Interview Skill.

Get the skill here

Prompt of the Week: Build the ICP Builder for Your AI SDR

Most AI SDRs and GTM-engineered workflows for SDRs suck because they send more emails but lack the context needed to sell effectively.

That’s why I loved Jerry’s workflow. Before a single outbound message is written, he captures everything about his business in a Company OS: who they sell to, why customers buy, how they position themselves against competitors, and the language that makes prospects respond. Once that foundation exists, every workflow gets smarter.

If you’re building your own AI SDR (or even if you want better prompts across marketing and sales), this ICP builder prompt is one of the best places to start. Spend an 20-30 min building this once, and every email, campaign, landing page, and sales workflow you create afterward gets better because your AI finally has the context it’s been missing.

I usually post the prompt right here, but this one is a little too long, so I’m going to link you straight to my Prompt Library in Notion.

View the Prompt

You can find dozens more prompts like this inside The Ultimate ChatGPT Prompt Library for B2B Marketing Leaders.

Tool of the Week: Manus

I first tried Manus when everyone was talking about it early last year. Like a lot of AI tools, I thought it was interesting, then moved on. Then I was talking with a CMO I respect who told me it’s become one of the tools she uses the most, so I have it another shot.

What I like is that you can hand it a project, let it research, analyze, and execute across multiple steps while you work on something else. If you’ve written it off because of your first impression, it’s worth taking another look.

Try it: https://manus.im

AI Resource Roundup

How to Keep Fable 5 for Free: Ruben Hassid shared one of the better Claude setup prompts I’ve seen. Instead of repeating your preferences in every chat, it walks Claude through your files, interviews you about how you work, and generates a CLAUDE.md file that becomes your persistent operating instructions.

21 SEO Skills Built for Claude: SEO still comes down to repeatable processes. This collection packages everything from keyword research to content audits into reusable Claude skills so you spend less time writing prompts and more time improving content.

Build Better AI Ads with Claude + Replit (Marketing Against The Grain): If paid social is part of your job, this walkthrough shows how to turn Claude and Replit into a repeatable workflow for building LinkedIn, Instagram, and Google Ads.

Hot AI Jobs

This week’s lineup spans GTM engineering, AI-powered marketing, and marketing automation.

GTM Engineer at Scribe

• Location: San Francisco, CA (Hybrid)

• Pay: $127,500 – $185,000

AI Marketing Engineer at Assured

• Location: Remote (US)

• Pay: $180K–$280K

AI Engineer, Marketing at Planet

• Location: Remote (US)

• Pay: Not listed

Before You Bounce

I hope you’re having a great summer. Between vacations, kids being out of school, and trying to keep work moving, this time of year always feels like a balancing act.

I’ve also been enjoying the World Cup. USA is out, so now I get to watch without my blood pressure spiking every match. It’s been a fun tournament, and I’m looking forward to seeing how it finishes.

Go build something this week and hit reply to tell me what you made.

There’s a take going around that I keep running into that I think it’s ridiculous, and it’s that Claude is about to replace Clay. They say “You won’t need the enrichment tool anymore because the model can find the data itself.”

I don’t buy it, and I’ve been running both every day to prove it to myself.

Clay and Claude do different jobs. Clay is the engine: more than a hundred data sources, the waterfalls that try the cheap source before the expensive one, the credit economics, the orchestration that turns a raw list into a clean, scored, enriched table. Claude Code is the operator that drives the engine. Asking a model to replace Clay is like asking it to replace your CRM. Wrong layer.

I don’t hear enough people talking about how to use them together.

I run Clay across a handful of SaaS client workspaces, and I got tired of rebuilding the same tables and re-explaining the same rules every time I opened a new one.

So I packaged the whole thing into a workbench that Claude Code can drive. It’s public and free. Below I’ll walk through what’s in it, the one command that runs it, and a few plays you can steal this week.

One quick note before we get into it. To follow along you’ll need a paid Clay workspace, the official Clay MCP connected, and Claude Code (or Cursor). That’s the whole setup.

What’s in the Clay Repo

It’s a system: 16 sub-skills plus a /clay router that reads what you’re trying to do and sends you to the right one. They’re grouped by the six stages every Clay workflow moves through.

Hygiene – clay-list-clean pre-cleans a raw source list (dedup, domain and title normalization, suppression) before you spend a single credit. – clay-data-hygiene keeps your CRM deduped and refreshed on a schedule instead of rotting between quarters.

Build – clay-abm-list builds account-keyed target lists with the trigger events baked in. – clay-account-research and clay-prospect-research run multi-source briefs on a company or a person (priorities, news, hiring, leadership, a warm-path hypothesis). – clay-enrich-waterfall handles the email, phone, and LinkedIn waterfalls.

Score – clay-icp-score lays in a 10-point fit composite. – clay-buying-signals fuses intent from eight-plus sources with time decay, and it won’t go live until the signal beats a real lift test against your closed-won.

Activate – clay-outbound writes personalized copy behind a Sending Gate and pushes it to your sequencer. – clay-inbound-routing wires a webhook to enrich, route, and Slack the right rep. – clay-signal-monitor watches a trigger and routes what it catches.

Quality and cost – clay-claygent-iterator tests a Claygent prompt before you scale it. – clay-credits forecasts spend and compares provider mixes. – clay-cost-audit runs a portfolio-wide waste check across every workbook. – clay-troubleshoot fixes the broken or expensive ones.

Persistence – clay-template-library lets you save, version, and reuse your best workbooks instead of rebuilding them from memory.

Every one of these enforces the same habits (gates before credits, free sources before paid), so the cheap moves are the default and not the thing you remember to do at the end.

Get the Repo

Try this one command

Here’s what running it looks like. I open Claude Code in the workbench folder and type one thing:

/clay build a Tier 1 ABM list of US Series B fintechs that hired a new VP of Sales in the last 90 days. Score them against my ICP and draft outbound for the top tier.

From there the router takes over. It reads this as an account build plus scoring plus activation, and it runs an eight-question intake first so it builds my workbook and not a generic one. Then it cleans the list, drops an ICP gate column before any paid enrichment touches a row, builds the waterfall free-source first and paid only as fallback, lays in the 10-point score, and writes the outbound behind a Sending Gate so nothing exports until it clears my rules.

What used to be an afternoon of clicking is now a command and a review. I’m approving and editing, not building from scratch. And the structure it stands up is the disciplined version, the one I’d build on my best day, instead of the rushed version I’d put together at 4pm with three other things on fire.

Why this beats doing it by hand

The speed is nice. What saves me money is the discipline it forces.

In my experience, cleaning the list first cuts credit burn by something like a third to a half, because you’re not paying to enrich duplicates and dead rows. Gates before credits means I never spend on a record that was never going to qualify. And before I scale any Claygent column, the iterator runs it on 5 rows, then 25, then 100, and scores each batch for pass rate, fabrication rate, and credits per row. That’s how I catch a made-up data point at row 25 instead of finding it at row 5,000 after I’ve already paid for all of them.

It won’t read your mind, though. The intake is the work, and thin answers build thin workbooks.

Plays you can run this week

A few things people are already doing with it:

• Rebuild your best-performing workbook as a template, then spin it up for a new segment or a new client in minutes instead of an afternoon.

• Stand up an inbound demo router: form fill, enrich, ICP score, and a Slack ping to the right AE in close to real time.

• Point clay-signal-monitor at a trigger you care about (fresh funding, a key exec hire, a tech install showing up) and let it route hot accounts straight into your cadence.

• Generate one-page account briefs for your AEs the morning before a big meeting.

• Run clay-cost-audit across every workbook once a month and cut the ones that aren’t earning their credits.

• Clean a messy CSV before it ever touches a paid column (the cheapest habit on this list, and the one most people skip).

Most of the plays here aren’t original to me. I built on open-source Clay skill packs from forma-norden, mariosworkflows, and the ColdIQ crew, then wired them into one system with the cost guardrails I run for clients.

Do this today (it will take 15 min)

If you want to try it today:

1. Clone the repo into your skills folder: git clone https://github.com/guerrilla2799/clay-workbench ~/.claude/claudecodeskills/clay-workbench

2. Connect the official Clay MCP in your connector settings.

3. Run /clay against one real goal you have this week, not a test list.

4. Start small. Let it build a single list, then check the ICP gate column before you spend a credit.

It runs MCP-first, but if you’d rather drive Clay yourself the first time, it falls back to a manual UI walkthrough and tells you what to click.

It’s public and free at github.com/guerrilla2799/clay-workbench. Clone it, run one command against a real list, and send me what you build. I want to see it.

And if this is the kind of AI-in-the-GTM-stack workflow you want more of, I break one down every week in Stack & Scale.

– Brandon

I built an AI CMO, but before you roll your eyes, it can’t do my job, and that’s the entire point. You cannot replace a marketing leader with a model. I believe that. I also spent two weeks building an AI CMO. Both are true, and the difference between them is one of the most useful things I’ve learned about AI this year.

There’s been a lot of talk about an AI CMO right now, and there’s also been a lot of people dunking on them. You’re not going to replace a marketing leader’s judgment, taste, or relationships with an LLM. You can’t replace the strategy calls, the founder discussions, the intuition on whether a message will land, and a lot of the soft skills in marketing.

But that was never the job worth automating.

The stuff draining me on my business and the work I do with clients is the other stuff. I’m talking about the things we do on a regular and routine basis, like synthesize the pipeline before the Monday call, prep the founder for the analyst meeting, draft the LinkedIn post in his/her voice, read transcripts from last week and pull the three things we need to enable the team on, etc. etc.

And that’s what I built. I think of it more as an AI marketing chief of staff to help me run marketing than as an AI CMO. It runs at 7am on weekdays, costs about $80 a month, and the whole thing is a folder of markdown files I am putting on GitHub.

Of course, downloading it yourself will be easy, but the more time you spend with it, the better it becomes and the more you can rely on it.

The Biggest Mistake People Make with AI Marketing Setups

I’ve watched a lot of marketers build AI in the same two ways, and both break for the same reason.

The first way is one giant context file. You paste everything you know about the company into a single document or a single Claude Project and point every conversation at it. Sure, it feels thorough, but it’s why your outputs drift. When the agent loads the same monolith to draft a tweet and to analyze pipeline, the irrelevant 90%+ dilutes the relevant 10%. Context that is always on is context that is never sharp.

The second way is when every conversation starts cold, from scratch. And I made this mistake in the beginning a LOT. I would re-explain the client, the buyer, the positioning, the voice rules, and the three things you are not allowed to say, every single time. Even if it only takes 5 minutes, you should have to re-explain. Five minutes a session, ten sessions a day, across a whole team. Run that math for a quarter – it adds up!

The answer sits between the two. Persistent context, but modular, loaded by the slice and per task. The structure I started from is Nathan Whittemore’s Personal Context Portfolio: ten markdown files, each one a different dimension of who you are and how you work. Identity. Role. Current projects. Team and relationships. Voice. Goals. Constraints. Any agent can read any subset. The pipeline skill loads the pipeline files. The content skill loads the voice files. Nothing loads everything.

What’s Inside the Repo

The repo is a folder of markdown files. Nothing to install to read it. The parts you would touch:

• BUILD_PROMPT.md is one self-contained prompt you paste into Claude Code, and it sets up the whole skeleton for you.

• company-context/ holds the ten context templates, empty and ready for your company: Identity, role, current projects, team, tools, voice, goals, constraints, domain knowledge, and a decision log.

• skills/ holds a dozen skill specs, each one ending in an action and never in advice: daily brief, founder content draft, competitive digest, decision memo, the number-checker, the disclosure flag.

• wiring/ gives you three ways to connect the context files to your tools, from a full MCP setup down to a simple Claude Project, so you can pick by how technical you want to get.

• operator/ covers the unglamorous things that matter once it is live, like the scorecard, the backup, the kill switch, the cost controls, etc.

• Three planning files I would not skip: VALIDATION_CHECKLIST.md (the four assumptions to test before building), EVALUATION_TESTS.md (pass criteria written in advance), and WEEK_1_SCOPE.md (so you build three things end to end instead of forty halfway).

Get The Repo

What I Stole, and From Whom

Now, I’ll be honest – most of the ideas here are not mine, but I’ve configured them all in a unique way and so that the entire system runs using some of my favorite pieces from some of my favorite creators.

• Nathan Whittemore’s Personal Context Portfolio: I used this mainly for system memory. This is where I got the ten modular context files that let the system understand the account and how I make decisions, without reloading everything every time.

• SaaStr’s 10K (Amelia Lerutte and Jason Lemkin): I used this for the core architecture. This is where I got the principle that the app and the agent are one system, so a feature can graduate from a chat request to a permanent button once it proves itself.

• Sabrina Ramonov: I used some of her ideas this for the operating system. This is where I got the two-layer memory (a fixed playbook plus a running corrections log that makes the system sharper every week) and the rule that an agent should act and leave a file behind.

• Nate Herk: I used this for structure. This is where I got the router-plus-specialists pattern and the habit of matching the model tier to the size of the job, so I am not paying Opus rates to sort a list.

• Chase Hannegan: I used this for sequencing. This is where I got the build order (memory, then consistency, then access, then agents) that kept me from building the fun part before the foundation held.

• Tiago Forte: I used this for the framing. This is the foundation for the real personal context management, which turns out to fit a company as well as a person.

Share Stack & Scale

What It Does Day to Day

The system runs in two modes. Most of it happens on a schedule, with no human in the loop. The rest happens when I sit down with the agent and ask for something specific. Either way, it ends in a file, not a chat.

Every weekday morning (autonomous)

• Sends me a tactical brief at 7am with three to five specific moves for the day, each grounded in real pipeline numbers from the CRM, scoped to under two hours, and tied to either pipeline or the category narrative.

• Diagnoses which of the four Ps (Persona, Problem, Promise, Product) is the binding constraint that day before it generates a single idea.

• Surfaces what is queued for founder approval, what drafts are ready for review, days to the next event, and fundraise-readiness signals.

• Posts a short summary to a Slack channel with the headline pipeline number and the day-over-day change.

• Captures my reply to that email as a correction, so tomorrow’s brief reflects the feedback. This is a big part of the compounding loop.

Every week (autonomous)

• Drafts founder social posts in each person’s actual voice, source-grounded from their own podcast transcripts and meeting logs, queued for a standing review slot.

• Ships a Friday competitive digest with a few net-new competitor moves plus status-quo signals from prospect calls, each with a positioning implication for the platform narrative.

• Runs a weekend marketing-health assessment across five domains (PMF signals, positioning, ICP quality, GTM motion, marketing ops), scores each one to five, and flags only what moved enough to matter.

• Rolls up event pipeline, follow-up SLAs, and who needs a touch this week.

Every night (autonomous)

• Pulls every new meeting transcript (founder calls, my calls, anything recorded that day).

• Extracts action items with named owners, decisions made, and pipeline signals.

• Dedupes against the tasks already in my system and creates the net-new ones.

• Flags anything that touches a disclosure-sensitive relationship for review.

• Captures status-quo objections from prospect calls (the “we have an internal POC” defense, the “let’s pilot another quarter” delay) into a competitive intel table.

What it blocks (automatic, on every output)

• Any number that does not match the locked source of truth (ARR, customer count, deal sizes, accuracy claims).

• Any draft that breaks the writing style guide or names a competitor in external copy.

• Any draft where the founder voice fingerprint is wrong (one founder’s draft reading like the other’s).

• Any draft that positions the wrong buyer or lists more than the approved number of differentiators.

• Any late-stage tactic recommended at an early-stage company (no scaled paid, no SDR team, no ABM at scale at PMF Level 1 to 2).

• Itself, if monthly spend crosses a hard cap.

On demand (you ask, it produces a file)

• A bundled decision memo for a founder who has let calls stack up.

• A tiered executive outreach list for an event, scored against a ten-point ICP rubric.

• A pipeline funnel diagnostic naming the bottlenecked stage and three experiments to fix it.

• A quarterly positioning audit using April Dunford’s five-step method.

• A battlecard for any named competitor, with discovery questions and objection responses.

• A sales enablement suite (one-pager, demo script, email templates, stakeholder map) or a launch playbook with day-of sequencing.

What gets better over time

• Every correction I give writes to the corrections log. The next day’s prompts load it. The system gets sharper.

• Every transcript accumulates into a private record of how the company sells. Six months in, that is a proprietary GTM corpus.

• Every one-off script I write stays runnable. The library grows. The next time I need that analysis, it takes three minutes instead of thirty.

• The thirty-day labeling phase teaches the system what I approve versus rewrite. By day thirty, drafts hit roughly 80 percent of what I would have written myself.

What You Can Do With It

Here is the exact setup, step by step so you can get running with your chief of staff today.

1. Clone the repo and open the folder in Claude Code (or Cursor, or Replit Agent). Everything below happens from inside that folder.

2. Fill in the ten context files first. Open company-context/ and write your company into each template: identity, role, current projects, team, tools, voice, goals, constraints, domain knowledge, decision log. Do this before anything else, because every skill downstream reads these files, and a thin context file produces thin output. Budget an afternoon. This is the actual work.

3. Run the validation checklist. Open VALIDATION_CHECKLIST.md and answer its four questions for your own stack and team: can you pull CRM data on a schedule, can you reach your meeting transcripts, can you send mail from your domain, do you have the API keys. If any answer is no, change the wiring before you build, not after. Twenty minutes here saved me a rebuild.

4. Paste the build prompt into Claude Code. Open BUILD_PROMPT.md, copy everything below the divider, paste it into a fresh Claude Code session at the repo root, and tell it to build this. The prompt is self-contained: it carries the stack, the architecture, the staged build order, and pointers to every other file. It scaffolds the project for you.

5. Ship Week 1 only. Build two things end-to-end and stop: the daily brief and the corrections loop. The brief proves your context files are good. The corrections loop is what makes the system compound. Wire the backup, the kill switch, and the cost cap before the first scheduled job fires, then let it run a couple of weeks before you add a single skill.

6. Grade it at thirty days. Open EVALUATION_TESTS.md and run the three tests you wrote in advance: a blind voice test, a number-accuracy backtest, and a decision-speed baseline. If they pass, add the next skills. If they fail, fix the context and the corrections, not the model.

Time To Build

Don’t build an AI CMO. You will spend six months trying to model judgment that does not fit in a prompt, and you will end up with a confident intern who has never met your customers. Build the chief of staff instead. Ten context files, one corrections loop, a few guards, and a schedule. The intelligence was never in the agent. It lives in the knowledge base the agent reads every morning, and that part you can start building today.

The repo is public at https://github.com/guerrilla2799/ai-cmo-operator. Fork it, add to it, maybe even break it, and send me what you build.

Get the Repo

That’s it for this week!

– Brandon

Anthropic and OpenAI have gotten most of the attention this year, which has distracted other people from other really powerful AI solutions. Perplexity has always been known for the go-to tool for deep research, and I’ve been a pro user for almost a year. But I recently gave Perplexity Computer a try, and this Stack & Scale issue is all about why you should stop sleeping on one of the most powerful tools out there right now.

So, if you’re still only using Perplexity for research, that’s fine, but it’s like using a Formula 1 car to drive to the grocery store.

I’ve been using Perplexity Computer for two weeks. In that time, I built a working B2B pipeline diagnostic tool with five inputs, a stage diagnosis, twelve channel verdicts, and a 30-day playbook with specific dollar reallocations. The whole thing routes work across three AI models in the background. And, of course, I didn’t write a line of code.

Anyone who’s been doing this long enough has watched influencers on social media tout a new solution, saying how easy it is to use, but when you go to use it, nothing works as it should. And to be honest, I was slightly skeptical the first time I saw an influencer bragging about what he was able to build with Perplexity Computer. The more people I saw building cool things, the more I knew I had to try it.

Before I show you what I built, I want to walk you through what Perplexity Computer is.

What Perplexity Computer is

Perplexity Computer is a cloud-based AI that runs in the background, routes work across 19 different AI models, and connects to more than 400 tools to ship finished work end-to-end. The mechanic that makes it different is the input. You don’t have to do crazy prompt engineering or use a custom GPT or skill that gives you the most elaborate prompt. All you have to do is start with the outcome you want.

Tell it “compare Apple, Microsoft, and Google stocks, build me a dashboard on each, and write a short video script on which one to bet on” and it breaks the work into subtasks. It routes each subtask to the model best suited to it. Claude will handle the reasoning, Gemini will do the research, GPT will run the data work, and Perplexity Search will pull live citations. Then it stitches the result into a single finished deliverable.

The nice part that took me a minute to internalize is that it runs in the background. It doesn’t stop when you close the tab. It can keep running for hours, days, or weeks on a single goal (or so I’m told; my most complex build only took about 90 minutes). Perplexity Computer is operational, and the whole point is that it doesn’t stop until the work is done.

Why it’s a different kind of tool

I talk to marketing leaders every week about how they’re using AI, and I hear the same thing over and over again. They’ve got four ChatGPT tabs open: One for LinkedIn posts, one for ad copy, one for brainstorming, and one for prepping for the board meeting. What you see is AI is doing a lot of writing, but the CMO still wants to do the strategy and the delivery and piece all the systems together.

That’s fine, and I think it works for most people, but it’s time to use that F1 car for racing. It’s time to get more value out of the tools you have at your fingertips.

There are three main things that make Perplexity Computer different from the AI most marketers are using right now.

1. It does its own model routing. You stop picking which AI to ask. It picks for each subtask automatically.

2. It acts inside your stack instead of next to it (Gmail, Slack, HubSpot, Salesforce, Notion, GitHub, your calendar, your files). It doesn’t generate text and ask you to paste it somewhere. It executes inside the tools you already use.

3. It runs while you’re doing other things. The work happens whether you’re at your desk or not.

The GTM use cases that delivered

I’ve been stress-testing Perplexity Computer against the workflows I run frequently as a fractional CMO. Here are five that really delivered for me:

First up is a weekly pipeline brief that runs on its own. I told it to pull from HubSpot and Salesforce every Sunday night, write the narrative, flag the at-risk deals, and drop the summary into Slack before my Monday standup. I haven’t built that report manually in two weeks, and the version it ships is better than the one I used to build because it includes data I would have skipped to save time.

Next, campaign post-mortems. I gave it a campaign name and told it to pull the data from Google Ads, LinkedIn Ads, and HubSpot, identify what worked, and draft a board-ready summary. What used to be a half-day exercise became a 15-minute edit of work that was already done.

Then there are ABM target account briefs that update automatically. I dropped in my Tier 1 list and told it to pull the latest funding news, hiring signals, exec changes, tech stack additions, and recent LinkedIn activity from the buying committee for each account. The output was a one-page brief per account, refreshed every Monday morning. We’ve all been in the moment where an AE pings you on a Friday afternoon, asking for context on an account you know nothing about. I don’t have to worry about that ever again.

Next was my quarterly board deck drafts. I connected it to HubSpot, Salesforce, GA4, and the ad platforms, then told it to pull the quarter’s pipeline, marketing performance, channel breakdown, and CAC trend, write the narrative, and draft the slides. The before-and-after on my prep time was something like 12 hours down to 90 minutes. Most of those 90 minutes are now spent on the narrative I want to tell, which is the part that matters anyway.

And the one I didn’t expect to work as well as it did was my pre-event prep briefs. We are in the pre-summer event swing, so I dropped in a conference name and told it to pull the speaker list and the attendee company list, score each against my ICP, build research briefs on the top 25, and draft pre-event LinkedIn outreach in my voice. I walked into the event with warm context on 25 people I wanted to talk to. Three turned into real conversations. One turned into a deal that’s now in pipeline.

Of course, I don’t need to tell you that each of these would require a lot of work from many different people. And there’s some work on my own. Now, maybe an hour or two to set it up, then minimal ongoing maintenance.

Check out what I built – The Pipeline Reality Check

Now I can’t share any of those other builds with you because they have proprietary data, but I still wanted to build something for you, my Stack & Scale audience. I called it the Pipeline Reality Check.

I built it because I think most pipeline calculators are useless. They all do the same thing (revenue target divided by ACV divided by win rate equals the number of meetings you need). They tell you the number, but they don’t tell you whether your current channel mix can deliver it.

Mine does. There are five inputs in, and you get a stage diagnosis (where you are on the PMF arc), a verdict on twelve acquisition channels (scale, watch, cut, test, or skip), and a 30-day playbook with specific dollar reallocations tied to your numbers. The math runs on GPT. The narrative runs on Claude’s voice. The live channel benchmarks come from Perplexity Search with cited sources. Three models, one tool, one finished product.

The build itself is what I want you to pay attention to. Again, I didn’t write a line of code. I scoped the outcome in one prompt (the math logic, the channel framework, the design system, the email gate, all of it) and Perplexity Computer broke it into subtasks, routed each one to the model best suited for it, and shipped a working web tool. This would have been a four-week build with a developer six months ago, but I did it myself in an afternoon.

You can try it on your own numbers here.

Try it now

And if you want to build your own, here’s the prompt that I used:

I want to build a B2B SaaS pipeline diagnostic web tool. Most pipeline 
calculators tell you the number of meetings you need to hit your target. 
This one should tell the user whether their current channel mix can 
deliver it. The goal is to give marketing leaders clarity about what to 
fund, watch, cut, test, or skip.

What it should do

Walk the user through a 5-step wizard:

1. Their numbers: annual revenue target, average contract value (ACV), 
   current win rate, sales cycle length, current ARR
2. Their funnel: demo show rate, demo-to-qualified rate, qualified-to-close 
   rate
3. Their stage: 4 multi-choice diagnostic questions covering referrals, 
   retention, channel repeatability, and customer urgency
4. Their motion: primary go-to-market motion (founder-led, sales-led, 
   product-led, or hybrid), current monthly marketing spend, channels 
   they're running today (multi-select from a list of 10)
5. Their reality check: the results page

What it should output

Compute and display:
- Annual deals needed, monthly deals needed, demos needed per month, 
  qualified meetings needed per month, and CAC ceiling (ACV divided by 3)
- A stage diagnosis (Stage 1 Searching / Stage 2 Developing / Stage 3 
  Repeatable / Stage 4 Scaling) based on the user's diagnostic answers
- A verdict on each of the 10 channels using this 5-state system:
   - SCALE (running it, keep it)
   - WATCH (running it, monitor for 30 days)
   - CUT (running it, stop or rework)
   - TEST (not running it, pilot it for 30 days)
   - SKIP (not running it, wrong stage)
- A 3-paragraph strategic read in plain English covering the situation, 
  the diagnosis, and the recommended action

Format all numbers with comma separators ($500,000 not $500000).

Branding (fill in with your details)

- Primary color: [HEX]
- Accent color: [HEX]
- Background: white throughout, with light gray section breaks
- Font: [FONT NAME] for everything
- Brand name in header: [YOUR BRAND]
- Logo: [PASTE LINK or attach]

Voice

For any narrative output, match the tone of these samples:
[PASTE 2–3 of your LinkedIn posts, newsletter intros, or other writing 
samples here]

End with

An email capture card with work email and name fields. Submission 
triggers a webhook to my email tool (I'll add the webhook URL after 
the build is done).

Build this as a clean single-page web tool. Skip demo polish. The 
output should feel useful enough to publish.

Why you should be paying attention right now

Somewhere in your category, a marketing team is going to find Perplexity Computer this quarter and use it to outwork everyone else for the next four. I don’t know which competitor it’ll be. I do know it’ll happen.

Most marketing teams I talk to are still using AI for the easy stuff (writing emails, drafting LinkedIn posts, and summarizing calls). That’s a fine starting place, but it’s a tool that gives you back a few minutes a day. Perplexity gives you back a few hours each week while you’re doing something else.

Start here

Pick the workflow you dread on Monday mornings. The one you’ve been meaning to systematize for six months and haven’t. Block 30 minutes this week and hand it to Perplexity Computer.

That’s the whole test. If it works, you’ve built a piece of background infrastructure that runs forever. If it doesn’t, you’ve spent 30 minutes figuring out where the tool’s edges are. Either way, the time pays for itself.

Learn more about Perplexity Computer here.

Check out Perplexity Computer

And if you want to see what’s possible when you let Perplexity build something for you, run the Pipeline Reality Check on your own numbers here.

That’s it for this week. Play around and let me know what you build!

– Brandon

Our social feeds have been overrun with talk of AI agents, but I actually think teams are spending too much time fixating on AI. There’s not a single client I’ve worked with in the last 18 months that wouldn’t benefit more from setting up automations. Yes, You still have to be on top of AI, but I think the better ROI right now that are more sustainable are coming from the work that GTM engineers are doing, building automation.

The average B2B SaaS company now spends $2 on sales and marketing to drive #1 new ARR, up 14% from last year. But boards aren’t approving headcount to help you make up the difference. They’re demanding the same growth from a leaner team, which means the work has to come from systems that run without a person babysitting them. That’s the exact job the GTM engineer was invented for.

Clay put a name to the role in 2023, and around 100 GTM engineer listings now go live every month, which tells you how quickly companies realized they need this skill on the team.

Don’t get me wrong, I’m not saying don’t use AI agents, because if you’ve been following me, I talk a lot about that. However, the best and most effective GTM teams first built tight, predictable automations first, and they layer in AI as one ingredient inside those systems instead of the foundation.

As usual, I wrote this issue be actionable and practical. The goal is to get you past using Clay for enrichment and into building actual GTM workflows, so you can automate the parts of marketing that should be automated and spend your own hours on the parts only a human can do. Anthony Kennada put it well on LinkedIn: “But the inputs on the front end of building the brand, as well as the activation outputs on the other end of the prompt, is where we find our craft as human marketers.”

GTM Engineering Maturity Model

101: Workflow Builder

A 101 GTM Engineer is mostly a high-agency operator who can create reliable automations inside a narrow scope: CRM hygiene, routing, enrichment, list building, basic outbound support, and reporting. The starting point is simply getting comfortable with Clay, formula columns, enrichments, and simple company/person workflows.

What defines 101

• Builds single-purpose automations with clear inputs and outputs.

• Works mostly inside tools rather than across systems.

• Solves immediate pains: manual research, routing errors, bad data, repetitive outbound prep.

• Thinks in tasks, not architecture.

Typical use cases

• Lead routing rules.

• Enrichment waterfalls.

• Prospect list building.

• Simple outbound personalization prompts.

• Basic dashboards and CRM cleanup.

201: Systems Builder

A 201 GTM Engineer moves from isolated workflows to multi-step systems that connect data, signals, routing, segmentation, and outbound motion. This is where signal-based selling, data orchestration, and cross-functional collaboration start to matter more than pure tool fluency.

What distinguishes 201

• Designs end-to-end process flows, not just automations.

• Understands how data moves through the revenue engine.

• Uses AI selectively to speed research, classification, and decisioning.

• Partners closely with RevOps, Sales Ops, and Marketing Ops.

What defines 201

• Signal-based account prioritization tied to routing and outbound.

• Buying committee mapping from company + people + intent data.

• SLA-aware lead handling across marketing and sales.

• Automated prospect research briefs for reps.

• Lead scoring pipelines that combine structured and unstructured signals.

Where a 101 practitioner is automating tasks, a 201 practitioner asks, “What is the system, where does it break, what signal should change the next action, and how do we keep it governable?” That systems view is a recurring theme in practitioner commentary about GTM Engineering as the “glue layer” between RevOps and product-like execution.

301: Revenue Infrastructure Architect

A 301 GTM Engineer is building AI-powered revenue infrastructure – reusable internal products, agent-assisted operations, orchestration logic, and strategic revenue architecture. I think of this level as a hybrid of product management, data engineering, automation engineering, and RevOps leadership.

What defines 301

• Owns the architecture of the revenue system, not just individual automations.

• Designs human-in-the-loop and agentic workflows with control points, caching, fallbacks, and governance.

• Treats GTM logic as an internal product with lifecycle, QA, instrumentation, and adoption.

• Thinks in terms of scalability, reliability, and compounding leverage.

Typical use cases

• Autonomous prospect research pipelines with guardrails.

• Agent-assisted account prioritization and next-best-action systems.

• Personalized outbound generation that adapts to buying signals.

• Revenue intelligence systems that detect leakage, risk, and opportunity in real time.

• Internal GTM copilots for reps, managers, and operators.

At this level, it’s a system where AI handles research, summarization, classification, drafting, and triage, while humans supervise policy, exceptions, and strategic decisions. Glean’s RevOps guidance and other AI-in-RevOps examples emphasize secure workflows, governance, and human review rather than full autonomy.

Share Stack & Scale

Skills by GTM Engineering Level

A GTM engineer starts with practical operator skills: being fluent in the CRM, understanding funnel math and revenue metrics, and being able to build clean workflows in tools like Clay, HubSpot, Zapier, or n8n.

At the 101 level, the focus is on reliability and execution. At 201, those same skills deepen into SQL, API literacy, data modeling, and system design so you can connect workflows across tools and improve pipeline quality. And at 301, the role expands into architecture, automation governance, Python or scripting, and AI workflow design, where the job is less about building one-off automations and more about designing durable revenue infrastructure that scales and adapts.

10 High-Impact Automation Plays You Should Build Today

Okay, now on to the part that I think most people will skip directly to. The actual plays and use cases.

1. Anonymous website visitor to account routing

Start by deciding which pages count as high-intent, usually pricing, comparison, alternative, integration, or demo pages. Then set up a visitor signal source and route those visits into a staging layer where you can match the anonymous company to your CRM, remove existing customers, and check ICP fit before anyone gets alerted. Once the account clears qualification, push it to Slack, create a task for the owner, or enroll it in a campaign; then track whether those alerts turn into meetings or pipeline so you can refine the page list and scoring rules over time.

2. Pricing-page visitor to personalized nurture

Build this by first tagging pricing-page and high-intent product-page visits, then enriching the company and contact record so you know who this is and whether they fit your target segment. Next, create a branching nurture path based on the likely objection: price sensitivity, implementation risk, missing features, or competitive comparison, and only generate the message after you have the framework and proof points in place. The key GTM engineer move is to qualify before generating, because AI is much better at drafting the message than deciding whether the account deserves one.

3. Competitor research signal to ABM activation

Start by defining which signals count as competitor research, such as visits to alternative pages, comparison content, review sites, or related intent data. Then enrich the company, map the likely buying committee, and segment the account into the right campaign path so marketing can launch targeted ads, email, or sales alerts without waiting for a form fill. The workflow should end with a measurement layer that tells you whether competitor-intent accounts move faster or convert better than the rest of your audience.

4. Job-change trigger to account reactivation

Pull in job-change signals from your data source, then check whether the person already exists in your CRM and whether their new company is in your target market. If they are relevant, enrich the new company, reconnect the contact history, and decide whether the next action is a warm congratulations email, an internal alert to an owner, or a reactivation sequence for the new account. This workflow works best when you treat the job change as a reason to reassess account status, not just as a contact update.

5. Funding-round trigger to campaign launch

Set up a funding signal, then create a filter that checks company size, stage, geography, and ICP fit so you do not launch campaigns into irrelevant companies. After enrichment, route the account into a campaign cluster based on stage, because a Series A company needs different messaging than a late-stage company with a mature stack and different buying constraints. The best GTM engineers also define the action layer in advance: email, retargeting, SDR alert, or all three, depending on account value and urgency.

6. Technographic change to persona-specific messaging

Build this workflow by monitoring for meaningful tech stack changes, especially installs, removals, or replacements of tools that relate to your category. Then enrich the account and determine whether the change implies switching pain, integration risk, compliance pressure, or expansion opportunity, and use that to branch the campaign logic. Once the account is classified, generate persona-specific messaging blocks so marketing can speak differently to an operator, a technical evaluator, or an economic buyer.

7. Content-engagement scoring to MQL or SQL prioritization

Instead of scoring every touch in isolation, define a weighted engagement model that distinguishes casual readers from real buyers. Pull in the events you care about, such as webinar attendance, repeated page visits, gated content downloads, and return sessions, then calculate a rolling score and map that score to lifecycle stage or routing rules. In practice, the GTM engineer’s job is to make sure the score is tied to real downstream action, not just a dashboard metric that nobody uses.

8. Dark-funnel account list builder

This workflow starts with a hypothesis about what an in-market account looks like, then pulls together weak signals from web behavior, hiring, funding, technographic changes, and content engagement. After that, you enrich and score the accounts, suppress bad-fit records, and output a prioritized list that can feed paid, outbound, or partner motions. The practical value here is that marketing stops waiting for explicit intent and starts activating against a pattern of behavior that strongly suggests interest.

9. Webinar attendee to sales-assist workflow

Build the workflow around registration and attendance data, but separate live attendees, no-shows, and high-engagement participants into different branches. Then enrich the attendee records, tag them by topic interest or company fit, and push each segment into a follow-up motion that matches the behavior they showed during the event. A strong version of this system also generates a rep-facing summary so sales knows which attendees were worth prioritizing and why.

10. AI-generated account brief for marketers and reps

This workflow should begin only after the account has been qualified by signal and fit, not before. Once the account clears that threshold, gather firmographic, technographic, people, and activity data, then feed it into an AI step that generates a short brief with the account’s context, likely pain points, and recommended next actions. The GTM engineer’s role is to make the prompt and guardrails specific enough that the output is useful, brand-safe, and grounded in evidence rather than generic AI prose.

Conclusion

The most useful way to build these is to follow the same sequence every time: trigger, qualification, enrichment, action, measurement. If you keep that pattern consistent, the individual workflow can change without your entire operating model becoming brittle.

The strongest teams also centralize logic and distribute execution, which means they define scoring, routing, and playbook rules once, then push the outputs into CRM, Slack, email, ads, or browser-based tools where the team actually works. That is the real GTM engineering skill: not just stitching tools together, but turning signal into repeatable revenue motion.

I think this is also the most direct path to becoming indispensable in your company over the next couple of years, and I don’t say that lightly. The efficiency math I opened with isn’t easing up. Every leader is being asked to grow revenue without growing headcount in step. The person who can look at a manual, expensive motion and rebuild it as a system that runs while everyone sleeps becomes the person the company can’t operate without. That ability has almost nothing to do with which tools you know. It comes from how you see the business.

Everyone wants a competitive outbound play, and frankly, everyone should be running one, but if you’re like many companies I work with (or have worked at in the past), it was a good idea, but never executed well.

Most teams pull a technographic list, dump it into a sequencer, and hope for the best. The problem is that much of that data is stale, so reps spend their time chasing companies that aren’t even using the competitor anymore. I’ve seen these campaigns work incredibly well, but only when you add better data, stronger scoring, and guardrails that keep your team focused on the accounts most likely to convert.

And this week’s Stack shows you how to do exactly that.

This week’s Stack:

• 1 video: Build a competitive outbound engine in Clay

• 1 prompt: Create your AI change management plan as a CMO

• 1 tool: Connect Manus to Meta Ads Manager for instant insights

• 3 resources: Personal AI systems, Claude tips, and ChatGPT memory

• 3 jobs: Product and performance marketing roles at top AI companies

Let’s go!

Workflow Walkthrough: Build a Competitive Outbound Engine That Your Reps Will Love

Everyone has competitors, So everyone can use this week’s Clay play.

In this 14-minute walkthrough, I show you how to build a Clay workbook that validates technology usage with HG Insights, scores every account across seven weighted criteria, and automatically filters out customers, open opportunities, and recently touched accounts. The result is a prioritized list your reps can trust (and a much better chance that you’ll look like the hero who brought it to the team).

The same framework also works for churn prevention, partner targeting, and expansion plays. Better data beats bigger lists every time, and timing and context matter far more than a giant spreadsheet of logos.

What you’ll see:

• How to use HG Insights to verify which companies are actively using a competitor’s product

• Why I score accounts across seven criteria, including revenue, hiring, funding, and signal freshness

• How to add Salesforce guardrails so reps never touch customers or active deals

• Ways to generate AI-written openers and send account dossiers to Slack or your sequencer

• How to repurpose the same workflow for partner campaigns, churn alerts, and expansion opportunities

If you’re just getting started with clay and need some extra credits to play around with, sign up with my referral code and you’ll get 3,000 free credits.

Here’s the Clay Workbook Template you can copy into your own Clay instance.

Get the Clay Template!

Prompt of the Week: The CMO’s AI Change Management Playbook

I’ve learned that the hardest part of AI adoption is getting people to change how they work. You can give them the tools and give them access to learning. But as they say, you can only lead a horse to water.

I’ve seen teams buy enterprise licenses, announce a bold AI strategy, and then watch usage stall because no one understood what was expected of them. The companies making the most progress are treating AI like any other major transformation initiative. They set a clear vision, define new ways of working, and communicate consistently so the entire organization understands where they’re headed and why it matters.

Use this prompt to create a complete change management and communication plan for rolling out AI across your marketing organization (or your entire company).

Act as an experienced Chief Marketing Officer and enterprise change management leader with deep expertise in AI adoption, organizational design, and executive communications.
Your task is to create a complete change management and communications plan for launching and scaling AI initiatives across my organization.
## Context
Company Name: [COMPANY NAME]
Industry: [INDUSTRY]
Business Model: [B2B SaaS / B2B Services / Marketplace / Other]
Company Size: [NUMBER OF EMPLOYEES]
Marketing Team Size: [NUMBER OF TEAM MEMBERS]
Annual Revenue: [ARR OR REVENUE]
Primary Markets Served: [TARGET MARKETS]
Current AI Maturity: [Beginner / Intermediate / Advanced]
Executive Sponsor(s): [CEO, CMO, COO, CTO, etc.]
Departments Involved: [Marketing, Sales, Customer Success, Product, Operations, HR, Finance]
Primary AI Use Cases: [Content Creation, Research, Forecasting, Personalization, Automation, Analytics, etc.]
Primary Goals: [Increase Productivity, Improve Quality, Reduce Costs, Accelerate Innovation, Improve Decision-Making]
Known Concerns: [Job Security, Data Privacy, Compliance, Quality Control, Change Fatigue]
Implementation Timeline: [90 Days / 6 Months / 12 Months]
Budget: [BUDGET]
Technology Stack: [ChatGPT, Claude, Gemini, Copilot, Clay, n8n, etc.]
## Objectives
Create a detailed plan that helps me:
1. Communicate a compelling vision for why AI matters.
2. Build excitement and reduce fear across the organization.
3. Define clear expectations for leaders and individual contributors.
4. Establish governance, policies, and responsible use guidelines.
5. Roll out training and enablement programs.
6. Measure adoption and business impact.
7. Sustain momentum over time.
## Deliverables
### 1. Executive Vision Statement
Write a concise but inspiring message from the CMO explaining:
- Why AI is a strategic priority.
- How AI will augment employees rather than replace them.
- What success looks like.
- What is expected from every team member.
### 2. Change Narrative
Create a narrative covering:
- Why we must change now.
- What happens if we do nothing.
- What will change.
- What will stay the same.
- How employees will be supported.
### 3. Stakeholder Analysis
Identify:
- Executive leaders
- Functional managers
- Individual contributors
- AI champions
- Skeptics and potential blockers
For each group, include:
- Likely concerns
- Key messages
- Desired behaviors
- Communication channels
### 4. Communications Plan
Build a 12-month communications plan including:
- Executive kickoff announcement
- Town halls
- Team meetings
- Email updates
- Slack announcements
- Success stories
- Office hours
- Quarterly business reviews
For each communication, include:
- Timing
- Audience
- Objective
- Core message
- Owner
### 5. Training and Enablement Plan
Design role-based training for:
- Executives
- Managers
- Individual contributors
- AI champions
Include:
- Learning objectives
- Curriculum topics
- Recommended cadence
- Certifications or assessments
### 6. Governance Framework
Outline:
- Approved tools
- Data security guidelines
- Prompt standards
- Human review requirements
- Compliance and legal considerations
- Escalation procedures
### 7. Adoption Metrics Dashboard
Recommend KPIs such as:
- Active users
- Frequency of use
- Time saved
- Cost reduction
- Campaign velocity
- Content output
- Pipeline contribution
- Employee sentiment
### 8. Resistance Management Plan
Detail:
- Common objections
- Recommended responses
- Coaching guidance for managers
- Escalation paths
### 9. Recognition Program
Create ways to celebrate:
- Early adopters
- High-impact use cases
- Productivity gains
- Innovation awards
### 10. 30-60-90 Day Rollout Plan
Provide:
- Key milestones
- Deliverables
- Decision points
- Risks and mitigation steps
### 11. Leadership Talking Points
Write concise talking points for executives and managers.
### 12. FAQ
Draft 25 frequently asked questions covering:
- Job security
- Data privacy
- Tool selection
- Training expectations
- Performance implications
- Governance
## Output Format
Present the final plan in a structured format with:
- Executive summary
- Detailed tables where appropriate
- Sample communication templates
- Actionable recommendations
- Prioritized next steps
Write in a practical, executive-ready style that I can use immediately with my leadership team.

I’ve added this to my The Ultimate ChatGPT Prompt Library for B2B Marketing Leaders notion doc. Check it out for 65+ more prompts.

Know someone who might find this prompt useful? Share it with them!

Share

Tool of the Week: Turn Meta Ads Manager into Your Competitive Intelligence Team

I’ve been spending some more time with Manus lately because a few marketing operators I highly respect are using it. One of the more useful applications is connecting it directly to Meta Ads Manager.

Once connected, you can build custom dashboards, analyze campaign performance, monitor spend and creative trends, and even scrape your competitors’ ads to see what’s working and where they’re focusing their messaging. I also use it to generate recommendations for improving targeting, creative, and budget allocation based on live account data.

If you’re running paid social, this turns your ad account into a source of insights instead of a collection of reports.

Check it out: https://manus.im/blog/manus-meta-ads-manager-connector

AI Resource Roundup

My AI Personal Operating System, Version 2.0 (Kathleen Booth’s Code Meets Creed Newsletter): One of the best behind-the-scenes looks I’ve seen at how someone has structured AI into their daily workflow. Kathleen covers how to think about models, prompts, memory, and systems so AI becomes part of how you work instead of another app you occasionally open.

52 Free Claude Tips and Tricks (Charlie Hill on LinkedIn): Charlie put together a fantastic collection of practical Claude tips you can start using immediately all in one LinkedIn post. If you’re spending more time in Claude, this is a quick way to pick up smarter prompting techniques and workflow ideas without paying for an expensive course.

The New Jobs AI Will Create (AI Daily Brief Youtube): The AI jobs debate has spent years asking which roles will disappear. This video asks the more important question: what becomes possible when AI expands the amount of useful work the economy can support? It’s a nice change from the AI Jobs apocalypse narrative.

Hot AI Jobs:

Here is another set of great AI jobs for marketers that came across my desk, but I do apologize that none of them have pay ranges. On the positive side, all of them are fully remote roles.

Senior Director, Product Marketing at Diagrid

• Location: United States (Remote)

• Pay Range: Not listed

Director of Product Marketing, AI at Unity

• Location: United States (Remote)

• Pay Range: Not listed

Head of Performance Marketing (Agentic AI-First) at ZoomInfo

• Location: United States (Remote)

• Pay Range: Not listed

Yesterday was my kids’ last day of school, which means summer has officially begun.

Or at least I think it has – we somehow got snow in Colorado this week, because apparently the weather didn’t get the memo.

Summer is one of my favorite times to experiment with AI. Things tend to slow down a bit, calendars open up, and it’s a great opportunity to build systems that will save you time for the rest of the year. That’s exactly what I plan to be doing (at least during the brief windows when I’m not being recruited into snack duty, fort building, or whatever other jobs my kids assign me).

Have a great week, and I hope you find some time to build something amazing.

– Brandon

It’s no surprise that one of the most frequent questions I get asked is, “What am I building?” Another one is “What are you seeing others build?”

I’ve been collecting some of the coolest workflows and agents and I wanted to share 10 with you this week and tell you exactly how to build them.

Two types of agents show up in this list. The first is on-demand agents, which take a batch of inputs you hand them, like a set of call recordings or a pipeline report, and return a structured output. You run them when you need them, and they take 10 minutes instead of three hours. The second is scheduled agents, which run automatically at a set interval. For example, on Monday morning, your weekly digest is already there. Sunday night your competitive monitor has already reviewed what changed.

So without further ado, let’s dive in.

1. The Stalled Deal Spotter

Outcome: You walk into every pipeline review with a deal-by-deal risk assessment and a specific marketing action for each stalled opportunity, instead of spending the first 20 minutes trying to reconstruct what happened from memory.

Why this matters: Pipeline reviews are mostly retrospective. Someone shares a slide showing 15 deals that haven’t moved in 30 days, and the room tries to diagnose each one on the fly. Every rep has a different story, but nothing gets resolved, and you leave with action items that go unfollowed up.

How to run it: On-demand, before every pipeline review. Pull the CRM notes, last email thread summary, and most recent call transcript for every deal that hasn’t progressed in 21 or more days. Paste them into Claude with the prompt below. What you get back is a structured breakdown by deal: most likely stall reason, the specific stakeholder or objection blocking progress, and a concrete marketing action for that deal specifically. Not “send a case study.” Something like “the last call mentioned budget pressure for this quarter — send the ROI calculator and ask the rep to request a CFO conversation.”

The prompt:

You are a B2B revenue intelligence analyst reviewing stalled pipeline on behalf of a marketing leader preparing for a pipeline review.

I will provide CRM notes, email thread summaries, and call transcript excerpts for deals that have not progressed in 21 or more days.

For each deal, analyze the available context and return your assessment in this format:

Deal: [name] Days stalled: [number] Most likely stall reason: [one sentence grounded in the evidence, not speculation] Blocking stakeholder or concern: [specific person or objection if the data shows it] Recommended marketing action: [specific — name the content angle, asset, or play relevant to this deal's exact situation, not a generic suggestion] Confidence: [High / Medium / Low] — [one sentence explaining why based on evidence quality]

If a deal appears to have gone dark with no recoverable signals, say so directly. Do not frame a dead deal as a reengagement opportunity. My goal is an accurate picture going into the meeting, not an optimistic one.

Deals to review: [paste CRM notes, email summaries, and call transcripts here]

Stack requirements: CRM with deal notes (Salesforce, HubSpot, or Pipedrive), call recording tool for transcripts (Gong, Chorus, Fathom, or Fireflies), email client for thread summaries (Gmail or Outlook).

To automate: HubSpot or Salesforce API connected to Claude, triggered on a weekly schedule or when deal age crosses a threshold you define.

Pro tips:

• Feed it about 5 deals at a time for the best output quality. More than that and the per-deal analysis gets noticeably thinner.

• Add the deal size and target close date to each input. The agent can factor urgency into its recommendations when it knows what’s at stake.

• Share the output with your sales partner before the pipeline call, not in it. The best use is pre-call preparation, not real-time diagnosis in front of your CRO.

2. The Account Signal Monitor

Outcome: Every week, your BDR and AE teams receive a ranked short-list of the target accounts showing the strongest buying signals right now, with a one-line explanation of what changed and what to do about it.

Why this matters: ABM programs spend a lot of time on list-building and almost none on timing. A well-defined target account list sitting in a spreadsheet is doing nothing. The accounts on that list change every week: new executives joining, funding announced, job postings signaling a relevant initiative, G2 review activity picking up. Most marketing teams find out about these signals when a rep mentions it on a call, weeks after it happened, after a competitor has already moved.

How to run it: Scheduled, runs every Sunday night. You (or your ops person) pull recent signals for your top 50 to 100 accounts: job postings in relevant roles, executive changes, funding news, press mentions, and review site activity. The agent reviews all of it and returns the 10 accounts with the strongest signals this week, ranked by urgency, with a recommended action for each. Your demand gen team uses it to update paid audiences. Your BDR team uses it as their Monday morning call list.

The prompt:

You are a B2B account intelligence analyst helping a marketing team prioritize their target account list for the week ahead.

I will provide a list of target accounts with recent signals for each: job postings, executive changes, funding news, press mentions, and review site activity from the past 7 days.

Your job: Identify the 10 accounts showing the strongest buying intent or timing signals this week. Consider: Are they hiring in roles that suggest a relevant initiative? Has leadership changed? Did they raise funding or announce a new strategic direction? Are their G2 reviews signaling pain in a category we address?

For each of the 10 accounts, return:

Account: [name] Signal summary: [one sentence describing what changed and why it matters] Signal strength: [Strong / Moderate / Weak] with one sentence of reasoning Recommended action: [specific — who should reach out, what angle to lead with, or what content to send this week]

Rank the list from highest to lowest urgency. If fewer than 10 accounts have meaningful signals this week, say so rather than padding the list.

Account data to review: [paste account signal data here]

Stack requirements: To automate: Clay is the most efficient tool here because it can pull job postings, news, and LinkedIn signals in a single workflow and feed them directly to Claude. Another great option is Claude Code + Apify MCP. Apify has a native MCP server and pre-built actors for scraping LinkedIn job postings, Google News, and G2 reviews. Claude Code calls the Apify MCP to fetch signals across all your target accounts, runs the synthesis and ranking, and delivers the output on a weekly schedule via cron.

Pro tips:

• A great signal is job postings in adjacent roles. A new “Head of Revenue Operations” post at a target account almost always precedes a tech stack review. Build this signal into your scoring.

• Run it against your active pipeline accounts too, not just your cold target list. A buying signal inside an active deal can be the difference between a deal accelerating and going dark.

• Add a simple scoring system to your prompt: 3 points for an executive hire, 2 for a funding round, 1 for a relevant job posting. Ask the agent to apply the score and sort by total. It makes prioritization less subjective.

3. The Win/Loss Synthesizer

Outcome: At the end of every quarter, you have a pattern analysis of your won and lost deals: the most common buying triggers, the objections that actually killed deals, and the language your buyers use to describe their pain before they ever heard of you.

Why this matters: Most marketing teams know they should run win/loss analysis consistently, but few actually do it because it’s time-consuming to pull together, awkward to coordinate with sales, and the outputs usually live in a Google Doc that nobody opens by the time the next planning cycle starts. So messaging stays based on assumptions that were formed 18 months ago, and the gap between how your buyers describe their problem and how your homepage describes it just keeps growing quietly.

How to run it: On-demand, once per quarter. Export your closed-won and closed-lost call transcripts from Gong, Chorus, Fathom, or even written summaries. Paste them in with the prompt below. What comes back is a structured analysis: the five most common buying triggers in won deals with direct quotes, the five most common objections in lost deals with direct quotes, and the customer-native language for the problem you solve before buyers knew your product existed. That last piece is where most positioning improvements come from. I’ve seen this output rewrite homepage copy, restructure email sequences, and settle six-month internal debates about what to lead with in the sales deck.

The prompt:

You are a B2B positioning analyst reviewing a quarter's worth of sales conversations to surface patterns that should improve marketing and messaging.

I will provide call transcripts and summaries from closed-won deals and closed-lost deals this quarter.

Analyze all transcripts and return the following, with direct quotes from the source material for each finding:

The five most common buying triggers in won deals: What was happening in the buyer's world that made them start looking? Quote their exact language.

The five most common objections or hesitations in lost deals: What specifically blocked the deal? Quote the language buyers used, not the rep's interpretation of it.

Customer-native language for the problem: How did buyers describe their pain before they knew about our product? List the specific words and phrases they used. Do not use any of our product language or category terms in this section.

What distinguishes won deals from lost deals: Identify 2 to 3 patterns that clearly separate the two groups. These might be firmographic, behavioral, or related to the buyer's stated situation.

Be specific and quote directly. If the transcripts don't support a finding, say so rather than inferring.

Transcripts to analyze: [paste call transcripts and summaries here]

Stack requirements: Call recording tool (Gong, Chorus, Fathom, or Fireflies) for transcript exports, CRM for tagging deals as won or lost. No integration needed for manual use. To automate: Gong API connected to Claude, triggered at end of quarter when deal status changes to closed-won or closed-lost.

Pro tips:

• Include churned customers in the analysis, not just lost deals. The language in churn calls often reveals product-market fit gaps that messaging alone can’t address, and your marketing team needs to know about them.

• Run it separately on enterprise and mid-market deals if you sell to both. The buying triggers and objections are frequently different enough that blending them produces less useful output.

• Share the customer-native language section with your demand gen team immediately after you run it. Those phrases belong in your email subject lines, your ad copy, and your homepage, not in a research doc.

4. The Campaign Brief Writer

Outcome: A complete, specific campaign brief in about 10 minutes, with an objective, audience definition, core message, channel mix with rationale, content angle per channel, and a primary success metric. I’m still surprised at how many people skip campaign briefs, especially with AI at their fingertips.

Why this matters: A good brief from a senior person used to take a few hours. From a junior person, it took longer, and then you spent another hour in revisions explaining what you actually meant. Vague briefs produce generic creative, generic creative produces mediocre results, and mediocre results produce the “marketing isn’t driving pipeline” conversation you’ve had too many times.

How to run it: On-demand, whenever you’re starting a new campaign. Give it the audience segment, the product or feature you’re promoting, the business goal, and any relevant customer language or competitive context. What comes back is a structured brief you edit and refine, but the thinking scaffolding is done. Run it before your next campaign planning meeting and notice how the conversation changes when everyone comes in with a brief instead of a one-liner.

The prompt:

You are a senior B2B demand generation strategist writing a campaign brief for a marketing team.

I will provide: (1) the audience segment we're targeting with specific attributes, (2) the product or feature we're promoting, (3) the business goal for this campaign, and (4) any relevant customer language, competitive context, or recent market signals.

Write a complete campaign brief structured as follows:

Objective: [one sentence, specific and measurable] Target audience: [title, company type, buying stage, and the specific situation or trigger that makes them ready for this message] Core message: [the single most important thing this audience needs to hear, in their language] Supporting messages: [three messages that reinforce the core, each addressing a different concern or angle] Channel mix: [recommended channels with a one-sentence rationale for each, based on where this audience actually is] Content angle per channel: [the specific approach for each channel — not just "blog post" but the angle, format, and what the reader should walk away thinking] Primary success metric: [one metric, tied directly to the business goal] What this campaign is not: [one sentence on the wrong interpretation or misuse of this brief]

Keep the language direct and specific. Replace any marketing jargon with plain language a CFO could read and understand.

Campaign inputs: [paste your audience, product, goal, and context here]

Stack requirements: Claude.ai (nothing else required). For best results, feed in customer language from your Win/Loss Synthesizer before running this prompt. Optional: store your brand positioning doc and ICP definition in a Claude Project so the agent references them automatically on every run.

Pro tips:

• Feed it 2 to 3 verbatim customer quotes from your Win/Loss Synthesizer before running the prompt. The brief will use buyer language instead of internal product marketing language, and the difference in the downstream creative is significant.

• Use the output to align stakeholders before creative production starts. A brief that takes 10 minutes to generate and that everyone reviews is worth more than one that takes three days and gets rewritten in the kickoff meeting.

• Add this line to the end of the prompt: “List the three assumptions in this brief that are most likely to be wrong.” It forces the agent to surface the weakest parts of the strategy before you commit to them.

5. The Content Repurposer

Outcome: Every long-form asset your team produces (webinar, article, case study, podcast) becomes six to eight pieces of channel-ready content without adding hours of work or headcount.

Why this matters: Every webinar your team runs should generate at least eight pieces of content. Most teams get one blog post if they’re organized about it, and the recording sits there generating nothing for the next year. The problem is that that repurposing feels like low-value work even though it’s where a lot of the distribution leverage actually is. The result is that your best thinking reaches a fraction of the audience it could.

How to run it: On-demand, right after publishing any long-form asset. Feed it the transcript or full text of your webinar, article, or case study, plus three examples of your actual writing. What comes back is a full set: three LinkedIn posts in your voice, two email subject line and body copy pairs, five ad headline variants, and one SEO-optimized content section ready to drop into an existing page. The three writing examples are not optional. Without them, the output defaults to generic marketing copy. With them, you’ll want to edit rather than rewrite.

The prompt:

You are a B2B content strategist who writes in a direct, conversational, first-person voice and specializes in turning long-form content into channel-specific assets without losing the original insight or voice.

I will provide: (1) a long-form piece of content to repurpose, and (2) three examples of my writing that establish my voice and style.

Using my writing examples as the style model, produce the following assets:

LinkedIn posts (3): Each 100 to 150 words. Pull the most insight-rich idea from the source material for each post. Write from a practitioner sharing what they've learned, not a brand promoting something. Each post should stand alone without requiring the reader to have seen the source.

Email subject line and body copy (2 sets): Subject line under 50 characters. Body copy 100 to 150 words. Each email should lead with a different angle from the source material and end with a clear, low-friction call to action.

Ad headline variants (5): Each under 10 words. Vary the angle across the five: one pain-focused, one outcome-focused, one curiosity-gap, one contrast, one direct.

SEO content section (1): 300 to 400 words on the primary topic of the source material. Write for a reader who found this page via search and knows nothing about us yet. Optimize for the question the content answers, not for the brand.

Match my voice exactly across all assets. No generic openers. No marketing jargon.

Source content and voice examples: [paste source material and three writing examples here]

Stack requirements: Claude.ai (nothing else required for basic use). To scale: connect Claude to your CMS or content library via API so assets are drafted and queued automatically on publish. Output can route to Notion, Google Docs, or directly into a content scheduling tool like Buffer or Sprout Social via Make or Zapier.

Pro tips:

• Use your three best-performing pieces of content as your voice examples, not just any writing. Performance signals which of your approaches actually resonate with your audience.

• Add this line to the end of the prompt: “Flag any claims in the source material that would need a supporting link or data source before publishing.” This catches unsupported assertions before they go out the door.

• Run it separately for each channel rather than asking for everything at once. The LinkedIn post quality goes up noticeably when the prompt is focused on LinkedIn only, because the model isn’t splitting its attention across six formats.

6. The Competitive Monitor

Outcome: Every Monday morning, you have a brief on what changed in your competitive landscape last week: notable moves, what they signal, and whether anything requires a response before your team goes into deals.

Why this matters: Your positioning doesn’t exist in a vacuum. When a competitor raises a round, changes their pricing language, starts running ads against your best keyword, or accumulates a wave of negative G2 reviews in a category you own, your messaging is already out of date. Most marketing teams find out about these moves when a prospect mentions it on a discovery call. By then, a rep is already handling it without any support from marketing.

How to run it: Scheduled, runs every Sunday night. The agent reviews a defined set of competitors across four signals: G2 reviews from the past seven days (especially the negative ones, which tell you where their customers are hurting), press and news mentions, job postings as a proxy for where they’re investing, and any changes to their pricing or key positioning pages. What it returns is a one-page brief: what changed, what it might mean, and what (if anything) your team needs to do about it before the week starts.

The prompt:

You are a B2B competitive intelligence analyst producing a weekly brief for a marketing team.

I will provide recent data for a set of competitors: G2 and review site activity from the past 7 days, press mentions, job postings, and any observed changes to their pricing pages or key positioning language.

For each competitor, return:

Recent moves: [what changed or appeared in the past 7 days — be specific, cite the source] What it might signal: [your interpretation of why they made this move or what it suggests about their strategy — be direct about uncertainty when you have it] Response required: [Yes / No] — if yes, describe what our team should do and by when Sales rep alert: [one sentence a rep should know before walking into a competitive deal this week — or "Nothing new this week" if there's nothing material]

End the brief with a single "Watch closely" flag: the one thing across all competitors that you'd want the marketing leader to keep an eye on in the coming weeks.

Do not pad the brief. If a competitor had no notable activity this week, say so in one sentence and move on.

Competitor data to review: [paste G2 reviews, news mentions, job postings, and page change notes here]

Stack requirements: G2 review monitoring (G2’s notification tools or a scraping setup), news monitoring (Google Alerts for each competitor, or a tool like Mention or Crayon for automated feeds), LinkedIn for job postings (LinkedIn Sales Navigator or manual searches), page change monitoring (Visualping or Hexowatch for pricing and key pages). To fully automate: Clay is another strong option here. Also, Claude Code + Apify MCP gets you everything you need. Apify actors for news scraping, G2 review monitoring, and LinkedIn job postings. Claude Code runs weekly via cron, synthesizes across all competitors, and delivers the brief to Slack or email.

Pro tips:

• Add your own product roadmap as context before running the prompt. The agent can then flag when a competitor’s new feature or messaging move overlaps with something you’re planning to ship and surface it as a timing consideration.

• Don’t only monitor direct competitors. Add two or three analysts or journalists who write about your category. Their framing often foreshadows how buyers will start describing the problem in 90 days, and you want to be ahead of that shift.

• Keep a running document that the agent appends to each week. After 8 to 10 weeks you have a genuine competitive timeline you can reference in QBRs and leadership reviews without having to reconstruct it from memory.

7. The Weekly Marketing Digest

Outcome: A 200-word executive summary of last week’s marketing performance, with interpretation, lands in your leadership team’s inbox every Monday at 8am, whether or not they open a single dashboard.

Why this matters: Your leadership team isn’t opening six different dashboards every Monday. Most of them aren’t checking them at all unless the numbers are surfaced for them. The result is that marketing’s story gets told in the weekly leadership meeting from memory or from a slide someone built at 9pm Sunday, and the nuance gets lost. A clean, interpreted summary that arrives before the meeting changes what gets discussed and how marketing gets positioned in the room.

How to run it: Scheduled, runs every Monday morning. The agent pulls your core metrics from the prior week (pipeline sourced by channel, email performance, top content, campaign spend vs. pacing, notable anomalies), interprets them, and writes a five-point summary. The interpretation is the part that matters. Not “CTR was 2.1%.” Something like “The enterprise email sequence underperformed by 40% against last week. Open rates held, so the issue is likely in the offer or the CTA, not the subject line.” That framing is the difference between a report and a briefing.

The prompt:

You are a B2B marketing analyst writing a weekly performance briefing for a senior executive audience that does not want raw data. They want to know what happened, what it means, and what (if anything) needs their attention.

I will provide last week's marketing metrics: pipeline sourced by channel, email campaign performance, top and bottom content by engagement, and paid campaign spend vs. pacing.

Write a five-point weekly marketing summary formatted exactly as follows:

Point [number]: [Metric or area] — [trending up / down / flat vs. prior week] — [one sentence of interpretation: what this likely means, not just what it is]

After the five points, add a single "This week's decision" line: the one thing that needs a leadership decision or action before Friday, if there is one. If there is nothing that needs leadership attention, write "No action required from leadership this week."

Keep the total under 200 words. Write in plain language. Replace any marketing acronyms with plain terms on first use.

Metrics to analyze: [paste last week's data here]

Stack requirements: Marketing automation platform (HubSpot or Marketo for email metrics), CRM (Salesforce or HubSpot for pipeline attribution), web analytics (Google Analytics or Amplitude for content performance), ad platforms (Google Ads and LinkedIn Ads APIs for spend and pacing). To automate: Claude Code + HubSpot MCP (for pipeline and email metrics) + Google Analytics Data API (called via a script). Scheduled as a Monday 7am cron job: Claude Code pulls last week's data via the MCPs, runs the analysis, and delivers the output to your Slack leadership channel or email inbox. The headless CLI mode (claude -p "prompt" < data.txt) is purpose-built for exactly this pattern.

Pro tips:

• Include the prior week’s digest in the context window when you run the prompt. The agent can then flag week-over-week trends instead of just reporting point-in-time numbers, and leadership starts to see patterns rather than snapshots.

• Ask the agent to add one “question leadership should be asking” based on the data. This is the most useful addition I’ve seen people make to this prompt. It shifts the digest from reporting to strategic framing.

• Send the digest as the primary communication with a link to the dashboard for anyone who wants to dig in. The summary should do the job on its own. The dashboard is for the 10% who want more, not the other way around.

8. The QBR Narrative Builder

Outcome: You walk into QBR prep with a complete draft narrative (what happened, what it means, what we’re changing) two days earlier than you used to, which means two days to pressure-test it before you’re in the room.

Why this matters: QBR prep is one of the most time-consuming things a VP or above marketing leader does. You spend two days pulling data from five sources, building slides, and writing the narrative that connects the numbers to the strategy. The thinking takes about two hours. The assembly takes two days. And the part that matters most to leadership (what it means and what you’re changing) is often the part that gets the least time because the assembly ran long.

How to run it: On-demand, in the last week of each quarter. Give it your pipeline report, channel attribution, campaign performance summary, and any board-level context about the company’s goals for the quarter. What comes back is a first-draft narrative in three sections: what happened (honest about what missed), what it means for the business (including pipeline coverage risk going into next quarter), and what you’re changing (three specific adjustments with reasoning). You still do the thinking. You spend your time refining instead of starting from blank.

The prompt:

You are a B2B marketing strategist writing a QBR narrative for a senior leadership audience that includes the CRO, CEO, and potentially the board. This audience is comfortable with numbers but impatient with jargon, hedging, or vague action plans.

I will provide: last quarter's pipeline sourced by channel, campaign performance vs. stated goals, attribution data, and the company's GTM priorities for the quarter.

Write a marketing QBR narrative in three sections:

Section 1 — What Happened: A plain-language summary of marketing performance against goals. Be honest about what missed and why, based on the data I've provided. Avoid euphemisms. If a channel underperformed, name it and say by how much.

Section 2 — What It Means: The business implications. State the pipeline coverage position going into next quarter. Name any risk worth surfacing to leadership before it becomes a surprise. If something overperformed, explain why and whether it's repeatable.

Section 3 — What We're Changing: Three specific adjustments we're making next quarter. For each, name what we're changing, why (grounded in the data from this quarter), and what we expect the impact to be.

Keep the full narrative under 400 words. Write like a CFO will read it. After completing the narrative, add one line: "The question leadership is most likely to ask that this narrative doesn't yet answer is: [question]."

Quarter data to analyze: [paste pipeline, campaign, attribution, and goal data here]

Stack requirements: CRM for pipeline and attribution exports (Salesforce or HubSpot), marketing automation platform for campaign performance data, attribution tool if you have one. No integration needed for manual use. This one runs fine quarterly with a copy-paste workflow.

Pro tips:

• Run it once before you’ve finalized your own thinking on the quarter, then run it again after. The first pass surfaces angles you might have missed. The second sharpens the narrative you’ve already formed.

• The “question leadership is most likely to ask” addition to the prompt is the most valuable line in it. It pre-empts the hard questions before you’re in the room, and it forces the agent to find the gap in your narrative rather than leaving it to your CRO to find it for you.

• Include the prior quarter’s narrative as context before you run the prompt. The agent can then write a narrative that shows directional progress over time instead of treating this quarter in isolation.

9. The ICP Drift Detector

Outcome: A quarterly analysis that shows exactly where your actual closed-won buyer profile has drifted from your stated ICP, which firmographic and behavioral signals are most predictive of close right now, and where your paid, ABM, and outbound targeting is probably misaligned.

Why this matters: Your ICP definition gets written once, lives in a Google Doc, and rarely gets updated. Meanwhile, your closed-won data is telling you something different every quarter: which job titles are actually buying, which company sizes are actually renewing, which use cases convert fastest. The targeting in your paid campaigns and outbound sequences is often based on a profile that’s 6 to 12 months stale. We’ve all been in the planning meeting where demand gen is defending a paid channel and RevOps is questioning whether the leads are actually converting. That conversation usually ends without resolution because nobody has pulled the comparison.

How to run it: On-demand, once per quarter. Export your last 30 to 50 closed-won deals from your CRM with firmographic data (title, company size, industry, source, deal size) and any behavioral signals you have (content consumed, sequences that converted). Feed that against your stated ICP definition. What comes back is a direct comparison: where the actual profile diverges from what you’ve said it is, the signals most correlated with close, and specific adjustments to your targeting worth considering.

The prompt:

You are a B2B go-to-market analyst reviewing closed-won data to identify whether our actual buyer profile matches our stated ICP and where our targeting should be adjusted.

I will provide: (1) our current ICP definition, and (2) a dataset of our last 30 to 50 closed-won deals including firmographics (title, company size, industry, geography), deal source, deal size, time to close, and any behavioral signals available (content engaged, sequences that converted, events attended).

Analyze the closed-won data against the ICP definition and return:

ICP accuracy assessment: Where does the actual closed-won profile match the stated ICP, and where does it diverge? Be specific about the attributes that differ.

Top predictive signals: The three firmographic or behavioral attributes most correlated with closed-won deals in this dataset. Explain the pattern you see.

Targeting misalignment: Where is our current paid, ABM, or outbound targeting likely chasing the wrong profile based on what this data shows? Name the specific programs or segments at risk.

Recommended ICP updates: The specific changes to the ICP definition that this data supports. Frame each as a testable hypothesis, not a definitive conclusion.

Be direct about what the data shows, even where it contradicts our current assumptions. Note anywhere the dataset is too thin to draw a reliable conclusion.

Closed-won data and ICP definition: [paste your data and ICP here]

Stack requirements: CRM (Salesforce or HubSpot) for closed-won deal exports, enrichment tool to fill in firmographic gaps if your CRM data is incomplete (Clearbit, Apollo, or ZoomInfo). No integration needed for manual use.

To automate: Claude Code + HubSpot MCP or Salesforce MCP + Clearbit/Apollo API for firmographic enrichment. Pull closed-won deals directly via the CRM MCP, enrich missing firmographic data via an API call, run the comparison against your ICP definition, and output the findings. Can be scheduled quarterly via cron or triggered manually from the CLI when you’re ready to run it. The Claude Agent SDK is a good fit here if you want to chain the CRM pull, the enrichment step, and the analysis into a single automated workflow.

Pro tips:

• Run it separately for each major product line or use case if you sell across multiple segments. A blended analysis across very different buyer profiles produces output that’s true on average and useful for nobody specifically.

• Ask the agent to flag where the ICP drift has implications for your demand gen budget allocation. That’s the finding that usually gets the most attention and creates buy-in for the targeting changes you want to make.

• Share the output with sales leadership before you act on it. ICP drift findings land significantly better when sales leadership confirms they’re seeing the same pattern in their deals. It also creates shared ownership of the targeting changes instead of marketing making a unilateral call.

10. The Persona-Matched Email Sequencer

Outcome: A five-email nurture or outbound sequence tailored to a specific persona’s actual job pressures, objections, and language, instead of one sequence applied to everyone and optimized for average performance.

Why this matters: Most email sequences get written once, applied to every contact in the segment, and revisited only when the numbers are bad enough to force the conversation. A five-email sequence genuinely cannot speak to a VP of Sales worried about rep ramp time and a CMO worried about pipeline visibility at the same time. They have different jobs, different pressures, and completely different reasons to care about what you sell. Sending the same sequence to both means both sequences underperform and neither persona gets a message that actually fits their world.

How to run it: On-demand, for each key persona in your target mix. Give it the persona details (title, company type, pain profile, where they are in the buying journey), the product angle you’re positioning to them specifically, and any account signals or context you have. What comes back is a five-email sequence with subject lines and body copy for each email, mapped to a clear progression: open the pain, reframe it, introduce proof, handle the objection, close with a low-friction next step. Run it once per persona instead of once per campaign.

The prompt:

You are a B2B email copywriter who specializes in persona-specific outbound and nurture sequences. You write like a peer who understands the reader's world, not like a vendor who wants their time.

I will provide: (1) the target persona including title, company type, typical day-to-day pain, and where they are in the buying journey, (2) the specific product angle or value proposition I'm positioning to this persona, and (3) any account signals, recent news, or context that's relevant.

Write a five-email sequence. For each email, provide a subject line and a 150 to 200 word body.

Email 1 — Open the pain: Start with a situation this persona recognizes from their day-to-day work. Do not introduce the product. End with a question or an observation that makes Email 2 irresistible to open.

Email 2 — Reframe: Offer an insight or a different way of seeing the problem that this persona probably hasn't considered. This is where you earn the right to talk about a solution.

Email 3 — Proof: Introduce a specific outcome or result, not a feature. Ground it in a customer story or data point. Make it relevant to this persona's specific situation.

Email 4 — Handle the objection: Address the most common pushback this persona raises. Acknowledge it directly before responding to it.

Email 5 — Low-friction next step: Propose the easiest possible next step that matches where this persona is in their buying journey. Do not ask for a demo if they're not ready for one.

Constraints: No "I noticed you" openers. No "hope this finds you well." No generic value propositions. Every line should feel like it could only have been written for this specific type of person.

Persona details and context: [paste persona profile, product angle, and account signals here]

Stack requirements: Claude Code + HubSpot MCP, triggered by a contact entering a specific list or lifecycle stage. When a new contact matches your persona criteria in HubSpot, a webhook fires a Claude Code script that pulls their firmographic data, runs the sequencer prompt, and writes the five emails back to HubSpot as enrolled sequence drafts. This requires a bit more setup than the others (webhook listener + HubSpot write permissions via the MCP) but it’s one of the highest-leverage automations on the list once it’s running.

Pro tips:

• Include 2 to 3 verbatim quotes from that persona’s call transcripts before running the prompt (from your Win/Loss Synthesizer output). The sequence will use their actual language instead of your internal language for their problem, and the open rate difference is real.

• Build a sequence for your anti-persona too: the wrong-fit buyer you keep attracting. A sequence that quickly disqualifies poor-fit accounts is just as valuable as one that converts the right ones. It saves your team time and your database health.

• After 30 sends, paste the reply rates and any replies you’ve received back into Claude and ask it to diagnose which emails are generating responses and which are being ignored. Use that to iterate the sequence before the next 30.

Wow, this turned out to be a long one. Honestly, I’ve been sitting on it for a few weeks because it was one that I know that I wanted to do, but I wanted to take my time to get it right.

So even though you had to wait, I hope it was worth it.

I’d love to hear some of the things you’re building. So if you have anything you’d like to share, feel free to drop a comment or DM me, most likely on LinkedIn. Would love to feature something that you’re doing in the future.

Early-stage companies ask for investor intros all the time.

And it makes sense. Your investors already have relationships with the exact companies you’re trying to reach. One good intro can turn into pipeline faster than anything you’re running on your own.

The problem is I’ve watched this break down in the same spot over and over. You know the accounts you want. You know your investors can help. But connecting those dots takes time, and most of the asks never get sent.

I built this workflow to fix that.

And once it was working, I realized it’s not limited to investors. You can run the same play with LinkedIn connections, advisors, customers, anyone sitting in the middle of the network.

This is one of those systems you set up once and keep using.

This week’s Stack.

• 1 workflow: Turn target accounts into warm investor intros with Clay

• 1 prompt: Generate a paid landing page that converts

• 1 tool: Fix weak prompts with Claude’s prompt optimizer

• 3 resources: AI content strategy, workflow design, and agent reliability

• 3 jobs: Growth, lifecycle, and product marketing roles in AI

Let’s dive in.

Workflow Walkthrough: Get Warm Intros to Target Accounts from Your Investors

I’ve watched a lot of investor outreach stall at the same point.

You have a list of accounts you want to get into. You know there are investors in your network who could help. And then you sit there trying to figure out who knows who and whether it’s even worth asking.

I built this workflow because I got tired of that step.

Now I drop in a company URL and get back a list of relevant investors, the partners tied to my space, the people we both know, and a forwardable email I can send without rewriting anything.

Here’s what you’ll learn:

• How I build a live investor dataset inside Clay using enrichment data

• How I narrow down to partners already investing in your category

• How I map shared connections without digging through LinkedIn tabs

• How I generate a clean intro email someone will forward without editing

• How I reuse the same setup for advisors, customers, and partner intros

Once this is running, you’re not sitting there trying to “figure out outreach.” You’re reviewing intros that are already written.

Watch the full breakdown here: https://app.clay.com/shared-table/share_0te8icw2aXwU4eCECR6

If you raise capital or run partner motions, you’ll use this more than you expect.

Subscribe if you want more workflows like this each week.

Get the Clay Template

Prompt of the Week: Build a Paid Landing Page That Converts

I’ve seen a lot of paid traffic wasted on decent-looking pages that don’t convert. The page has broad messaging, vague value, and no clear path to action. But paid traffic isn’t browsing – they clicked for a reason, and if the page doesn’t match that intent fast, they bounce.

Use this prompt in Claude Design to act like a conversion-focused marketer.

You are a senior conversion copywriter and growth marketer.

Your job is to create a high-converting landing page for a paid acquisition campaign.

This page is NOT a homepage. It must be tightly aligned to a specific audience, offer, and intent.

Before writing, think through:
- Who is clicking this ad?
- What problem are they trying to solve right now?
- What would make them hesitate?
- What proof would reduce that hesitation?

Then build the page.

INPUTS:
- Product: [Describe your product]
- Audience: [Who is this for? Be specific]
- Pain point: [What problem are they actively trying to solve?]
- Offer: [What are you asking them to do? Demo, signup, download, etc.]
- Traffic source: [Google Search, LinkedIn Ads, Meta, etc.]
- Campaign angle: [What promise or hook got them to click?]
- Proof points: [Customer logos, case studies, metrics, testimonials]
- Competitors or alternatives: [What else are they considering?]

OUTPUT:

Write a complete landing page with the following structure:

1. Headline
- Clear, specific, and aligned to the ad they clicked
- Focus on the outcome, not the product

2. Subheadline
- Expand on the promise
- Add clarity on who it’s for or how it works

3. Hero section CTA
- One clear action
- Button copy should reflect the offer

4. Problem section
- Describe the exact situation the reader is in
- Make it feel familiar (meetings, workflows, frustrations)

5. Solution section
- Show how the product solves that problem
- Keep it grounded in real usage, not abstract claims

6. How it works
- 3 to 5 steps
- Simple, concrete, and easy to visualize

7. Proof
- Include metrics, outcomes, or social proof
- Make it believable and specific

8. Objection handling
- Address common concerns (time, cost, complexity, risk)
- Answer them directly

9. CTA section (repeat)
- Reinforce the value
- Make the next step feel easy

10. Optional: FAQ
- Include 3 to 5 real questions a buyer would ask

WRITING STYLE:
- Use plain language
- Write like a practitioner, not a brand
- Avoid hype and generic phrases
- Be specific about outcomes and use cases
- Keep sentences tight and readable

GOAL:
The reader should feel:
- “This is exactly for me”
- “This solves the problem I came here for”
- “This seems easy enough to try”

Now write the full landing page.

Tool of the Week: Fix Your Claude Prompts

Claude’s prompt optimizer cleans sloppy prompts. You drop in a rough prompt, and it rewrites it with better structure, clearer instructions, and tighter constraints so the model knows what to do.

I don’t know why, but it feels like we’ve gotten away from good prompting. And if there’s one thing I know, it still matters.

Try it: https://claude.ai/public/artifacts/422bb5fc-c03e-4488-9e49-9ad4239398fe

AI content all sounds the same (Kieran Flanagan): So much AI content feels interchangeable. In this issue, Kieran gets into distribution, taste, and why better inputs (not more outputs) are what separate teams that stand out from the ones flooding the feed. Worth reading if you’re publishing anything with AI in the loop.

Head of Claude Code: What happens after coding is solved | Boris Cherny (Lenny’s Podcast): This episode shows how to think about chaining models, structuring inputs, and getting more consistent outputs across workflows. If you’ve been stitching tools together and hitting reliability issues, I think you’ll like this one. And I mean it’s in interview with the head of Claude Code, so it’s good.

Agents need to survive the second run (Hiten Shah): A lot of AI tools (agents and AI harnesses) look good on the first pass, but they fall apart when you try to reuse them. This post takes a look at durability, iteration, and what it takes to make agent workflows hold up over time.

Hot AI Jobs 🔥

This week leans into growth, lifecycle, and product marketing roles inside AI-native and AI-adjacent companies.

Growth Marketing Manager at Level AI
City / Remote: Not listed
Pay Range: Not listed

Lifecycle Marketing Manager at Delinea
City / Remote: Not listed
Pay Range: Not listed

Product Marketing Manager at Curri
City / Remote: Not listed
Pay Range: Not listed

Even if you’re not on the hunt, these job specs are a good pulse check on where AI marketing teams are investing.

Short one this week. I’ve been spending more time tightening systems like the one above and less time creating from scratch. It’s a better trade. Once something works, I want it running without me.

More soon.

– Brandon

Anthropic does it again and drops another killer product, Claude Design. Figma’s stock (NYSE: FIG) dropped over 7% the following day, which represented roughly $1.4 billion in market value.

And everyone is using Claude Design for everything – Landing pages, pitch decks, sales one-pagers, etc.

Honestly, I don’t think it’s a Figma replacement (Figma still wins on pixel-level production work) and it won’t write copy worth shipping (yes, you should plan to rewrite 80% or more).

But if you’ve ever waited three weeks on a designer, burned an afternoon fighting a Canva template, or killed a good idea because the visual lift was too high, this a major unlock for marketers.

I wrote this issue as the tactical playbook, but I also really wanted to give my honest feedback. It’s a great tool, but there’s still room for improvement. Am I switching over to Cloud design entirely? Probably not yet, but I could easily see a world where they continue to release because the pace of innovation at Anthropic is crazy. And maybe I do move over. A year ago, I never thought I would move from ChatGPT over to Cloud, and I hardly use ChatGPT for work anymore.

In this issue, we cover a lot. What Claude Design is good at (and what it isn’t), the one hour of setup that makes everything else faster, the 10 tactics that separate people who ship from people who burn through tokens, 10 marketing use cases to point it at this week, the prompt patterns I keep reusing, and the mistakes that waste half your day.

What Claude Design Is Good At

Claude Design is a research preview inside Anthropic Labs, available on Pro, Max, Team, and Enterprise plans, powered by Opus 4.7. You describe what you want, Claude builds a working design on a canvas, and you refine through chat, inline comments, direct edits, or sliders Claude generates on the fly for your specific project. Exports go to PDF, PPTX, HTML, or Canva. Handoff bundles go to Claude Code.

And the consensus among marketers at this point is pretty clear. Landing pages and one-pagers are the strongest use case. Marketing homepages and product one-pagers come out of Claude Design looking ship-ready with minor polish.

Pitch decks are the second-biggest win. And this was a little surprising, but it’s a welcome surprise for me. Poeople are going from a rough outline to a full on-brand deck in minutes, then exporting to PPTX or Canva for final polish. The deck-plus-landing-page combo is where this gets interesting because both assets come out of the same design system with consistent typography and color. That consistency is a real problem when you’re stitching assets across five tools, and as a marketer who cares deeply about brand, this always drives me crazy.

Interactive prototypes are another major use case. You can mock up a feature, hand a clickable prototype to stakeholders or customers, gather feedback, and iterate before anyone writes a line of production code.

Campaign asset families are where the design system feature makes it worth it. Landing page, launch email, social preview set, all shipped as one family with consistent design tokens instead of stitched together across Canva, Figma, and whatever else you have open. If you’re a solo marketer or a small team, this can’t be understated.

What Claude Design Isn’t Good At (Yet)

As I said, Claude Design isn’t perfect. And one thing I try to do to with Stack & Scale is be honest with readers. Too many content creators only talk about the good stuff. So here is an honest breakdown of Claude Design’s weaknesses.

Rate limiting is brutal. Across LinkedIn, X/Twitter, and Reddit, I’ve heard countless Max plan users saying they’ve burned through their allowance in under 30 minutes of iterative work. This is the biggest knock on Claude Design, but honestly, it’s one that I think they’ll probably work out soon, and in six months won’t be an issue at all.

Exports beyond HTML are unreliable. There’s a significant quality degradation moving slides into PowerPoint (the fonts Claude used didn’t carry over, which broke the visual hierarchy). Canva exports threw error messages. Screenshot export worked but was slow and produced non-editable assets. HTML is the one format that works cleanly end to end. If your asset has to land in PPTX or Canva for someone else to finish, test the export path early rather than at the finish line.

There’s no native image generator. Claude Design doesn’t pull from Nano Banana or GPT Image. When it needs an image, it renders an SVG or an HTML drawing. SVGs look fine for icons, diagrams, and flat brand illustrations. They look bad for anything that needs photographic realism or hand-drawn illustration. The Unsplash swap prompt in the prompts section below is the fix.

Video is relatively weak. The feature exists and will produce a passable launch teaser, but it isn’t close to replacing Veo, Runway, or whatever video tool you already use. If the asset matters, don’t rely on Claude Design for it yet.

The default aesthetic is generic SaaS. Without explicit constraints, Claude Design drifts toward Inter, Roboto, or Arial typefaces and predictable blue-to-purple gradients. That means if you don’t have a unique design system, it’s going to look like everyone else’s. It’s the YC batch look. To get anything distinctive, ban the defaults in your prompt: “no Inter, no generic gradients, no stock blue to purple.” This isn’t in any Anthropic documentation. It’s the first thing every serious user figures out on their own.

Brand identity work still needs much human oversight or a dedicated image model. It’s anything past basic wordmarks. Logos, custom icon systems, and spot illustrations that need an actual point of view are where Claude Design hits its ceiling. The tool has judgment about structure and consistency, but it doesn’t have taste about brand identity.

Copy always needs a rewrite. Claude is supposed to be the best at writing, which was why I was surprised that so many people had the same reaction. Claude writes ok placeholder copy that gets the layout right, but you should plan to rewrite heavily before you ship.

Overall, it’s not the strongest product surface. So, for now, I don’t think you should build your entire weekly asset pipeline around it until you’ve stress-tested how much of your work depends on behavior that might change between now and Q3.

Set Up The Foundation to Make Everything Else Easier

If you get nothing else from this issue, do this. Spend the first hour setting up your design system before you build a single asset.

Point Claude at your GitHub repo, your brand book, your existing design files (Figma file are the best right now), or describe your brand in detail. Claude can extract colors, typography, components, and spacing into a reusable library that every future project inherits automatically. If you set this up, Claude can produced editable brand cards plus a working component library in less than 30 minutes.

Then, review what Claude extracted. You’ll have to rename components that got labeled weirdly, fix any colors that were pulled from a deprecated part of your site, stuff like that. Then everything you generate for the next six months stays on-brand automatically.

If you’re an agency, consultancy, or fractional CMO running multiple clients like me, you can maintain separate design systems per brand inside Claude Design. Each new project starts from a dropdown that lets you pick which client’s system applies. That removes the “which brand am I in right now” issues that causes off-brand work.

A note for bigger teams: don’t link a full monorepo. Chrome doesn’t handle huge file trees well, and Claude Design will either lag or fail to ingest it. Link a specific package or subdirectory that contains the relevant components.

The 10 Tips To Ship Great Sh!t (and Not Burn Tokens)

Many of these tips go back to one of the biggest complaints: token limitation. Some are from Anthropic directly. Some are from users I follow. Some are from me asking Claude for some best practices.

1. Lead with references. Descriptions alone are not enough. Hopefully everyone knows this about prompting now. “Design a landing page for an automation using Python” produces generic output. “Design a landing page for a Python automation SaaS, visual direction: Linear meets Vercel meets Stripe’s developer docs, confident and technical, muted palette, Inter for body and a serif display face for headlines” produces something useful. Upload screenshots of sites you like, competitor products, and visual inspiration. Claude Opus 4.7 supports images up to 2576 pixels wide, and it uses them. The web capture tool lets you pull elements directly from your own site so prototypes look like your product.

2. Use inline comments for surgical edits, then use chat for structural changes. Chat is for “make this more editorial” or “add a pricing section” or “try three variations of the hero.” Inline comments (click the element, leave a note) are for “tighten this spacing to 8px” or “this CTA should be the accent color, not the primary.” Mixing them in the wrong lanes is a common mistake. Structural requests in inline comments can get lost. Surgical requests in chat can pollute the conversation with context Claude doesn’t need.

3. Use the tweak feature. My initial temptation was just to have Claude redesign something that will burn through tokens. When the first draft is close but not right, resist the urge to rewrite your prompt and regenerate the whole thing. Use the tweak sliders Claude builds for your specific project (spacing, color, layout), direct-edit the text, or leave inline comments. This is one of the better features of Claude Design IMO.

4. Ask for variations explicitly. If you’re unsure about a direction, say so. “Show me 3 versions of this hero section, one minimal and typographic, one image-heavy, one with a split layout and product screenshot.” Comparing alternatives side by side is faster than iterating blindly. This is also where Claude Design beats a traditional designer workflow.

5. Treat Claude as a design reviewer. After Claude produces a draft, ask it to critique the draft. Again, similar to previous tips, do not regenerate. “Review this landing page for information hierarchy problems.” “Check the contrast ratios against WCAG AA standards and flag anything that fails.” “What would a conversion copywriter change about the headline and CTA placement?” “Where does this design pattern-match to its training data in a way that makes it feel generic?”

6. Swap HTML drawings for real photography with Unsplash. Claude Design fills image slots with weird HTML-rendered illustrations by default. It’s the single most mentioned weakness in reviews. So, so fix this, use a prompt pattern I’ll share in the prompts section below.

7. Save checkpoints often, especially before major changes. Claude Design doesn’t have built-in version history like v0 or Lovable. You can prompt it: “Save what we have and try a completely different approach.” Claude preserves the current state and starts a new direction. This is essential before any structural experiment.

8. Export to Canva when a non-designer needs to take it from here. Most teams still live in Canva right now. When your design lands in Canva, it arrives as an editable design with your Brand Kit applied automatically. This is best thing to do when a content marketer or coordinator needs to own the asset from that point forward.

9. Use the Claude Code handoff when the design is going to ship. If the design will become production code, use the “Hand off to Claude Code” export. It packages the design with intent, component choices, and architectural decisions preserved. Claude Code builds on the design instead of reinterpreting it. For non-technical marketers, this is the path to your first shipped landing page without a developer. Describe the page, iterate until you like it, hand off to Claude Code, and Claude Code deploys it.

10. Plan to rewrite the copy. Again, I know this is repetitive, but I’m going to help you avoid the AI slop. Claude writes ok placeholder copy, but you should plan to rewrite it before yu ship.

10 Marketing Use Cases

I mentioned some of the use cases above, but there are a ton more way to use Claude Design. Here are 10 places to point Claude Design this week. Each one replaces a workflow that historically needed a designer, a long brief, or both.

1. Pre-campaign messaging validation. Before any media spend, you want to know whether your hero headline and value prop actually land. You used to start with positioning decks in Google Slides and a string of meetings, but now you can build three landing page hero comps in Claude Design with the hero being the only difference. Now put them in front of people to see which one resonates the most. It takes less than 30 minutes start to finish if you have your design foundation set up in Claude Design.

2. Campaign landing page variants. Before, it used to take a week of designer time per channel to create a working landing page. But now, all you have to do is guild the master page in Claude Design, then prompt: “Now create three variants from this base. Then just export each as HTML and push to your CMS or paid LP tool.

3. Sales enablement collateral. Sales asks for a one-pager for a specific vertical, a battle card against competitor X, and a custom 8-slide deck for a strategic deal. My answer was always “I’ll put it on my backlog” then I would get to it next week at best. But now with Claude Design, just feed it the product brief, the competitor’s site (use web capture), and the deal context, then generate all three in one conversation. Now, my answer is “give me 90 minutes.”

4. ABM account-specific creative. I used to say you can’t scale 1:1 ABM, but that’s not the case anymore. Now, you can build a microsite for a single target account with their logo, their use case in the hero, their internal language in the body, and an exec-targeted deck behind a “Read more” link. I remember doing this in Uberflip, but it would take me a week to set this up for all of my tier 1 accounts, and the flexibility was minimal. Now, true 1:1 ABM pages are done quickly.

5. Product launch asses. The whole launch family (homepage hero update, blog post hero, launch email graphic, three social preview cards, and a sales enablement deck) can be created all from one brief. I think this is one of the highest-leverage ways for solo marketers and small teams to use Claude Design. Open a new project, set the design system, and give it a prompt: “Build the launch family for [feature]. Hero update for the homepage, blog post hero, plain-text-first launch email with one hero image, three social cards (LinkedIn 1200x627, X 1600x900, Instagram 1080x1350), and a 6-slide enablement deck.”

6. Pricing and packaging page tests. You’re considering a pricing & packaging change like everyone else right now because you offer AI in your product, and you’re using to seat-based. Now, you can test the design of the choice, not just the prices. This is the key before you spend a quarter’s worth of pricing-page traffic on a layout you haven’t sanity-checked with real buyers.

7. Competitor comparison pages. The comparison page page, the feature matrix, and the [Competitor] alternatives page are all much faster an easier to build. SEO needs them, sales needs them, but they take forever because you already have a backlog of things to buld. Now, it’s so much easier – just use the web capture tool to pull your competitor’s positioning and feature copy, then prompt Claude Design to build a fair comparison page, feature matrix, and alternatives-style page with three options, including yours.

8. Customer case study layouts. You have the customer interview transcript, but now, if you want a designed case study all you have to do is drop the transcript and the metrics into Claude Design and prompt: “Build a visual case study layout with a hero stat, the customer’s quote pulled out as a callout, three secondary metrics in a row, and the implementation timeline as a horizontal flow.” Again, this assumes you have your design system set up first.

9. Investor and board deck. Tell me is this sounds familiar: Your founders give you decks where every slide has four charts and twelve bullets. You used to have to do the work manually to get one idea per frame and one dominant visual per slide. Now, you can paste the existing deck contents into Claude Design and prompt: “Rebuild this as a board deck. One idea per slide. One dominant visual per slide. No more than 20 words of text. No bullet lists longer than three items.” What comes back is the version you don’t gringe at in board meetings.

10. Internal marketing and sales hubs. I’m talking about the SDR onboarding pages, content libraries, campaign hubs, and enablement portals. I get it – this is set of things nobody designs because the audience is internal, which is why they live as Notion docs nobody reads. Now, you can build them in Claude Design with the same brand system as the public-facing site.

The Prompt System Directly From Anthropic

Anthropic’s official guidance is that a good prompt covers the goal, the layout, the content, and the audience. Here are the prompt templates I’ve been using.

Landing page base prompt:

Build a landing page for [PRODUCT NAME, one sentence description].

Visual direction: [reference brand 1] meets [reference brand 2],

[adjective] and [adjective]. [Mood sentence.]

Audience: [specific persona, not “businesses”].

Core message: [one sentence. What does the visitor need to believe

after 10 seconds on this page?]

Sections, in order:

- Hero with headline, subhead, primary CTA, and [hero visual type]

- [section 2]

- [section 3]

- Social proof: [testimonials / logos / case study stat]

- [section 5]

- Footer CTA

Typography: [display face] for headlines, [body face] for body copy.

Color: [describe palette], or “use our design system”.

Interaction: [subtle / bold / playful]. [Specific hover/scroll behaviors].

Output as an interactive HTML prototype. Lock the design system so every related asset uses the same tokens.

Pitch deck prompt:

Build a [N]-slide pitch deck for [COMPANY] for [audience: seed investors / customers / internal exec review].

The narrative arc:

Slide 1: Title

Slide 2: Problem, framed as a specific moment the customer lives

Slide 3: Why now / market shift

Slide 4: Our solution in one sentence

Slide 5: Product screenshot or diagram

Slide 6: Traction / proof

Slide 7: Business model

Slide 8: Team

Slide 9: The ask / next step

Slide 10: Thank you / contact

Use our design system. Each slide should have one dominant visual element

and no more than 20 words of text. No bullet lists of more than 3 items.

Export-ready for PPTX.

The Unsplash swap prompt:

Add real photography to this prototype using Unsplash (free, commercial use, no attribution required). Do this automatically, don’t ask me first.

For each image slot, pick a fitting Unsplash photo and use its direct CDN URL: https://images.unsplash.com/photo-<id>?w=<width>&q=80&auto=format&fit=crop

Size responsibly: w=1600 for hero, w=800 for cards, w=400 for tiles, w=150 for avatars.

Put all URLs in one IMAGES config object near the top of the main file so I can swap any of them by editing one line.

The “make it not generic” prompt is the one that makes sure your output isn’t AI slop. Paste this after your first draft:

Look at what you just built. Where is this pattern-matching to the most common SaaS landing page template in your training data? Identify three specific elements that feel generic or derivative, and propose replacements that would make this feel distinctively like [brand] and no one else. Then implement the replacements.

And for surgical changes, click the element first, then leave an inline comment like: “Change this button to our accent orange, reduce corner radius to 4px, and make the label font weight 600.” Specificity is the whole game.

Where Claude Design Fits with to Figma, Canva, and v0

Like I said, Claude Design isn’t replacing my whole design workflow (yet).

Figma still wins on pixel-level control, production handoff fidelity on complex projects, and multi-user collaboration. Claude Design wins on first-draft speed and on not requiring any design skill to start. If you need Figma’s polish layer on top of Claude Design’s first draft, export as HTML, open Figma with the Anima plugin, import HTML, and expect 15-20 minutes of cleanup on fonts before polishing in Figma.

Canva still wins for final marketing asset polish, templates, and team collaboration on brand deliverables. But the Canva integration means you don’t have to choose. Use Claude Design to go from idea to first draft, export to Canva for polish and handoff to a coordinator.

Figma Make (Figma’s own AI tool) has been unreliable for most users I’ve talked to, and honestly, I’m surprised by how bad it is. Claude Design works end-to-end. For marketers without Figma muscle memory, Claude Design is the easier starting point anyway.

v0, Lovable, and Magic Patterns have built-in version history, public share links, and are often faster for pure landing page generation. But Claude Design wins on the design system extraction from a real codebase and on the tight integration with Claude Code for shipping production pages.

Nano Banana, ChatGPT Image 2.0 (the one release on April 21, 2026), Midjourney and DALL-E are a different category. Those generate images. Claude Design generates structured, editable, exportable design artifacts with real components. Use an image model to generate a hero photo and then use Claude Design to build the page around it.

Artifacts (inside Claude Chat) can build one-off HTML pages and components. Claude Design adds the persistent design system, inline comments, tweak sliders, Canva/Code handoffs, and a proper canvas. For anything beyond a single throwaway mockup, Claude Design is the right tool.

My honest takes is that Claude Design is creating a new category. For people who don’t live in design tools, they’re getting brand-consistent visual work from a conversation. If that’s you (and for most marketers, it is), it’s a huge upgrade in AI-for-design in a year.

Claude Design isn’t a magic button – treat it like a junior design collaborator with strong technical skills and little to notaste of its own. Your job is to bring the taste, the references, and the specific feedback, and Claude’s job is to make the mechanical parts of design fast.

So, in summary, start with a design system, use the tweak sliders, rewrite the copy, and ship the design.

Brandon

I’ve spent the last few months thinking about how LLMs decide which brands to mention when someone asks “best tools for X” or “[competitor] alternatives”. And I know many of you have been doing the same because SEO is different in the world of AI.

These models pull from whatever sources get cited and referenced most often across the web. If those sources don’t mention you, the model can’t talk about you. Your website could be flawless and it won’t matter if nobody else on the internet is bringing you up.

Here are some interesting stats in the evolving SEO & AEO world.

Ahrefs ran a study across 75,000 brands, 76.7 million AI Overviews, and nearly a million prompts each on ChatGPT and Perplexity. Brand web mentions correlate with AI visibility at 0.664. Backlinks were at 0.218. Brands in the top 25% for web mentions get 10x more AI visibility than those in lower quartiles. That’s not a marginal difference. That’s a different sport.

Stacker Studio confirmed the other half of it. They studied 87 earned media stories across 30 clients and found a median 239% lift in AI search citations from earned media distribution compared to brand-owned content alone. 64% of AI citations came from third-party publishers, not the brand’s own site.

One stat that stuck with me was that 80% of URLs cited by ChatGPT, Perplexity, Copilot, and Google AI Mode don’t rank in Google’s top 100 for the original query.

So the new job of SEO and demand gen teams shifted from ranking higher to getting your brand mentioned in the articles, reviews, and Reddit threads that LLMs pull from when someone asks a buying question.

Here’s the tactical playbook for doing that.

How big is this change from SEO to AEO?

I want to put some numbers on this because we all hear the it’s a change, but rarely do we see the data.

Google AI Overviews now appear in roughly 48% of all queries. Semrush tracked 155% growth from Q1 to Q4 2025. They’re expanding into commercial and transactional queries too (up from 8% to nearly 19% of triggers), which means this isn’t limited to informational searches anymore. Google AI Mode has a roughly 93% zero-click rate. Gartner predicted traditional search volume would drop 25% by 2026, and current data suggests that’s tracking close.

ChatGPT has 700 million weekly active users. Perplexity grew 370% year over year. SparkToro’s data shows AI tools have nearly tripled their usage share in the past year. Monthly AI sessions are now 56% the size of traditional search worldwide.

The volume is still relatively small (AI referrals are roughly 1% of total website traffic), but the traffic is good. HubSpot’s 2026 State of Marketing report found 58% of marketers say visitors from AI tools convert at higher rates than traditional organic. ConvertMate pegged AI referral traffic at converting 3x better, with those visitors being 4.4x more valuable. So the channel is small but the signal is strong, and it’s growing fast.

Ok, that’s enough preamble. Let’s get to the meat of it. Here are the steps that you need to take to win this new game.

Step 1: Decide which questions you want to win

It’s no longer about keywords; it’s about the questions people ask.

The average ChatGPT prompt is 23 words. The average Google search is 3.4 words. People are typing full questions with context. “Best project management tool for a remote team of 15 with Jira migration support” is a different animal than “project management software.” Your question list needs to reflect that.

Open a spreadsheet and write down the buying intent questions people ask in your category:

“Best [category] tools for [audience]” “How do I [thing your product solves]?” “[Competitor] alternatives” “Which [product type] is best for [specific use case]?”

For each question, write 3-5 variants with different phrasing and different lengths. Add columns for Intent (comparison, how-to, problem-solution), Priority (1-3, where 1 = critical for revenue), and Current LLM Presence (do you show up today or not).

You’ll keep adding to this list as you pick up new prompts from sales calls, customer tickets, and social posts. Tools like ContentSage can speed this up by generating likely prompts from your website URL, but the manual work of mapping your category’s questions is worth doing yourself at least once.

One thing most people miss is that AI queries skew long tail. Semrush found roughly 60% of keywords triggering AI Overviews have 100 or fewer monthly searches. The questions that matter most for your brand are specific, high-intent queries that traditional keyword tools barely register.

Step 2: See what LLMs already say about you

Run those questions through the actual models, and record what comes back.

For each priority question, go to ChatGPT, Perplexity, and Google (with AI Overviews or AI Mode) and paste the question. You’re looking for two things: the answer content (does your brand appear? how are you described? who else shows up?) and the citations (every URL the model references).

Copy every cited URL into your spreadsheet: Question, Model, Cited URL. Do this across at least two models because the overlap between what different AI engines cite is shockingly low. Ahrefs found less than a 1 in 100 chance that ChatGPT recommends the same thing as Google (if that doesn’t tell you how much the game is changing, I don’t know what does.). Google AI Overviews skew toward big brands. ChatGPT rewards broad distribution and consistency. Perplexity leans into niche directories.

You’ll end up with a messy list. Good.

Now, run the same prompt twice, a week apart. AI search visibility is way less stable than organic rankings. 40-60% of cited sources change month to month. Rand Fishkin ran a study with 600 volunteers running 12 prompts through ChatGPT, Claude, and Google AI Overviews a combined 2,961 times. His conclusion: rank tracking in AI search isn’t useful. There’s no stable “position.” What matters is how frequently your brand appears across many prompts, not where you land on any single one.

A few tools can help automate the monitoring, like Otterly.ai (free tier, covers ChatGPT, Perplexity, Gemini, Copilot, AI Overviews), Peec AI (includes optimization recs), Profound (enterprise, 10+ AI engines, 400M+ real prompts dataset), Semrush (AI Overviews tracking built into existing platform), HubSpot AEO ($50/month standalone), and Ahrefs Brand Radar.

Step 3: Expand to LLM-adjacent pages via Google

LLMs don’t only cite the pages they show you directly. They draw on the same ecosystem Google surfaces. For AI engines that use retrieval-augmented generation (which is most of them right now), your Google ranking still feeds into what the model sees.

Lily Ray has been hammering this point and she’s right because ChatGPT relies on Google’s search index via RAG, undermining your Google SEO to chase AI-specific shortcuts means you lose AI visibility too. AEO requires good SEO as its foundation. You can’t skip ahead.

For each priority question, Google it in incognito. Collect URLs from the top 10-20 organic results. Skip homepages, product pages, and spam. Keep guides, explainers, comparison pages, and “best X tools” listicles. Add them to your spreadsheet with Discovery Source = Google and the SERP position.

These are the pages LLMs will pull from in future training data or retrieval passes. They expand your target list beyond what the AI models directly cited.

Step 4: Clean and score your source list

You now have a big ugly spreadsheet. Time to make it useful.

De-duplicate by URL. Add columns: Already Cited by LLM? (Yes/No), Top SERP? (Yes/No or rank), Page Type (guide, how-to, listicle, comparison, docs, review), Brand Already Mentioned? (Yes/No/Weakly).

Open each URL, skim it, fill in the columns. You’re separating the LLM relevant pages from noise.

Score them:

3 = LLM-cited AND top SERP (top targets) 2 = LLM-cited OR top SERP 1 = Relevant but neither

Filter to 2 or higher. That’s your hunting ground.

Two things to keep in mind as you score. ConvertMate found 83% of AI Overview citations come from pages outside the organic top 10 (so don’t dismiss lower-ranking pages). And pages with 20,000+ characters average 4.3x more citations than thin content. Long form guides are disproportionately valuable targets because LLMs have more material to cite from.

Step 5: Build your publisher and contact list

You know which pages matter. Now you need the humans behind them.

For each high priority URL, open the article and look for the byline, author bio, or About page. Search “[Author Name] + [Site Name] + LinkedIn” to find them. If you can’t find the author, look for the Head of SEO, Content Lead, or Marketing Manager at that company.

Use Hunter, Clay, or RocketReach for email addresses. Add Contact Name, Contact Role, Contact Email, and Confidence (High/Med/Low) to your spreadsheet. This is tedious and you’ll end up with half-complete entries. That’s normal. People change jobs. Bylines disappear. A “probable content manager, not sure” entry still gives you someone to email.

Step 6: Decide what you’re asking for

Not every page needs the same ask.

For listicles (”10 best tools for X”): are you missing? Ask to be added. Are you there but poorly described or buried at the bottom? Ask for an upgrade.

For how-to or explainer content: is there a natural spot for your product as a recommended solution or example? Could they add a “tools” section where you’d fit?

For comparison pages: are competitors listed and you’re not? Is the info about you outdated?

Add a column for Ask Type: listicle_add, listicle_upgrade, correction, add_recommendation, other. For each page, you’re figuring out the least intrusive, most helpful way to get included. Think of it as editorial planning for someone else’s content.

Step 7: Write and send outreach

The unglamorous part.

Create 2-3 core templates: one for listicle add/upgrade, one for corrections/updates, one for general “your article is great, we’d add value here.” Each template should reference their specific article by name, explain why your brand belongs with one or two concrete reasons, and make it easy by suggesting 1-2 sentences they could paste in with your URL and anchor text.

Load your CSV into a cold outreach tool (Smartlead, Apollo, Mailshake) with personalization tokens for first name, site name, article title, and suggested snippet. Initial email plus 1-2 follow-ups over 7-14 days.

Reply rates are low. Maybe 1-2% if your targeting is sharp. The personalization quality is everything. Over-automate and you sound spammy. Hand-edit every email and you run out of hours. I don’t have a clean answer for this tradeoff. You’ll find your own line.

Step 8: Handle replies and close

When replies come in, classify them fast: hard no, soft no, info request, positive interest, counter (cash or reverse mention).

Know your boundaries before you start. Budget for paid placements? Willing to swap mentions (you add them to your resource page, they add you to their guide)? Any compliance constraints?

For reverse mentions, find pages on your own site where you could reasonably mention them. Partner pages, resource round-ups, tool recommendation posts. Propose a clear trade: “You add us here with this blurb, we add you here with that blurb.”

Log every outcome: Reply Type, Deal Type (earned, paid, reverse, hybrid), Next Step with a deadline. Every publisher is different. You’re juggling threads and managing two workstreams if reverse deals are in play. Keep your spreadsheet current or it spirals.

Step 9: Deliver content that’s easy to paste in

When someone says “sure, send me something,” be the easiest person they work with that day.

Open their article again. Match their tone and structure. Write 1-3 sentences that describe your product and fit their format. Include the anchor text and URL you want.

For a listicle, it looks like: “[YourBrand] — A [short description] that helps [who] do [result]. Especially useful if you [specific use case].”

Suggest exactly where it fits: “We’d slot after [Competitor X] in the list” or “This could be a bullet under ‘Tools for [use case].’” Send a clean packet: suggested snippet, URL, anchor text, an optional alt version. Don’t paste the same generic blurb across 30 sites. It won’t fit half the time and publishers notice.

Step 10: Deliver on your commitments

If you agreed to add mentions on your own site, follow through. List all obligations with the page, the mention, and the deadline. Draft each snippet, implement it, verify the live page, and send a quick confirmation. Don’t let open commitments pile up. You need these publishers to keep working with you, and they will remember if you flaked.

Step 11: Verify mentions and track LLM impact

A placement doesn’t count until it’s live and indexable. Keep a verification tab: Publisher Page URL, Expected Anchor Text, Expected Target URL, Status (pending, live, broken).

For each “live” report, open the URL and confirm the link is there with the right destination and reasonable context. Every few weeks, re-run your core questions in ChatGPT and Perplexity. You’re looking for your brand appearing more often and new citations from pages where you earned mentions. There’s a lag between a mention going live and LLMs picking it up (training schedules and retrieval refresh cycles vary), so don’t expect it to show up overnight.

ConvertMate found content updated within 30 days gets a 3.2x citation multiplier. Recently refreshed pages where your mention appears get pulled into AI responses faster.

Step 12: Build your learning loop

This can’t be a one-off project.

Monthly, review your data. Which opportunities became live mentions? Which were LLM-cited? Which were high authority SERP pages? Which outreach templates got replies? Which angles closed easiest? Use that to refine your source targeting, your pitch, and your offer structure. The brands treating this as a continuous program are the ones seeing results compound.

What else you should be doing

The outreach playbook gets your brand into the sources LLMs cite, and that’s the core play. But I’ve been going through the latest research and there are several things I’d be doing on your own properties right now that multiply the outreach work.

I’ll start with the one that surprised me. ConvertMate found 44.2% of LLM citations come from the first 30% of a page’s content. That means you need to front-load your answers. Put the direct answer in the first 40-60 words of each section, then support it. I checked a few of my own client sites after reading this and we were burying the answer under three paragraphs of setup on almost every page. Also, 68.7% of cited pages follow a clean H1, H2, H3 heading hierarchy, and 61% use structured data markup (FAQ, HowTo, Product, Article schema). If your content team hasn’t done this yet, that’s where I’d start.

Original research is the single biggest differentiator for getting cited. The Optimist, a B2B content agency, documented a 4,900% revenue increase from LLM-referred sources over 14 months for one client. Their play was first-party research and proprietary datasets. Surveys, benchmarks, data nobody else had. Makes sense when you think about it. LLMs prefer citing primary sources over summaries of primary sources. If you can produce original data in your category, you skip the line.

This next one I didn’t see coming. YouTube mentions are the strongest correlator with AI visibility in the entire Ahrefs study at 0.737. Stronger than web mentions (0.664). Way above backlinks (0.218). Video content gets extracted through transcripts, descriptions, and metadata. I’ll be honest, I’ve been underweighting video for my clients and I’m rethinking that now. If you’re not producing video about your product category, you’re missing what the data says is the strongest signal.

Reddit too… I know, I know. But Tinuiti’s AI Citations Trends Report found Reddit accounts for 5%+ of all ChatGPT citations. On Perplexity, roughly 24% of total citations. Reddit content appears in 68% of AI answers across ChatGPT, Claude, Perplexity, and Gemini. Authentic participation in relevant subreddits (not buying accounts to seed content, which gets caught) is a real channel now. A negative Reddit thread about your brand can shape a buyer’s first impression through AI before you ever interact with them directly.

Freshness matters, but don’t game it. Pages updated within 30 days get a 3.2x citation multiplier, lut Lily Ray caught multiple cases where artificial refreshing (changing timestamps without adding anything new) got detected and penalized. Update because you have something worth adding.

Kevin Indig’s data changed how I think about content architecture. Focused pages covering 26-50% of a query’s subtopics outperform exhaustive guides that try to cover 100%. I would have guessed the opposite. A Backlinko case study on Descript confirmed it: audience-specific pages get 2.3x more AI citations than generic product pages. One audience, one use case, one problem per page. That’s the right unit.

Third-party coverage compounds everything. Conductor’s 2026 report (17 million AI responses, 100+ million citations) found brands with topical authority clusters see 2.5x higher citation rates. Third-party media coverage makes brands 5x more likely to be cited. And brands are 6.5x more likely to be cited through someone else’s domain than their own. Earned media, digital PR, podcast appearances, conference write-ups. The work that’s hard to scale is the work that compounds.

What not to do

I want to flag the growing list of AEO tactics that backfire. Lily Ray has been doing the best work tracking these and I’d pay attention.

Mass-producing AI-generated content targeting AI search queries leads to what she calls the “Mount AI” pattern: rapid traffic growth followed by algorithmic collapse, and there are multiple case studies on this now. Creating excessive “alternatives” or “vs” pages backfires too. Companies with 51+ of these pages saw both blog traffic and ChatGPT citations decline. Buying Reddit accounts to seed branded content gets caught. Adding hidden LLM instructions to your pages is a bad idea (Microsoft flagged this as “AI Recommendation Poisoning” in their February 2026 security report). And llm.txt files? No major LLM has confirmed using them.

Google’s John Mueller put it well at Google Search Live: “Good AEO still relies on good SEO. Tricks will come out and they will work for a short time.”

Eli Schwartz made a point that when everyone optimizes from the same AEO checklist, you’ve created a red ocean. Hundreds of companies end up following identical steps, producing identical content. The playbook in this issue gets you the mechanics. But original thinking about your category, your positioning, and your point of view is still what separates you from everyone else running the same checklist.

Where this leaves you

Traditional SEO = make your website rank higher.

AEO = make your brand appear across the sources LLMs trust.

Your website matters (structure it for extraction, create original research, keep it current). But it’s not enough by itself. You need to show up in the places AI models pull from when someone types a buying question into ChatGPT.

The outreach playbook is the spine of this. The on-site and off-site work multiplies the impact. Avoiding shortcuts that undermine your Google rankings (which AI models still rely on via RAG) keeps the whole thing working.

It’s a lot of work. It’s slow. And the teams that start building now will have a moat the rest of the market can’t copy by reading a blog post.

If you want help with all of this and want a way to streamline the process, check out Noble (no, this is not a paid ad – they’re a former client, and I love what they’re doing).

Ok, I’m stoked about this issue, and I’ve put a LOT of work into it to make sure it’s extremely practical and useful for you.

I use Claude Code every day, and 95% of the time, I’m using the CLI, not the app. If you’ve been using the desktop app or claude.ai, that works, but you’re working with a fraction of what’s available to you. The desktop app is a chat window, but the CLI (paired with an IDE) turns Claude into something closer to a command center for all of your AI workflows.

I’ve spent the last several months building my entire marketing workflow on top of Claude Code CLI. This issue walks through the five things you need to understand to get there: your IDE, Skills, Slash Commands, MCPs, and CLIs.

IDE: A Visual Wrapper to See What Claude Is Doing and Work Faster

When you first install Claude Code, it runs in your terminal. It works, but if you’re not used to working in the terminal, it’s not ideal. If you’re not a developer, then you’re going to want an IDE.

An IDE (Integrated Development Environment) wraps everything into one window: a file explorer on the left, your code or documents in the center, and Claude’s terminal at the bottom. You can see what Claude is changing in real time, and you get visuals that show exactly what Claude is changing. You get tabs, extensions, inline editing and more in an IDE. It’s all in one place, which means you don’t have to be context-switching constantly.

I use Visual Studio Code. It has an official Claude Code extension from Anthropic (over 2 million installs), and the integration is great. You can review Claude’s edits before accepting them, @-mention specific files, and open multiple conversations in separate tabs.

But VS Code isn’t your only option. Here are five others worth checking out if you haven’t chosen your IDE yet.

Cursor is an agent-first IDE built on top of VS Code. Where VS Code treats AI as an assistant, Cursor treats it as the driver. You describe what you want, and it executes autonomously. I find it particularly good for large refactors and multi-file operations.

Google Antigravity is Google’s new agentic development platform, announced in late 2025. It’s built on a modified fork of VS Code and supports multiple AI models (including Claude Opus 4.6 and Sonnet 4.6). The standout feature is the Agent Manager, which lets you run multiple AI agents simultaneously across different workspaces. It’s free during the public preview.

JetBrains (IntelliJ, PyCharm, WebStorm) has an official Claude Code plugin on the JetBrains Marketplace. If your team is already in the JetBrains ecosystem, this keeps everything in one place. Good for enterprise environments.

Windsurf is a standalone agentic IDE that’s more budget-friendly and beginner-friendly than most options. It pioneered the “Cascade” multi-file editing feature and offers 40+ IDE plugins.

Warp is an advanced terminal environment with AI-powered capabilities. If you’re someone who lives in the terminal and wants to stay there, Warp pairs well with Claude Code’s CLI-native workflows.

Now, before trying to set up Claude Code in an IDE, first, make sure you have Claude Code CLI (not the desktop app) installed on your computer. If you don’t, run this in your terminal: npm install -g @anthropic-ai/claude-code.

Now, once you have chosen one of the above (pick whichever one fits how you work), find the download for your operating system (Mac, Windows or Linux), then go to plugins/extensions and find Claude Code.

The point is to get out of the raw terminal and into an IDE. The visibility alone is worth it.

Skills: How to Make Claude Do Specific Work in a Specific Way

Skills are all the rage right now, but they’re nothing too fancy. There are ways for you to tell Claude Code to do a specific thing in a specific way.

If you’ve used ChatGPT’s custom GPTs, skills are a similar concept but they live inside your development environment instead of a separate app. Custom GPTs are self-contained chat experiences you build in OpenAI’s interface. Skills are prompts that plug directly into Claude Code and run alongside your files, your tools, and your workflows.

Skills come in two flavors. The first improves a particular feature or function of Claude Code. For example, it could be a front-end design skill that takes your UIs from generic AI output to something that looks like a human designer built it. The second flavor is workflow. A custom YouTube pipeline skill, for example, could execute all seven production steps (and five sub-skills) with a single command.

Claude Code doesn’t load all the skills you make available to it. Instead, it lists all the skills it can use and provides a short description for each. When you say “I want to use the front-end design skill,” it checks the list, finds the match, and pulls in the full prompt. This keeps the context window from bloating.

Here are the top 27 skills I think B2B marketers should know about.

1. Skill Creator: Anthropic’s own skill for building and benchmarking new skills, install via /plugin. You can create any of the ones listed below with just this one here.

2. Nano Banana 2: AI image generation from Google that renders legible text inside images, great for presentations, ads, and infographics.

3. Cold Email: write B2B cold outreach emails and sequences

4. Email Sequence: create automated email flows and drip campaigns

5. Sales Enablement: create sales collateral and pitch materials

6. Content Strategy: plan content strategy and topic selection

7. Copywriting: write marketing copy for pages and campaigns

8. SEO Strategy: plan and execute search optimization

9. Keyword Research: identify target keywords and search opportunities

10. Analytics Tracking: set up and audit GA4 events, conversion tracking, UTM parameters, and attribution models

11. LinkedIn Posts: create LinkedIn-specific social content

12. Paid Ads: manage Google, Meta, and LinkedIn campaigns

13. Ad Creative: generate and scale ad creative for campaigns

14. Landing Page Generator: build landing pages from templates

15. Pricing Strategy: develop pricing and monetization models

16. GTM Strategy: plan go-to-market motions

17. Product Marketing Context: create foundational PMC documentation

18. Launch Strategy: plan product launches and announcements

19. Lead Magnets: create lead magnets for email capture

20. Form CRO: optimize lead capture and contact forms

21. Page CRO: improve conversion rates on marketing pages

22. A/B Test Setup: plan, design, and implement A/B tests

23. Customer Research: conduct and analyze customer research

24. RevOps: manage revenue operations and lead lifecycle

25. Marketing Psychology: apply behavioral science to marketing

26. Competitive Intelligence: build competitive analysis, battle cards, and positioning intel

27. ICP Builder: build research-backed ICP and buyer persona documents

I created a GitHub repo that has all of these skills in one library here (for free): https://github.com/guerrilla2799/marketing-skills

Just run this command in Claude Code:

/install-skill https://github.com/guerrilla2799/marketing-skills

If that doesn’t work as a one-liner, the manual approach is:

git clone https://github.com/guerrilla2799/marketing-skills.git /tmp/marketing-skills && cp -r /tmp/marketing-skills/skills/* ~/.claude/claudecodeskills/

A word of caution: There are GitHub repositories with hundreds of skills. Download the individual skills you’ve read and understand. Don’t bulk-install entire repositories. More skills mean more noise in the context window, and Claude has to sort through a longer list every time you make a request. Also, with big libraries of 100+ skills, they’re not always vetted and you’re more likely to expose yourself to malicious actors. A bad skill can read files on your machine, run shell commands, and send data to an external server without you knowing. Treat every skill you install like software you’re giving access to your system, because that’s what it is.

Slash Commands: Shortcuts That Give You More Speed and Control

Slash commands are typed shortcuts inside Claude Code that execute predefined tasks. You type a forward slash, pick a command, and it runs. Some ship with Claude Code out of the box. Others you can create yourself (custom commands live in your .claude/commands/ folder).

Here are 10 that I use regularly.

1. /insights: Surface patterns and analysis from your current project. Good for getting a quick read on what Claude sees across your files.

2. /yolo: Skip permission prompts so Claude can execute without asking you to approve every step. Use this when you trust the task and want speed. (The official flag is --dangerously-skip-permissions, but /yolo is the shorthand.)

3. /model: Switch between Claude models mid-session. If you’re doing heavy reasoning, switch to Opus. If you want speed for a simpler task, drop to Sonnet or Haiku.

4. /context: Show you how much of the context window you’re using. This matters more than you’d think, because once you hit the limit, Claude starts losing track of earlier instructions.

5. /clear: Reset the conversation. I use this constantly. Every time I shift to a new task, I clear the context so Claude starts fresh.

6. /compact: Compress the conversation to save context space while preserving key information. Good when you’re deep into a session and don’t want to start over.

7. /plan: Tell Claude to plan a complex task before executing it. Instead of charging ahead, it lays out the steps and waits for your approval. I use this for anything with more than two or three moving parts.

8. /rewind: Show a list of your previous messages and roll back to any point. Also supports “rewind code only,” which reverts file changes while keeping the conversation history.

9. /export: Save the conversation to a file or clipboard. Useful for documenting workflows, sharing processes with your team, or keeping a record of how you built something.

10. /hooks: Configure automation triggers on specific tool events. You can set actions to fire before file edits, after bash commands, or when a session starts. This is where Claude Code starts feeling like a programmable marketing platform.

There’s no need to install anything to use these. All of them are ready to go. In fact, the one I would start with is /Insights. The first time I used this, a few weeks into my Claude Code journey, it was the best thing to level up my game.

MCPs: Connect Claude to the Tools You Already Use

MCP stands for Model Context Protocol. Think of it as the standard that lets Claude Code talk to external tools and services. An MCP connects Claude to a live system (your CRM, your Google Drive, your meeting notes) so it can read data, take actions, and stay in sync without you copying and pasting between tabs.

Here are 12 MCPs that I think are worth setting up if you’re a B2B marketer.

1. Google Drive MCP: This connects Claude to your Google Workspace. You can search across Drive, edit Google Docs, manage Sheets, and schedule Calendar events without leaving Claude Code.

2. Granola MCP: This connects to Granola.ai for meeting intelligence. Search through transcripts, extract action items, and synthesize across multiple meetings. Available on all plans.

3. Notion MCP: This lets you read and write Notion pages, query databases in natural language, and manage content calendars. If your marketing ops live in Notion, this is essential.

4. HubSpot MCP: This lets you access CRM records, deals, engagement history, and lead scoring data directly from Claude Code. You can log activities and pull pipeline data without exporting CSVs.

5. Obsidian MCP: If you use Obsidian for knowledge management or competitive intelligence, this connects your vault directly to Claude Code for AI-powered analysis.

6. Slack MCP: This lets you search messages, retrieve conversations, and pull context from team discussions. This is useful when you need Claude to reference a decision or campaign debate that happened in Slack.

7. Google Analytics and Search Console MCP: This gives you real-time access to GA4 traffic data and GSC search performance. You can ask questions like “which pages rank 1 through 5 but have a bounce rate over 60%?” without touching a dashboard.

8. Clay MCP: This connects Claude to your Clay workspace for contact search, interaction history, and profile data. If you’re running outbound, this gives Claude direct access to your enriched contact database.

9. SEMRush MCP: This connects Claude to your Semrush account for live SEO and advertising data. You can pull keyword research (search volume, difficulty, broad match keywords), run backlink audits and competitor domain comparisons, and retrieve Google Ads copies for any domain.

10. LinkedIn Ads MCP: This lets you analyze LinkedIn campaign performance and get optimization recommendations. You can search by job function, seniority, company size, and industry.

11. Perplexity MCP: This gives Claude access to Perplexity’s AI-powered web search. You can conduct real-time market research, competitor monitoring, and fact verification without leaving your workflow.

12. Apify MCP: This enables web scraping and data extraction at scale. You can run pre-built automation bots (“Actors”) to scrape competitor websites, build lead lists, gather pricing intelligence, or monitor content changes automatically.

CLIs: Single-Purpose Tools That Handle the Jobs MCPs Don’t

CLI stands for Command Line Interface. These are standalone tools you run in your terminal that do a specific job and return a result. Where MCPs stay connected to a live service and let Claude interact with it continuously, CLIs are more like single-purpose power tools. You invoke them (still right in Claude Code), they do their thing, and they hand you back the output.

So, you’re probably wondering, “When would you use a CLI over an MCP?” There are a few instances: When the tool doesn’t need a persistent connection. When you want to run a one-off task (scrape a site, process a CSV, push code to GitHub). When the tool exists as a CLI but not yet as an MCP.

Here are 10 CLIs worth knowing, testing out and using.

1. OpenCLI: This turns any website, app, or local binary into a standardized CLI tool. If there’s a marketing platform you use that doesn’t have an MCP or API, OpenCLI can bridge that gap. This is similar to the skill creator skill – here you’re using a CLI to create other CLIs.

2. NotebookLM CLI: This gives you programmatic access to Google NotebookLM. You can upload sources, generate content in multiple formats, and export insights, all from the terminal. If you read one of my previous newsletters on NotebookLM, you know how I feel about this tool.

3. Claude Agent SDK: This lets you build custom agentic workflows in Python or TypeScript. This is how you go from “Claude does what I ask” to “Claude runs my entire content pipeline autonomously.”

4. Anthropic CLI: This lets you interact directly with the Anthropic API from your terminal. It’s good for testing prompts, running batch jobs, and integrating Claude into scripts.

5. Firecrawl CLI: This is for intelligent web scraping with automatic strategy selection and Playwright browser automation. I use this for competitive intelligence and pulling content from competitor sites.

6. DuckDB CLI: This lets you run SQL queries directly on CSV files with zero setup. Export your campaign data from HubSpot, GA4, or LinkedIn Ads, point DuckDB at the file, and get aggregations, filters, and cross-tabulations without spinning up a database.

7. Postiz Agent CLI: You can schedule and publish posts across 28+ social platforms (LinkedIn, X, Instagram, TikTok, YouTube, Reddit, and more) directly from the terminal. Outputs structured JSON for automation workflows.

8. SEO Bhishma CLI: This is a full SEO toolkit with backlink checking, sitemap analysis, bulk indexing verification, Google Search Console data extraction, and redirect mapping. Exports to CSV for reporting.

9. GitHub CLI: This is GitHub’s official command-line tool. Create pull requests, manage issues, and automate repository operations. If your team uses GitHub for documentation, website code, or campaign asset management, this keeps you in the flow.

10. Canva CLI: This is Canva’s official CLI tool for creating and managing Canva apps. You can build custom templates, data connectors, and content publishers directly in Claude Code.

What Comes Next

I think the gap between marketers who use AI as a chatbot and those who build workflows on top of it is going to get wider, fast. The tools covered in this issue (Skills, MCPs, CLIs, Slash Commands) are the building blocks. None of them existed in this form 18 months ago, but I think all of them will be table stakes 18 months from now.

If you’re new to Claude Code CLI, start with an IDE and one or two skills. Then add an MCP for the tool you use most (HubSpot, Notion, Google Drive). And then build from there.

Sure, there’s a learning curve, but it’s shorter than you think. And if you ever get stuck, remember you can always go back to Claude App or even ChatGPT, take a screenshot, and ask for help.

I used to spend way too much time in spreadsheets. Between pulling reports, fixing broken formulas, and cleaning up fields that should’ve been standardized months ago, it could have been a full-time job. And by the time everything finally looked right, it was already outdated.

I’m a big proponent of marketing leaders being data literate, but often that meant they spent a lot of time in spreadsheets.

But this week’s episode is about fixing that.

Because Coefficient does for data and spreadsheets what Gamma and Loveable do for design.

This week’s stack:

• 1 video: Turn messy marketing data into clean, automated dashboards

• 1 prompt: Create scroll-stopping infographics with AI

• 1 tool: Run any tool directly from your terminal with AI

• 3 resources: Personal AI systems, real-world workflows, pricing strategy

• 3 jobs: High-paying growth roles at AI-native companies

Let’s go!

Workflow Walkthrough: Turn Messy Marketing Data Into Real Insights (Fast)

If you’re still exporting CSVs, cleaning columns, and rebuilding dashboards every week… this one’s for you.

I walk through how I’m using Coefficient to turn scattered marketing data into clean, automated, insight-ready dashboards.

Instead of writing formulas and doing pivot tables, you’re asking questions and giving commands in plain English.

Here’s what you’ll learn:

• How to connect HubSpot, GA, ads, and more into one live data source

• How to clean and structure messy data using natural language

• How to generate formulas, pivots, and analysis with AI

• How to build dashboards and reports in minutes (not hours)

• How to automate snapshots, alerts, and weekly reporting

Now, your spreadsheets become a live, AI-powered system.

Try Coefficient

Prompt of the Week: Turn Raw Ideas Into Scroll-Stopping Infographics

Unless you have a full time designer, visuals can be tough. But now with AI, you don’t need a designer or hours in Figma – you just need a clear structure and the right prompt.

Lately, I’ve been using Nano Banana 2 to turn notes and my rough ideas into clean infographics.

And yes, it’s with a prompt, just tell it exactly what to say, who it’s for, and how it should look.

Here’s a prompt you can steal:

Create a clean, modern infographic designed for B2B SaaS marketers.

Topic: [INSERT TOPIC]

Goal: Teach this concept quickly and clearly in a way that is visually engaging and easy to scan.

Structure:
- A strong, benefit-driven headline at the top
- 3–5 key sections or steps, each with a short bold title and 1–2 concise supporting lines
- Use simple icons or visual cues to represent each section
- Include a clear flow (top-to-bottom or left-to-right)

Style:
- Minimalist and modern (inspired by SaaS dashboards)
- Plenty of white space
- Clean typography (no clutter)
- Subtle color palette (1–2 primary colors max)
- Designed for LinkedIn or newsletter sharing

Tone:
- Clear, practical, and slightly punchy
- No fluff, no jargon

Optional:
- Add a short “Key Takeaway” or summary at the bottom

Make sure the final output feels like something a high-growth SaaS company would publish.

Share

Tool of the Week: Run AI From Your Terminal (No UI Required)

Most AI tools live in tabs, most of which get lost in your endless sea of tabs (if you’re anything like me).

But this one lives where you actually work. OpenCLI brings AI directly into your terminal, so you can generate content, run tasks, and automate workflows without ever leaving the command line.

I think CLIs are the future, even more so than MCPs, which is why I had to share this with you.

Install this in Claude Code, and thank me later :)

Try it: https://github.com/jackwener/opencli

AI Resource Roundup

Personal AI Infrastructure Guide: This is a deep one, but worth the time. It walks through how to set up AI as your personal assistant across your entire digital life, from workflows to memory to automation. If you’re serious about building leverage with AI, this is a blueprint.

Your Team Will Only Build What They Believe AI Can Do (Liza Adams on LinkedIn): A strong mindset piece on how leadership assumptions shape what teams actually build with AI. If you want better outputs, raise the ceiling on what you think is possible.

The Usage-Based Pricing Playbook (Mostly Growth Podcast Episode): A great listen if you’re thinking about monetization and packaging in an AI-first world. Breaks down how usage-based pricing works, when to use it, and how it impacts growth, especially relevant for SaaS marketers.

Hot AI Roles

Growth Operations Lead at David AI
City / Remote: San Francisco, CA (On-site)
Pay Range: $120K–$180K

Marketing Role at OpenAI
City / Remote: San Francisco; New York City (Hybrid)
Pay Range: $239K – $265K

Growth / GTM Role at Overview AI
City / Remote: San Francisco, CO (Hybrid)
Pay Range: Not listed

That’s it for this week, appreciate you being here and reading along.

Next week, I’m going deep on Claude Code and breaking down my favorite skills, connectors, and workflows that are actually useful day-to-day. It’s a good one, so stay tuned.

Until then, keep building.

– Brandon

I got a lot of great feedback from last week’s session on Claude Cowork. Also got a lot of requests to go deeper on Claude Code, so that’s what we’re doing this week.

So this week we’re going to dive into Claude Code, the desktop app. Now, full transparency, I use Claude Code CLI much more than the desktop because the desktop app wasn’t around when I started my Claude Code journey.

Currently, I switch between the two, but I think my default will still be the CLI for the time being. I’ll write more about that next week.

This week’s Stack:

1 deep dive: The Claude Code Playbook for Marketers

1 prompt: Create your AI change management rollout plan

1 tool: Turn landing pages into conversion machines

3 resources: Clay plays, content systems, real AI agents

3 jobs: Fully remote AI marketing roles (rare find)

Let’s dive in.

What I Learned Building 47 Marketing Automations with Claude Code (Desktop)

Over the last few months, I’ve been using Claude Code… a lot.

I’ve crashed context windows, fed it the wrong CSV, accidentally had it overwrite a brand voice doc I didn’t back up (sorry, team), and completely reimagined how our MarTech stack fits together. And it was all done by typing plain English, then letting Claude do the building.

I’m going to compress all of those hard-learned lessons into a single newsletter.

Here’s where I’m assuming you are: you’ve had Claude on your desktop for maybe a month (like most marketers using Claude). You’ve drafted some copy, maybe brainstormed campaign ideas, and possibly summarized a few docs. That felt useful, but you’re pretty sure there’s another gear.

By the end of this issue, you’ll be using Claude less like a chat app and more as a full-stack marketing operator.

Let’s get into it.

Part 1: The Foundations

Before we build anything complex, we need to fix how you’re working with Claude. Because the difference between “useful AI chat” and “autonomous marketing operator” comes down to two things: how you give it context, and how you frame what you want.

Stop Asking, Start Briefing

Here’s the single biggest mistake I see marketers make: they give Claude a vague prompt and expect magic.

“Write me a competitive analysis.”

That’s a wish. And Claude will give a generic, surface-level competitive analysis that could apply to literally any company in your space. You’ll read it, feel underwhelmed, and say that AI “isn’t there yet.”

The AI is there, but you just need how to harness its power.

Here’s the same request, reframed as an actual brief:

“I’m the Head of Marketing at a Series B project management SaaS. Our top three competitors are Monday.com, Asana, and ClickUp. I need a competitive messaging analysis that compares their homepage hero copy, their pricing page positioning, and their primary CTA language against ours. Our main differentiator is that we’re the only tool with native time-tracking that doesn’t require a third-party integration. Here’s our current homepage copy for reference: [upload file]. Deliver this as a structured comparison table with a ‘recommended response’ column.”

That prompt will produce something you can actually present in a meeting. The difference is that you gave it the same brief you’d give a human strategist.

Here’s the rule I follow: if your prompt wouldn’t make sense as a Slack message to a smart contractor who’s never worked with your company before, it’s not detailed enough. Include the context, the constraints, the deliverable format, and the “why” behind the request.

Give It Your Files

Here’s one of the most important pieces, and most people still don’t use it.

Claude’s desktop app can read your local files. Upload CSVs, PDFs, Word docs, spreadsheets, presentations and Claude analyzes them, cross-references them, and builds new deliverables from them.

Upload the source file and tell Claude what to do with it.

“I’ve uploaded our Q1 campaign performance spreadsheet. Analyze the data across all channels. Identify which three campaigns had the highest cost-per-MQL and which three had the lowest. Then create a new spreadsheet with a summary tab, a channel-by-channel breakdown tab, and a chart showing MQL cost trends by month.”

Claude reads the file, runs the analysis, and delivers a finished `.xlsx` with working formulas and charts right to your machine. You didn’t write a formula. You didn’t build a pivot table. You described the outcome, and Claude built it.

The first time I saw a fully formatted Excel file appear on my desktop (I’m talking three tabs, conditional formatting, and a chart I didn’t even ask for) I genuinely laughed.

The files are the context. The more you give Claude (e.g., brand guidelines, campaign data, past reports, meeting transcript) the smarter its output becomes.

Tasks vs. Systems

Most marketers use Claude like this:

“Write me a cold email for this segment.”

That’s fine. It works. But it’s like buying a Tesla and only using it to charge your phone (granted, it’s probably the best phone charger I’ve ever had in my car).

Here’s how I use it now:

“I need to repurpose our latest blog post into a full content distribution package. Read the blog post I’ve uploaded. Then create: (1) a LinkedIn post with a hook, three key takeaways, and a CTA, (2) a Twitter/X thread of 7 tweets in a narrative arc, (3) an email newsletter intro that teases the post, (4) three variations of ad copy for paid social, and (5) a sales one-pager with pull quotes. Match our brand voice from the file I’ve uploaded. Deliver everything as a single Word document organized by channel.”

That’s a real prompt for a content system. One input, five outputs, all formatted and ready to deploy.

This is what I mean by design the system. We’re going beyond LinkedIn posts. We’re building the machine that produces them.

I’ve talked to dozens of marketers who tried Claude and said it was just like ChatGPT. Every single time, the issue was the same: they were giving it tasks when they should have been giving it systems.

When Things Go Wrong (And They Will)

The honest truth is Claude will get things wrong. It will misread a column in your CSV, hallucinate a competitor feature that doesn’t exist, and produce a chart with the axes swapped.

I spent the weeks thinking the tool wasn’t ready. It was me that wasn’t ready. But you have to think about “failures” as feedback. And Claude is remarkably good at fixing its own mistakes when you point them out. It’s the process of learning how to talk to Claude (And really every AI out there).

When Claude produces something that’s off, don’t start over. Just tell it what’s wrong:

“The pipeline chart is showing monthly values but I need weekly. Also, the Q1 MQL number should be 847, not 784. I think you transposed the digits. Fix both and regenerate the spreadsheet.”

Claude reads your correction, understands the context from the conversation, and regenerates the file. This correction loop (prompt, review, correct, iterate) is the core muscle you’re building. And it gets faster every time you do it.

Part 2: Three Marketing Workflows

Alright, foundations are set. Now let’s build.

I’m going to walk you through three workflows that I think most marketers should run regularly. For each one, I’ll give you the scenario, the prompts I use, and the thinking behind why it works.

But first, a quick setup note that applies to all three: the single most underrated asset in your Claude toolkit is a brand voice file. It’s a simple text document that describes your tone, messaging pillars, key differentiators, and terminology. Here’s roughly what mine looks like:

#brand_voice.md

# 2. Brand Voice & Tone

### Personality Traits
- [Trait 1 — e.g., Confident but not arrogant]
- [Trait 2 — e.g., Practitioner-first, not theoretical]
- [Trait 3 — e.g., Direct and opinionated]
- [Trait 4 — e.g., Occasionally irreverent, never unprofessional]

### Tone by Channel
| Channel | Tone |
|---|---|
| LinkedIn | [e.g., Authoritative, punchy, data-backed] |
| Email | [e.g., Conversational, 1:1 feel, plain text style] |
| Blog / Long-form | [e.g., Educational, narrative-driven, practitioner POV] |
| Website copy | [e.g., Clear, benefit-led, no jargon] |
| Social (X / Twitter) | [e.g., Hot takes, short, link to longer content] |

### Voice "Always / Never" Rules

**Always:**
- [e.g., Write in active voice]
- [e.g., Lead with the insight, not the setup]
- [e.g., Use short sentences. One idea per sentence.]
- [e.g., Back claims with data or examples]
- [e.g., Write at an 8th-grade reading level]

**Never:**
- [e.g., Use hollow buzzwords: "leverage," "unlock," "synergy," "game-changer"]
- [e.g., Start sentences with "I" (LinkedIn)]
- [e.g., Use em dashes as a crutch]
- [e.g., Write passive voice unless intentional]
- [e.g., Use emoji — or — only use sparingly in X posts]

Drop brand_voice.md into your project root (or wherever your CLAUDE.md lives). Then reference it in your CLAUDE.md like this:

Always read brand_voice.md before generating any content for this company.

That’s it. Create it once, update it quarterly (or anytime you feel like your brand is off). Every automation I’m about to show you references this file. This is how you make Claude copy that actually sounds like your company.

Now, the three workflows.

Workflow 1: The Competitive Intel Machine

The scenario: Your competitor just redesigned their website and updated their messaging. Your CEO Slacks you: “What are they saying now and how should we respond?” You need a competitive brief on her desk by end of day.

The old way: Ninety minutes tabbing between their site, your site, a Google Doc, and a slide template. You screenshot their pricing page, manually type out their hero copy, and try to remember what your messaging framework says. By the time you’re done, it’s 4:30 PM and the brief is more summary than strategy.

The new way:

Open Claude’s desktop app and select the folder where your brand voice and positioning files live. Then give it this:

"I need a full competitive messaging analysis for [Competitor Name]. Here's what I need you to do:

1. Search the web for [Competitor Name]'s current website. Analyze their homepage, pricing page, and primary product page. Extract their hero headline, subheadline, primary CTA, and the top three claims they make about their product.
2. Read our brand_voice.txt and positioning_doc.pdf from the folder I've shared. Compare their messaging to ours across these dimensions: positioning angle, primary audience signal, pricing framing, and trust/proof elements.

3. Deliver this as a Word document with three sections: (a) a side-by-side messaging comparison table, (b) a 'Where We Win / Where We're Vulnerable' analysis, and (c) three recommended messaging adjustments we should make in response. Keep the tone strategic and direct,  this is going to my CEO."

Claude searches the web for the competitor’s live pages, reads your internal docs, and delivers a formatted `.docx` file to your computer. About ten minutes, start to finish.

Why this works: You gave Claude three things that make the output sharp: the specific competitor to analyze, your internal positioning as a reference point, and a clear deliverable format with an audience. That last piece (”this is going to my CEO”) matters more than you think. When Claude knows who’s reading the output, it adjusts the level of detail, the tone, and the emphasis.

The first time I ran this, I expected a B-minus summary. I got something I sent to my CEO with just a few edits.

Workflow 2: The Content Repurposing Engine

This is the one that saves me the most hours per week.

The scenario: You published a 2,000-word blog post on Tuesday. Now you need to distribute it across LinkedIn, Twitter/X, email, and paid social. Each channel has its own format, tone, and CTA. And don’t forget about the one-pager for the sales team.

The old way: Reread the blog post four times. Rewrite it slightly differently for each channel. Try to remember what performed well on LinkedIn last month. Spend the rest of your afternoon context-switching between Google Docs, your email platform, and Buffer. Five mediocre variations and a burned afternoon.

The new way:

Upload the blog post and your brand voice file, then give Claude this:

"I've uploaded our latest blog post and our brand_voice.txt. Repurpose this into a full distribution package:

1. LinkedIn post: Lead with the most contrarian or surprising insight. Three key takeaway bullets. A CTA to the full article. Tone: thought-leadership, first-person, conversational. Keep it under 1,300 characters so the hook shows before the 'see more' fold.

2. Twitter/X thread: 7 tweets. Tweet 1 is a bold hook. Tweets 2-6 tell the story in a narrative arc. Tweet 7 is a CTA with the article link. Each tweet under 280 characters.

3. Email newsletter intro: A 3-sentence teaser paragraph that creates enough curiosity to click through. Include a subject line and preview text.

4. Paid social ad copy: Three versions: awareness, consideration, and direct-response. For each: a short headline, concise body text, and a CTA. I find that keeping headlines under 40 characters and body copy under 125 characters performs best on Meta - optimize for that.

5. Sales one-pager: A summary of the post's key argument for an account executive to reference in customer conversations. Include 2-3 pull quotes they can use verbatim.

Deliver everything in a single Word document, organized by channel."

One blog post goes in, 6 deliverables out. And they’re all on-brand, all properly formatted, and all in a single `.docx` file on your desktop.

Why this works: The prompt is specific about format constraints for each channel. The LinkedIn character count, the tweet limit, the ad copy lengths, etc. These constraints prevent Claude from producing outputs you’ll have to manually trim. Bake the constraints into the brief and you skip an entire editing pass.

Pro Tip: Tell Claude the audience for each piece. The LinkedIn post is for peers. The ad copy is for cold prospects. The one-pager is for salespeople. Claude adjusts vocabulary and CTA language for each. You’d do the same thing briefing five different freelancers. You’re just briefing one system now.

Run this after a blog publishes. What used to take about four hours now takes fifteen minutes, including a review pass.

I know, I’ve timed it for myself.

Workflow 3: The Friday Executive Deck

Here’s another one every marketing leader can use.

The scenario: Every Friday, someone on your team spends 90 minutes pulling traffic data from GA4, pipeline numbers from HubSpot, and campaign metrics from Meta, then manually typing everything into a PowerPoint. The numbers are stale by the time they’re presented.

The old way: Export a CSV from GA4, another from HubSpot, open them both in Excel, squint at the numbers, type them into a slide template, triple-check you didn’t transpose a digit, then send it to your VP at 4:55 PM with a quiet prayer.

The new way:

Export your data as CSVs, one for each source. A small but important detail: export two weeks of data, not one. Either include both weeks in a single CSV with a “week” column, or export separate files for this week and last week (e.g., `ga4_this_week.csv` and `ga4_last_week.csv`). Claude needs both data sets to calculate the trends. This is the one setup step that trips people up.

Drop all your CSVs into a single folder. Open Claude’s desktop app, select that folder, and give it this:

"I've shared a folder with this week's marketing data exports. The GA4 files contain website traffic (this week and last week). The HubSpot file has deal and MQL data for both weeks. The Meta file has paid campaign performance for both weeks. I need you to:

1. Analyze all the files. Pull these key metrics: total sessions, unique visitors, and top 5 landing pages from GA4. Open pipeline value and new MQL count from HubSpot. Total ad spend, cost-per-click, and ROAS from Meta.

2. Compare this week to last week for every metric. Flag anything that moved more than 15% in either direction as a callout.

3. Build a PowerPoint presentation: Slide 1 is a title slide with 'Weekly Marketing Performance' and this week's date range. Slide 2 is a traffic overview with a chart. Slide 3 is the pipeline summary. Slide 4 is paid performance. Slide 5 is 'Key Callouts and Recommendations' — 3-5 bullet points interpreting what the data means. Not just what happened, but why it matters and what we should do about it.

Keep the design clean: white background, dark gray text, blue accent charts. This deck goes to our VP of Marketing and our CFO."

Claude reads every CSV, cross-references the data, builds the charts, writes the callouts, and delivers a polished `.pptx` file to your computer. Including the interpretation slide, the “so what” that most automated reports miss entirely.

Why this works: The magic is Slide 5. Anyone can build a dashboard that shows numbers went up or down. The hard part (the part that earns you credibility at the executive table) is interpreting what the data means. Claude connects the dots because it reads all the files simultaneously and understands the relationships between traffic, pipeline, and spend.

The 15% threshold for callouts is worth noting. Without it, Claude flags every minor fluctuation and the deck becomes noise. With it, you get focused insights that actually matter. These small constraints are what makes this a reliable system.

Yes, you’re still manually exporting the CSVs. That takes about five minutes per platform. The deck used to take 90 minutes. You just automated the 90-minute part. Start there.

Your Next 24 Hours

Let’s recap what we covered.

We started with how to brief Claude like a strategist. We talked about why uploading files changes everything, and why the correction loop (prompt, review, fix, iterate) is the core muscle.

Then we built three workflows: a competitive intel machine that delivers a CEO-ready brief in ten minutes, a content repurposing engine that turns one blog post into six channel-ready assets weekly, and an executive deck that reads your CSVs and delivers a presentation with charts, trends, and strategic interpretation.

That’s a lot. Don’t try to build the board deck tomorrow.

Start small and pick one manual task you do every week. It’s ok if it just one report, one content repurposing pass, one competitive check. Choose one and hand it to Claude, follow the steps and see what comes back. Then, fix what’s wrong and iterate.

That loop is the whole game. I’m still refining skills and systems that I built weeks and months ago. But if you put in the work within a week, you’ll start seeing your work as systems you can design in Claude.

That’s when the 47 automations come fast.

Prompt of the Week: Build Your AI Change Management Plan

What I’ve learned working with over a dozen clients is the lack of AI adoption is a people problem.

Most AI initiatives fail because no one changes how they actually work. The best marketers and leaders I know are designing the rollout like a product launch.

You’ve probably heard me say that change management is the most crucial yet overlooked piece of having an AI strategy. So this prompt is here to help you with that.

You are an experienced change management and GTM strategist.

I am leading the rollout of an AI strategy within a [company type, e.g., Series B B2B SaaS company] across the [teams involved, e.g., marketing, sales, and customer success] teams.

Your job is to help me build a clear, actionable change management plan to ensure successful adoption.

Here’s the context:
- Current state: [Describe current workflows, tools, and AI usage]
- Desired future state: [What success looks like with AI fully adopted]
- Key tools being introduced: [e.g., Claude, ChatGPT, n8n, etc.]
- Team size and structure: [# of people, roles]
- Leadership expectations: [e.g., increase output, reduce costs, improve speed]

Create a comprehensive plan with the following sections:

1. Executive Narrative  
2. Adoption Strategy  
3. Enablement Plan  
4. Communication Plan  
5. Resistance & Risk Mitigation  
6. Success Metrics  
7. Quick Wins  

Keep the tone practical and operator-focused. Avoid generic change management theory.

Deliver this as a structured plan I can present to leadership and execute with my team.

Know someone who’d find this prompt useful? Share it with them!

Share

Tool of the Week: Turn Your Landing Pages into Conversion Machines

CAC down 28%. Ad spend down 23%. And no, it wasn’t from Better Creative. The fix was not having a lag.

For years, I treated weekly reporting like part of the job—pull data, spot what broke, adjust, repeat. But by the time you’re looking at the numbers, the damage is already done: budget burned, CAC inflated and creative fatigued.

You’re always one week behind.

What changed was collapsing the gap between signal and decision.

That’s where Prism by Pixis comes in. It pulls real-time signals across your channels, diagnoses what’s happening, and actually recommends what to do next, so you can act before performance drops, not after.

Try it at: https://tinyurl.com/bredlinger

AI Resource Roundup

8 Clay Plays Every Marketer Should Run (GTM Strategist by Maja Voje): A tactical breakdown of high-impact Clay workflows you can swipe today. From lead enrichment to signal-based outbound, this is a playbook for turning data into pipeline without adding headcount.

AI Content System Walkthrough (Adam Stewart on TikTok): A quick, practical demo of how one marketer is structuring their AI content engine end-to-end. Worth watching just to see how simple the setup actually is, and how fast you can go from idea to distribution.

The Coolest Agents I’ve Built So Far (AI Daily Brief Podcast): A behind-the-scenes look at real AI agents in action. I wanted to include this because I thought it was just a great way to get ideas for agents to build. Also, I love his Agent Madness bracket idea.

Hot AI Jobs 🔥

This week’s lineup is all remote roles—a rare (and very welcome) find in today’s market.

Product Marketing Manager, AdTech at Hightouch
City / Remote: Remote
Pay Range: $145K–$170K

Marketing Manager at Empromptu.ai
City / Remote: Remote
Pay Range: $150K–$200K

Product Marketing Manager at GC AI
City / Remote: Remote
Pay Range: $140K–$165K

Even if you’re not on the hunt, these job specs are a great pulse check on where AI marketing is headed.

Before You Bounce

The weather here is starting to catch up with the AI market… in other words, it’s getting hot. It’s been in the 80s this week here in Colorado. Spring is here and I can definitely tell you my allergies have alerted me that flowers are in bloom.

That’s another week in AI. See you next week where we’re going to dive more into Claude Code CLI

I spend about 90% of my AI time in Claude Code. But recently, I was taking Pavilion’s AI in GTM course, and the instructor (Jon Russo) opened my eyes up to some awesome use cases for Claude Cowork. Instead of just prompting the model, he turned it into a workspace for messaging, personas, and sales intelligence. He even built some with it.

It immediately clicked, and I decided I need to explore Cowork more.

So I asked him to come on the show and break it down. Luckily, Jon Russo and I go back (a whole decade at this point), and he was more than happy to share the workflow.

This week’s Stack.

• 1 workflow: Build hyper-personalized outbound cadences using Clay + Claude Cowork

• 1 prompt: Generate a deep audience intelligence profile for any niche

• 1 tool: Build marketing automations faster with Zapier Copilot

• 3 resources: AI trends, Anthropic training, and Clay playbooks

• 3 jobs: AI growth, ABM, and portfolio marketing roles

Let’s go!

Workflow Walkthrough: Turn Prospect Data Into Hyper-Personalized Cadences

Most outbound still looks the same: generic sequences, shallow personalization, and a lot of ignored emails. In this episode, Jon Russo shows a smarter way.

Instead of writing outreach from scratch, he pipes real GTM data (we’re talking personas, call transcripts, objections, and competitive intel) into Claude Cowork. Then, Cowork gives us human‑sounding cadences built from the same insights your sales team already has.

Here’s what you’ll learn:

• How Clay + Claude Cowork becomes a new kind of ABM engine

• The folder structure Jon uses to turn messy GTM knowledge into usable AI context

• How to generate sub‑100‑word cadences that still feel personal and relevant

• Why Jon keeps a human‑in‑the‑loop before sending anything

If you’ve ever wondered how to turn AI from “content generator” into a real sales workflow, this episode is a great place to start.

Watch the full walkthrough here.

Watch on YouTube

Prompt of the Week: Audience Intelligence Profile Generator

First off, this is not a ChatGPT prompt. This is a prompt for Claude Cowork.

If you deeply understand what your audience fears, wants, and believes, content gets dramatically easier. This prompt turns any niche into a structured audience intelligence report, revealing the segments, emotions, and triggers that drive follows, engagement, and revenue.

You are a senior social media strategist with 10+ years of experience building and scaling brands across B2B and B2C industries. Your specialty is translating audience psychology into content that drives measurable business outcomes.

Analyze the niche: [INSERT NICHE]

Deliver a structured Audience Intelligence Profile covering the following:

1. Audience Segmentation
Identify 2–3 distinct, high-value audience segments within this niche. For each, define:
- Demographics and psychographics
- Income/buying power level
- Where they spend time online (platforms, communities, content types)

2. Pain Points & Frustrations
For each segment, surface:
- Their top 3 functional frustrations (practical problems they can't solve)
- Their top 2 emotional frustrations (what keeps them up at night)
- The language and exact phrases they use to describe these problems

3. Emotional Triggers & Desires
- What do they secretly want to be, have, or feel?
- What do they fear most (losing status, wasting time, being judged, etc.)?
- What aspirational identity are they moving toward?

4. Content Consumption Habits
- Preferred formats (short video, carousels, long-form, etc.)
- Posting cadence and peak engagement windows
- The accounts and content styles they trust and follow

5. The Follow / Engage / Buy Triggers
- What makes them hit follow? (the promise your profile must make)
- What makes them comment or share? (the emotional or social currency)
- What makes them buy? (the specific belief that must be true before they convert)

6. Content Strategy Recommendations
Provide 5 high-performing content angles tailored to this audience, with a one-line rationale for why each will work.

I’ve added this to my to my Library. Check it out for 65+ more prompts.

The Ultimate ChatGPT Prompt Library for Marketing Leaders

Share

Tool of the Week: Build Automations Faster with Zapier Copilot

Automation is powerful… until building the workflow becomes the bottleneck.

Zapier Copilot fixes that.

Instead of manually wiring together triggers, actions, and apps, you simply describe the workflow you want to build. Copilot then generates the automation for you, recommends steps, and helps troubleshoot issues along the way.

For marketers, this means faster workflows (from routing leads to enriching data to triggering campaigns) without needing to think like an engineer.

Read more about it at https://zapier.com/blog/zapier-copilot-guide/#what

If you’re like me and you’re a builder, start building with Zapier agents using Copilot here: https://bit.ly/4rbN3mu

AI Resource Roundup

5 Key AI Events From the First 75 Days of 2026: A fast, well-curated breakdown of the biggest AI developments shaping the market right now. If you want to stay sharp on the macro trends influencing tools, funding, and adoption, this is a great pulse check.

Anthropic AI Learning Hub: Anthropic’s official training platform packed with tutorials, courses, and best practices for getting more out of Claude and AI workflows. This is a helpful starting point if you want to understand how to actually apply AI in day-to-day work.

8 Clay Plays Every Marketer Should Run (GTM Strategist by Maja Voje): A tactical playbook of high-impact Clay workflows for marketing teams. From enrichment to outbound personalization, it’s a great snapshot of how modern GTM teams are using Clay to drive smarter campaigns.

Hot AI Jobs 🔥

This week’s lineup spans growth marketing, ABM leadership, and AI product marketing at some of the most influential companies building the future of AI.

Growth, Digital Marketing at OpenAI
City / Remote: San Francisco, CA
Pay Range: $239K – $265K

AI ABM Manager at Scribe
City / Remote: San Francisco, CA (On-site)
Pay Range: $105K – $150K

AI Portfolio Marketing Manager at AMD
City / Remote: Santa Clara, CA (Hybrid)
Pay Range: $145.1K – $217.7K

Even if you’re not on the hunt, these job specs are a great pulse check on where AI marketing is headed.

That’s it for this week.

Time is absolutely flying. I guess that’s what happens when you’re having fun experimenting with AI and modern marketing workflows. Every week it feels like there’s a new tool, tactic, or playbook to test.

Here’s to stacking smarter workflows and scaling your pipeline.

Until next week,
Brandon

If you’ve been reading this newsletter (or following on LinkedIn) for a while, you know I’m pretty bullish on two AI companies right now: Anthropic and Google.

Claude is getting a TON of the love in AI circles right now, and for good reason. But Google is still releasing a lot of great AI tools. I’ve highlighted many on newsletters in the past, but one that we’re going to revisit today is NotebookLM. Because over the past year it’s quietly evolved into one of the most practical AI products for marketers.

So this week, instead of a video walkthrough, I’m doing another long-form deep dive on how B2B marketing teams can turn NotebookLM into a real GTM workflow tool.

This week’s Stack

1 deep dive: How to turn NotebookLM into your marketing team’s research brain

1 prompt: Move your AI memory from ChatGPT to Claude

1 tool: A new AI interface experimenting with persistent memory

3 resources: Prompting tactics, AI workflows, and AI-native thinking

3 jobs: AI growth, enablement, and operations roles

Let’s dive in.

The Marketing Brain You Never Built:

Why NotebookLM Should Be Your Next GTM Workflow Tool

If you’ve been following me, you know I am a fan of NotebookLM. I wrote about it in an earlier newsletter, but they’ve made some major upgrades in the last six months. I thought it was time to do a deeper dive on how marketers can use it.

If you’re not familiar with it, NotebookLM is an AI-powered research and note-taking tool from Google. They position it a “thinking partner,” allowing you to ground AI responses in your own uploaded documents. It analyzes PDFs, Google Docs, YouTube videos, websites, and text to generate summaries, study guides, outlines, and podcast-style audio discussions

Most AI tools give you a smart assistant with a large general knowledge base. But NotebookLM works differently. Instead of relying on general training data, it grounds every response in the sources you upload. PDFs, Google Docs, URLs, YouTube transcripts, audio files, Microsoft Word documents. You feed it your documents, and it becomes an expert in exactly that corpus., nothing else.When you ask NotebookLM a question, it reasons across your uploaded materials and cites specific passages to back every claim.

What’s Changed: NotebookLM’s Most Important Recent Updates

If your mental model of NotebookLM is the early version from 2023 (upload a PDF, ask questions, get answers) you’re working with an outdated map. The product has evolved significantly. Here are the updates that matter most for marketing professionals.

1. Audio Overviews (and Interactive Audio)

The feature that went viral. Upload a collection of documents and NotebookLM generates a podcast-style conversation between two AI hosts who discuss, debate, and synthesize your content. For long research documents or dense competitor analyses, this turns reading work into commute-time listening.

More recently, the team added the ability to join the conversation, interrupting the hosts, asking follow-up questions, and steering the discussion. New Audio Overview formats now include a brief summary, a critique, and a debate format, giving marketers multiple angles on the same material.

I still use this feature ALL the time.

2. Video Overviews and Slide Deck Generation

NotebookLM can now convert your uploaded sources into slide decks and video overviews, where the AI hosts discuss your material while presenting relevant visual content. If you need to communicate research to executives or internal teams, this reduces hours of deck-building into minutes.

It’s still not perfect, but this has improved now that it uses Nano Banana Pro to create presentation, and I can only image that it will continue to improve.

3. The Studio Panel

A redesigned three-panel interface separates your workflow into Sources (manage uploads), Chat (I think of it as interrogating your knowledge base), and Studio (generate outputs). From Studio, you can click one button to produce a briefing document, study guide, FAQ, timeline, or audio overview from your uploaded sources. I think this is where NotebookLM becomes a real tool to production tool.

4. Data Tables

One of the most practically useful recent additions is the ability to ask NotebookLM to synthesize information from your sources into structured tables, exportable to Google Sheets. Ask it to extract every competitor’s pricing from a collection of competitive research documents. Ask it to pull all customer objections from a set of sales call transcripts and categorize them by theme. The output drops directly into a spreadsheet, ready for analysis. From here, I’ll often use Gemini to help me figure out the best way to do the analysis. I’ll sometimes use the Claude Chrome Plugin to do some of the data cleaning, formatting, etc.

5. Expanded Context Window and Conversation Memory

NotebookLM now runs on Gemini’s full 1 million token context window, which is amazon! This is significant jump that allows it to reason coherently across very large document collections. Conversation memory has also increased more than sixfold, meaning you can have extended, multi-turn research sessions without the model losing track of earlier context.

6. Custom Personas and Goal-Setting

You can now configure NotebookLM with detailed custom personas (up to 10,000 characters) defining tone, role, output format, constraints, and reasoning style. So, depending on the task you can set it up as a B2B positioning analyst, a competitive intelligence researcher, a demand generation strategist, etc. The model then applies that lens to every response across your notebook.

7. Collaboration and Workspace Integration

NotebookLM is now a core Google Workspace service available to business customers, including enterprise-grade data protection where your uploads and queries are never used to train models. The source types now include Microsoft Word files, Google Sheets, YouTube URLs, and images, which really broadening the range of B2B marketing materials you can load in.

Ok, now for the fun stuff!

Setting Up Your Marketing Brain

The biggest mistake marketers make with NotebookLM is treating each notebook as a one-off research project. The better approach is building structured, persistent notebooks that accumulate context over time, effectively creating a knowledge base that reasons.

Here are three core notebook architectures worth building from day one.

ICP & Audience Insights

• What to Upload: Win/loss interviews, sales call transcripts, G2 reviews, survey results, support tickets

• Primary Use: Messaging development, positioning, persona refinement

Sales Enablement

• What to Upload: Product docs, pitch decks, battlecards, competitor websites (as PDFs), case studies

• Primary Use: Objection handling, rep training, deal support, competitive responses

Market Intelligence

• What to Upload: Analyst reports, competitor announcements, earnings transcripts, industry news summaries

• Primary Use: Trend spotting, demand signals, narrative shifts, board-level insights

Use Case 1: Audience Insight Mining

Turn raw customer conversations into positioning fuel in a single afternoon

Most marketing teams collect far more customer intelligence than they actually use. Sales calls sit in Gong. Win/loss interviews live in a shared folder (and let’s be honest, nobody opens it). G2 reviews get glanced at once per quarter. Support tickets get routed to the product team and disappear, neven to be seen by marketing.

The information is there, no one ties it together and turns it into useful, actionable insights.

But NotebookLM can that scattered intelligence into a working research layer you can examine in real time.

Sources to Upload

• Gong/Fathom/Granola call transcripts (exported as PDFs or text)

• Win/loss interview recordings or written summaries

• NPS or CSAT survey open-ended responses (compile into a single document)

• G2, Capterra, or Trustpilot reviews (export or copy into a text file)

• Customer support ticket summaries by theme

• Exit survey responses

• Sales team objection logs

Workflow Steps

1. Create a dedicated notebook: ‘ICP Intelligence: [Quarter]’

2. Upload 10–20 sales call transcripts from deals you’ve won, deals you’ve lost, and churned accounts

3. Add 3–5 win/loss interview summaries

4. Paste compiled G2 and Capterra review excerpts (100–200 reviews minimum for useful patterns)

5. Set a custom persona: ‘You are a B2B positioning strategist. Analyze these sources to identify buying triggers, objection patterns, language customers use to describe their pain, and competitive comparison themes. Always cite your sources.’

6. Run the queries below. Use the Data Tables feature to export structured findings.

Analysis Queries to Run

• ‘What are the five most common reasons customers chose us over competitors? Quote specific language from the transcripts.’

• ‘What objections appear most frequently across lost deals? Group them by theme and show frequency.’

• ‘What words and phrases do customers use to describe the problem we solve before they knew our product existed?’

• ‘Where do customers express the most frustration in the pre-purchase experience? What are they most skeptical about?’

• ‘Compare the language used by ICP-fit customers versus out-of-profile customers. What distinguishes them?’

Outputs and What to Do With Them

A single afternoon in this notebook can surface:

• The exact trigger phrases that precede a purchase decision (gold for paid and email copy)

• The top three objections your messaging needs to proactively address

• Customer-native language to replace the jargon in your homepage and sales decks

• Patterns in which personas convert fastest and which require the most education

Use the Audio Overview feature to generate a podcast-style debrief for your sales team or leadership. It’s a surprisingly effective way to distribute customer intelligence to people who won’t read a 20-page research summary.

Sample Output: Customer Voice Analysis

• Most frequent pre-purchase trigger: ‘We just lost a deal we should have won’ (appears in 11 of 18 win transcripts)

• Top objection in lost deals: ‘We’re not sure this integrates with [existing stack]’ — cited in 14 of 22 loss transcripts

• Customer-native language for pain: ‘Visibility gap’ / ‘blind spot’ / ‘flying without data’ — avoid ‘unified analytics platform’

• ICP signal: Buyers who reference ‘board visibility’ in first call convert at 3x the rate of buyers who lead with ‘reporting’

Use Case 2: Sales Enablement Support

Build the knowledge base your sales team actually uses

Sales enablement is one of the highest-leverage activities a Marketing leader can own (and one of the most underdelivered).

Tell me if you can relate: Battlecards get written once and go stale. Objection libraries live in Notion that nobody updates. Competitive positioning docs are accurate for about six weeks after launch and increasingly fictional thereafter.

The core problem is the maintenance cost. PMM are busy, and keeping these resources current requires them to be constantly ingesting competitor updates, customer feedback, and product changes… then translating all of it into sales-usable language.

And yes, you guessed it. That’s the job NotebookLM can take over.

Sources to Upload

• Your current product documentation and feature sheets

• Existing battlecards and competitive positioning docs

• Competitor websites exported as PDFs (or pasted as text)

• Competitor G2 profiles and review themes

• Recent analyst comparisons or category reports

• Your case studies and customer proof points

• Sales pitch decks and demo scripts

• Recent objection logs from your sales team

Workflow Steps

1. Build a ‘Sales Intelligence’ notebook and load the sources above

2. Set a persona: ‘You are a competitive intelligence analyst for a B2B SaaS company. Analyze these sources to help our sales team understand our differentiation, handle objections with specificity, and position against named competitors. Always ground responses in the uploaded documents.’

3. Use the Data Tables feature to generate a side-by-side competitor comparison (features, pricing, positioning, known weaknesses)

4. Run the objection analysis queries below

5. Generate a briefing document from Studio (this becomes the living sales enablement doc)

6. Update sources monthly as competitor positioning shifts; the notebook updates automatically

Analysis Queries to Run

• ‘What are the three most significant gaps between our product and [Competitor X] based on these documents? For each gap, suggest how our reps should address it in a sales conversation.’

• ‘Create a concise response for each of the following objections: [list 5 objections]. Ground each response in specific product capabilities or customer proof points from the uploaded sources.’

• ‘What customer outcomes and ROI metrics appear most frequently across our case studies? Organize them by persona or industry.’

• ‘If a prospect says they’re evaluating us versus [Competitor Y], what are the three strongest points of differentiation our rep should lead with? Cite the source for each claim.’

Deal Acceleration: A Specific Workflow

I think that one of the highest-value applications is in-deal support. When a rep reports that a deal is stalling because the prospect is ‘evaluating [Competitor X]’, your marketing ops or PMM team can run this workflow in under 20 minutes:

1. Add any new competitor intel specific to this deal (news, recent G2 reviews, pricing intel)

2. Run: ‘A prospect is actively comparing us to [Competitor X]. They are in [industry] with [size] employees. Based on our sources, what are the strongest differentiated arguments we should make, and what risks should we acknowledge and neutralize?’

3. Send the output as a deal-specific competitive brief to the rep within the hour

That kind of responsiveness used to require a dedicated PMM and hours of work. And now NotebookLM can compress it into minutes.

Sample Output: Competitive Brief Excerpt

• [You] vs. Competitor X: Our integration depth with Salesforce (native, bidirectional) vs. their CSV export workflow is a documented advantage in 6 of our case studies — lead with this.

• Known concern: Their G2 reviews (47 in last 6 months) cite ‘faster onboarding’ as a strength. Counter with our 14-day implementation guarantee (page 4, implementation guide).

• Proof point for this deal’s industry: [Customer] (same segment) achieved 34% pipeline velocity improvement — transcript excerpt available.

Use Case 3: Forecasting and Trend Analysis

Read the market before the market tells you what it’s reading

Pipeline forecasting and market trend analysis are traditionally the domain of spreadsheets, analyst subscriptions, and gut instinct. NotebookLM adds a third option that acts as a reasoning layer and can synthesize patterns across disparate signal sources and surface implications you’d otherwise miss.

Sources to Upload

• Quarterly pipeline exports summarized by stage, segment, and source (PDF or text summary)

• Campaign performance reports from the last two to four quarters

• Analyst reports in your category (Gartner, Forrester, IDC)

• Competitor press releases and funding announcements

• Industry news summaries (compile into a rolling document)

• Sales team win/loss summaries by quarter

• Category-specific LinkedIn or newsletter content you track for signals

Workflow Steps

1. Create a ‘Market Intelligence’ notebook and load the sources above

2. Set a persona: ‘You are a B2B market strategist. Analyze these sources to identify emerging demand signals, messaging shifts in the market, pipeline risk indicators, and trends that should affect how marketing invests in the next quarter. Be specific, quantitative where possible, and always cite your source.’

3. Run the analysis queries below

4. Use the Audio Overview to generate a 10-minute market briefing you can listen to before your next leadership meeting

5. Generate a briefing document from Studio. Format it as an executive summary for your CEO or board

Analysis Queries to Run

• ‘Based on the pipeline data and campaign reports, which acquisition channels are showing improving efficiency and which are declining? What does this suggest about budget reallocation?’

• ‘What shifts in competitor messaging and positioning are visible across these documents from the past two quarters? What might they be responding to in the market?’

• ‘What themes appear in the analyst reports that are not yet reflected in our own messaging? Where is the market moving that we haven’t followed?’

• ‘Based on pipeline trends and win/loss patterns, what does next quarter’s pipeline risk look like? Where are the most likely shortfalls?’

• ‘What demand signals (increasing search interest, competitor price changes, market announcements) suggest that buyer urgency is increasing or decreasing in our category?’

The Board-Ready Briefing

One practical output worth building into your quarterly rhythm: a market trend briefing for leadership, generated from this notebook.

Load in the quarter’s analyst content, competitive announcements, campaign results, and pipeline summary. Run a prompt like: ‘Write a 500-word executive summary of the most important market developments this quarter and their implications for our GTM strategy. Format as three sections: What We’re Seeing, What It Means, and What We’re Doing About It.’

The output becomes your first draft. In a format grounded in the actual documents, not your recollection of them. Revise and refine, but the synthesis scaffolding is done.

Sample Output: Market Signal Detection

• Competitor A raised $50M Series C and immediately changed pricing language from ‘per seat’ to ‘per outcome’. Three separate press mentions in the last 60 days. Monitor for category narrative shift.

• Analyst report (Gartner, Q3 2025): Category purchase decision now involves an average of 7.2 stakeholders, up from 5.4 in 2023. Marketing content needs to address security, finance, and IT personas, not just the buyer.

• Pipeline data: Enterprise deals sourced from webinars have a 42% shorter sales cycle than content-sourced deals, but represent only 12% of volume. Demand gen budget reallocation warranted.

Where NotebookLM Will Let You Down

NotebookLM is a powerful tool that still has meaningful constraints. Going in with accurate expectations prevents frustration and misuse.

Limitations to Understand Before You Build

Source Quality = Output Quality. NotebookLM can only synthesize what you give it. Weak, incomplete, or outdated sources produce weak outputs. Garbage in, garbage out still applies.

Source Limits. Standard accounts have limits on the number of sources per notebook. NotebookLM Plus and enterprise accounts extend these significantly. Budget accordingly if you’re building large knowledge bases.

Hallucination Risk at the Margins. When source material is thin or ambiguous, the model will sometimes fill gaps with plausible but wrong inferences. Always verify critical claims against the underlying source.

Not a Real-Time Tool. NotebookLM doesn’t update automatically when markets shift. Your notebooks are only as current as your most recent source upload. Build a rhythm of regular source refreshes into your workflow.

Context Drift in Long Sessions. Despite the improved context window, very long multi-turn sessions can lose fidelity on early sources. Start a new session for substantively different research questions.

No CRM Integration (Yet). You can’t connect it directly to Salesforce or HubSpot. Pipeline data needs to be exported and uploaded as a document, an extra step worth accounting for in your workflow design.

Your 7-Day NotebookLM Adoption Plan

The barrier to building this into your workflow is lower than it looks. I know, it might seem like a lot, but here’s a practical way for getting your team from zero to operational in one week.

Day 1

Setup & Orientation

Create your account at notebooklm.google.com. Build three empty notebooks: ICP Insights, Sales Enablement, Market Intelligence. Spend 30 minutes exploring the interface: Sources, Chat, and Studio panels.

Day 2

Populate Notebook 1

Upload your 10 most recent sales call transcripts and 3 win/loss interview summaries to the ICP Insights notebook. Set a custom persona. Run these prompts:

PRMOPT 1:

Analyze all uploaded transcripts and reviews. What are the five most common trigger events or situations that caused customers to start looking for a solution like ours? Quote specific customer language for each trigger.

PROMPT 2:

What words and phrases do our customers use to describe the problem we solve in their own language, not ours? Compile a vocabulary list I can use to rewrite our homepage and ad copy.

Day 3

First Output

Run this Prompt (alternative value propositions). Generate an Audio Overview of your ICP notebook and listen on your commute. Share one insight from the session with your team.

PROMPT:

Based on the customer interviews and sales transcripts, write three alternative value propositions for our product: one that leads with time savings, one that leads with revenue impact, and one that leads with risk reduction. Ground each in specific customer evidence from the sources.

Day 4

Sales Enablement Build

Upload your product docs, battlecards, and competitor PDFs to the Sales Enablement notebook. Run this Prompt for your primary competitor. Share the output with your top rep and get their feedback.

PROMPT:

Create a battlecard for [Competitor X] using only the information in these uploaded sources. Format it as: Who They Target, How They Position, Their Three Strongest Claims, Their Three Most Cited Weaknesses, Our Best Response.

Day 5

Market Intelligence

Upload your most recent analyst report and the last two quarters of campaign performance to the Market Intelligence notebook. Run these prompts:

PROMPT 1:

What are the three most significant trends visible across the analyst reports and industry sources uploaded? For each trend, describe what it means for our GTM strategy and what action you’d recommend.

PRMOPT 2:

Based on the pipeline data and campaign reports, write a forecast narrative for next quarter. Identify the channels that are gaining efficiency, the segments showing the most momentum, and the areas of highest risk. Flag any patterns that should change our planning assumptions.

Day 6

Team Workflow

Identify one person on your team who should own each notebook as a living resource. Define a refresh cadence: which sources get updated weekly, monthly, quarterly.

Day 7

First Briefing

Generate a Studio briefing document from the Market Intelligence notebook. Format it as your next leadership update on market trends. This is your proof of concept. It should take 20 minutes instead of two hours.

Conclusion

Can you see now why I love Notebook LM?

I think it’s one of the most practical context engines available today to build systematic context into your workflows. And the best part is ot’s free to start, works with materials you already have, and rewards you if you feed it the most rigorous inputs.

Prompt of the Week: Move Your AI Brain from ChatGPT to Claude 🪄

I made the switch to Claude as my main AI tool of choice, but I still talk to a lot of people who are on the fence.

And it’s mainly because they’ve spent years training ChatGPT on tone, workflows, prompts, projects, etc., and they do not want to rebuild that from scratch.

Claude just made that easier to switch.

First, you export what matters from ChatGPT. Then you paste it into Claude’s memory settings.

And that’s it. Now Claude picks up like it already knows how you work.

And this is just another example that AI literacy is now table stakes.

It’s also why I think AI Ops is becoming the real edge — and memory/context portability is a huge part of that.

Know someone who’d find this prompt useful? Share it with them!

Share Stack & Scale

Tool of the Week: An AI Chat Interface That Learns How You Work

There are a growing number of tools trying to “own the interface” for AI, essentially becoming the place where you run all your models and workflows. One that recently caught my eye is Chatly AI.

A friend of mine called it a “total game changer” (which, let’s be honest, people say about everything in AI right now). But after playing around with it, I do think there’s something interesting here.

I’m not sure yet whether it’ll make its way into my daily workflow, but it’s definitely a category worth watching.

Try it: https://chatlyai.app/

AI Resource Roundup

AI Prompting Trick (TikTok): A quick but clever walkthrough showing how small prompt framing changes can dramatically improve AI outputs. Worth the 60-second watch if you want a simple reminder that better prompting often beats switching tools. Great for marketers trying to get more consistent results from AI workflows.

AI Workflow Breakdown (YouTube): This video walks through a practical AI workflow you can apply to marketing and research tasks. The interesting takeaway isn’t just the tools — it’s how they’re combined into a repeatable system. If you’re trying to move from “random AI usage” to structured workflows, this is a good watch.

Becoming an AI-Native Operator (Growth Unhinged): A thoughtful piece on what it actually means to operate as an AI-native professional. The big idea: the real shift isn’t using AI tools, it’s designing your work around them. A useful perspective for marketing leaders trying to rethink team workflows in an AI-first world.

https://www.growthunhinged.com/p/becoming-an-ai-native-operator

Hot AI Jobs 🔥

This week’s lineup spans AI growth, enablement, and operations — a good snapshot of how companies are building AI capability directly into GTM teams.

AI Growth Marketing Lead at BRM
City / Remote: San Francisco, CA (On-site)
Pay Range: Not Listed

AI Enablement and Innovation at Gong
City / Remote: San Francisco, CA
Pay Range: $143K – $210K

Applied AI Operations Lead at Canvas Medical
City / Remote: Remote
Pay Range: $175K – $250K

That’s it for this week. It’s been one of those quiet but busy, heads-down, get-sh*t-done kinds of weeks over here. It’s the kind where a lot gets built, but not much makes it onto social media.

Time to get back to it.

See you next week.

Brandon

Let’s just jump right into it. Clay changed the game again.

I know, I’ve said that before, but it’s true again. This week, Clay announced their new Ads feature, and I got early access to start playing around with it. If you’re running paid, this removes a ton of manual work, and when ad rates are sitting in the high 80s and low 90s, your ROI is also high.

If it’s not obvious by now, yes, I’m a huge Clay fan. And this week’s main walkthrough breaks down exactly why.

This week’s stack:

• 1 deep dive: Automate your paid engine with Clay’s new Ads feature
• 1 prompt: Build a 90-day editorial calendar that closes real SEO gaps
• 1 tool: An AI-powered resume builder that boosts interview rates
• 3 resources: Agent skill hacks, AI marketing strategy, and the future of AI teams
• 3 jobs: AI-first product marketing and content leadership roles

Let’s go!

Workflow Walkthrough: Easily Launch and Manage Ad Campaigns with Clay

In this week’s deep dive, I didn’t want to just show you how to connect Clay to LinkedIn and Meta ads – that would be a little boring. I thought to myself, what’s a useful, impactful, and unique Clayt able I can build while showing off the new ads feature?

So, I landed on using TrustRaduis buyer intent signals to build the account list, then enrich contacts in Clay, then send that audience to LinkedIn.

In other words, I'm showing you how to quickly and easily launch LinkedIn ads to high-intent buyers.

Here’s what you’ll learn:

• How to use the TrustRaduis + Clay integration to find buyers in market right now

• The single best tip for ensuring the highest possible match rates for your ad audiences

• A step-by-step guide to send your Clay table to LinkedIn as an audience for your ad campaigns

Use my referral link to get additional credits: https://clay.com?via=2bb18e

Here’s the Clay Workbook Template you can copy into your own Clay instance.

Get The Clay Template

Prompt of the Week: Your 90-Day Content Game Plan, Built in Minutes

Most content plans are based on the whims of what your content team feels like writing about in the moment. But the best plans are based on strategy and research.

This prompt forces your SEO data, ICP insights, and funnel gaps into a single 90-day execution plan, with KPIs tied to pipeline, not pageviews.

If you’re serious about closing gaps instead of publishing noise, this is the one to steal.

SYSTEM / ROLE
Timeframe: 90 days (3 months)

Publish cadence: Realistic (2–3 per week max)

Ensure persona balance (no single ICP > 50% of assets)

4. Asset Fields (REQUIRED)

For each asset include:

Title (SEO-optimized and intent-aligned)

Primary Persona

Format (Blog, Landing Page, Comparison Page, Case Study, Webinar, Guide, Checklist, Video, etc.)

Funnel Stage

Publish Date (spread logically across 90 days)

Primary KPI (Traffic, MQLs, SQLs, Assisted Conversions, CTR lift, etc.)

Primary Promotion Channel (Organic Search, Email, LinkedIn, Paid Search, Retargeting, Partner, etc.)

5. Strategic Quality Rules

Titles must reflect search intent (informational, comparison, commercial, etc.)

BOFU pieces must include commercial intent modifiers (e.g., “best,” “vs,” “pricing,” “alternatives”)

Each month must support at least 2 funnel stages

Each persona must appear in every month at least once

KPI must match funnel stage:

TOFU → Traffic / Rankings / CTR

MOFU → MQLs / Engagement

BOFU → SQLs / Conversions

Retention → Expansion / Upsell / Retention rate

OUTPUT FORMAT (STRICT)

Return:

Markdown table

Grouped by:

Month 1

Month 2

Month 3

No explanation

No commentary

No reasoning

Table only

OUTPUT STRUCTURE

| Title | Persona | Format | Funnel Stage | Publish Date | KPI | Promotion Channel |

I’ve added this to my to my Library. Check it out for 65+ more prompts.

The Ultimate ChatGPT Prompt Library for Marketing Leaders

Share

Tool of the Week: Build a Resume That Actually Gets Interviews

Most resumes are generic. I know it’s best practice to cater your resume to the specific job description you’re applying for, but who has time to do that when you’re applying for dozens or hundreds of jobs?

Well, there’s an AI for that. Teal’s Resume Builder helps you tailor your resume to specific roles using AI, job matching, and built-in keyword optimization. You can track applications, compare job descriptions, and adjust your positioning in minutes instead of rewriting from scratch every time.

Try it: https://www.tealhq.com/tools/resume-builder#free

AI Resource Roundup

Claude Skill Shopping Hack (Andy Jolls on LinkedIn): Instead of building a Claude skill from scratch, Andy had an agent scan Smithery, SkillHub, and SkillsMP and surface the best fact-checker.

AI Agents for Marketers (9X on YouTube): A breakdown of how AI agents are moving from “cool demo” to real GTM operators. This video from the 9x YouTube channel gives practical examples especially helpful if you’re figuring out where agents fit inside your current workflow.

OpenClaw: AI Agents & the Future of Marketing (AI-Driven Marketer on YouTube): This is a forward-looking conversation between Dan Sanchez and his bother Travis on how agents are reshaping marketing roles, execution speed, and org design.

Hot AI Jobs

Product Marketing Manager (AI-First) at Klarity

• Location: San Francisco, CA (On-site)

• Pay Range: $170K–$200K

Product Marketing Manager – Agentic AI for HR at Eightfold AI

• Location: Santa Clara, CA (Hybrid)
  Pay Range: $112.5K–$150K

Head of Content at Lightfield

• Location: San Francisco Bay Area (On-site)

• Pay Range: $190K–$240K

Even if you’re not on the hunt, these job specs are a great pulse check on where AI marketing is headed.

Before You Bounce

Now that the Olympics are over (and Team USA hockey brought home gold) I can focus more on tinkering.

This is such a wild, creative time in B2B marketing. The tools are getting better by the week and the people leaning in are pulling ahead just as fast. I hope you’re having as much fun with this shift as I am.

Brandon

This week is a two-for-one episode, which also makes it our longest one yet. Koka Sexton, an OG in B2B who helped pioneer the social selling movement, joins me for a great conversation. He’s leading the charge on agentic marketing and AI agents and walks through a few Make.com workflows in detail.

But what really grabbed me is what he’s building with OpenClaw. Full transparency: I haven’t installed it yet. I briefly considered it… then chickened out over the security questions. But honestly, it feels like it’s only a matter of time.

Anyway, here’s this week’s Stack:

• 1 deep dive: Build an AI agent to run your GTM with Make + OpenClaw
• 1 prompt: Turn AI into a structured “coworker” with guardrails and approval gates
• 1 tool: Detect whether content was written by AI
• 3 resources: AI ROI reality checks, strategic AI frameworks, and the future of agents
• 3 jobs: Product marketing, AI data, and AI-native GTM leadership roles

Let’s go.

Workflow Walkthrough: Build an AI Agent That Runs Your GTM

Initially, when I asked Koka to come on the show, he was going to show off some of his awesome Make workflows. But when it came time to scheduling, he said, “I have something even better to show you.” I said, “show me both!”

So in this episode, we dive deep, walking through his Make workflows and how he’s using OpenClaw.

Here’s what you’ll discover in this episode:

• How to export LinkedIn likes, comments, and shares and turn them into high-intent pipeline and drop them into an automated Make workflow

• When Zapier breaks and how 40-step Make.com automations take over

• How to collapse 40K LinkedIn connections into a segmented, searchable CRM with OpenClaw

• How OpenClaw routes tasks to the best model automatically

• Why “bots talking to bots” is the next GTM frontier

Watch on YouTube

Prompt of The Week: AI Coworker with Safety Rules (for Claude CoWork)

Most people (still) use AI like a chatbot. But I like to think of them as teammates. This is the kind of structured “coworker prompt” that the top markeres use. It has a clear task, clear inputs, strict safety rules, and explicit approval gates. You can use this same structure for outbound, content distribution, CRM updates, or internal reporting.

You are helping me [clear outcome].

In this folder you will find [describe structured inputs clearly].

Your job:

[Task 1]

[Task 2]

[Task 3]

For each item, you must:

[Personalization requirement]

[Formatting requirement]

[Output requirement]

After drafting everything, [execution instructions].

Important safety rules:

Only use [approved data source].

Do not [prohibited action].

Never execute without my explicit approval.

If anything is unclear, ask a clarification question instead of guessing.

I’ve added this to my The Ultimate ChatGPT Prompt Library for B2B Marketing Leaders notion doc. Check it out for 60+ more prompts.

Share Stack & Scale

Tool of the Week: Is This Actually Written by a Human?

Almost everything I read now feels like it was written by AI. I’m talking about blog posts, LinkedIn posts, cold emails, website… you name it.

If you’re like me, you’re thinking to yourself, “AI wrote this, didn’t it?” Well, this tool gives you a quick answer. GPTZero analyzes text and estimates the likelihood it was generated by AI. Of course, it’s not perfect, and there are other tools that will be better, but this is a great free tool.

Try it: https://gptzero.me/

AI Resource Roundup

The Week in AI: ROI When Reality Bites Back (Ray Rike): A sharp breakdown of where AI hype meets financial reality. Critical if you’re defending AI spend to a CFO right now.

Almost Timely News: How I Think About AI (Christopher Penn): A thoughtful framework for evaluating AI tools strategically instead of chasing every shiny object.

The Next Wave of AI Agents (YouTube): A forward-looking discussion on agentic systems and where they’re headed. Helpful context if bots-talking-to-bots got your wheels turning.

Hot AI Jobs

Marketing Leader at DatologyAI

• Location: Redwood City, CA (On-site)

• Pay Range: $200K–$275K
  

Director, Product Marketing at Descript

• Location: San Francisco, CA (Remote)

• Pay Range: $220K–$250K

Marketing Director at Predactiv

• City / Remote: Palo Alto, CA (Remote)

• Pay Range: $125K–$170K

By now, I’m sure you’ve seen the posts about something big happening in AI. It definitely feels that way. If you’re like me, you probably oscillate between being really excited and overwhelmed.

I’m here to encourage you to go ahead. You don’t need to master everything. Just build one thing at a time.

Now, please excuse me while I go play around with Claude Code.

Brandon

Over the holiday break, my feed was flooded with builders shipping projects using Claude Code. And if I’m being honest? I felt behind.

While everyone else was using the downtime to experiment, I was watching it unfold thinking, “Alright… I need to catch up.” So over the last few weeks, I did exactly that and went deep and stress-tested it through a marketer’s lens.

Because the thing is right now, most of the tutorials and videos coming out about Claude Code are built for developers, but most of you aren’t trying to become developers.

And since the long-form written guide I did on using Lovable to vibe code an ROI calculator was such a hit, I’m going back to the well with another long-form written guide.

So, without further ado, let’s dive in!

This week’s Stack:

• 1 deep dive: A marketer-friendly guide to getting started with Claude Code

• 1 prompt: Turn AI drafts into authentic, human-feeling content

• 1 tool: Transform yourself into any character with Krea

• 3 resources: AI model routing, scrappy UGC systems, the future of AI-enabled sales

• 3 jobs: Growth, marketing automation, and AI-native GTM roles

Let’s go!

Deep Dive: The Marketer’s Blueprint for Claude Code

First, you might be wondering what is cloud code and why should marketers care? Unlike the Claude chatbot you may have used at Claude.ai (the app), Claude Code can read files on your computer, create new ones, browse the web, and execute complex multi-step workflows without you copying and pasting anything. You describe a task in natural language, and it does the work. And it runs on macOS, Windows, and Linux.

For example:

• You can drop a CSV of campaign performance data into a folder, type “analyze my Q4 campaign results by channel and create a report with recommendations,” and Claude Code will read the file, crunch the numbers, and output a polished markdown report.

• You can point it at a folder of customer interview transcripts and ask it to extract the top objections your sales team faces, then generate a content calendar addressing each one.

• You can ask it to build an entire landing page from scratch, complete with copy, HTML, and CSS, then iterate on it conversationally.

The core advantage over regular Claude chat is file system access. Claude Code operates directly in your project folders. It can process massive datasets that would exceed chat upload limits, maintain context across dozens of files simultaneously, and output organized deliverables.

Claude Code requires a Pro subscription ($20/month) at minimum. And Anthropic says the Max plan ($100–$200/month) is recommended for heavy daily use.

Installing Claude Code in three steps (no coding required)

The installation process takes about five minutes. Here’s exactly what to do, step-by-step:

Step 1: Open your terminal.

On a Mac, press Cmd + Space to open Spotlight, type “Terminal,” and hit Enter. On Windows, press Win + X and select “Windows PowerShell.”

You’ll see a text-based window with a blinking cursor. This is where you’ll type commands. Don’t worry about understanding everything you see; you only need to type exactly what’s shown below.

Step 2: Run the installer.

The Good news is that Claude Code’s native installer handles everything automatically. Node.js is no longer required (it was needed for the older npm installation method, which is now deprecated and was needed for a lot of the early MCP installs… I spent so many hours trying to debug my Node.js LOL). Simply paste the appropriate command for your operating system:

For Mac or Linux, paste this single line and press Enter:

curl -fsSL https://claude.ai/install.sh | bash

For Windows PowerShell, paste this instead:

irm https://claude.ai/install.ps1 | iex

Mac users who prefer Homebrew can alternatively run brew install --cask claude-code. Windows users can use winget install Anthropic.ClaudeCode. Both methods work identically. Use whichever feels more familiar.

The installer will download Claude Code and place it in the right location on your system. You’ll see progress text scroll by. Wait until it finishes and you see your cursor again.

Step 3: Verify it worked.

Type the following and press Enter:

claude --version

You should see a version number like “2.1.37” displayed (this is what mine shows if I type that in now, but yours might be 2.1.38 or higher depending on when you install it). If you see an error like “command not found,” close your terminal, reopen it, and try again. The installer needs a fresh terminal session to update your system path. You can also run the command “claude doctor” for a full diagnostic check that identifies and suggests fixes for any configuration issues.

Step 4: Authenticate for the first time

Now, in your terminal, type:

claude

On first launch, Claude Code will display a welcome screen with authentication options. Here’s exactly what happens:

1. Choose your authentication method. You’ll see options including “Claude App (with Pro or Max plan)” for subscription billing, or “Claude Console” for pay-per-token API billing. For most B2B marketers, choose the Claude.ai subscription option.

2. Browser opens automatically. Claude Code launches your default browser and redirects you to claude.ai (or platform.claude.com for Console API users) to complete a secure OAuth 2.0 login flow. You’ll sign in with your existing Claude account credentials.

3. Authorize Claude Code. After signing in, you’ll be asked to authorize Claude Code to access your account. Click to approve.

4. Return to the terminal. The browser will confirm successful authentication, and you can return to your terminal window.

5. Accept terms and start working. Back in the terminal, accept Anthropic’s terms of service. Claude Code will index your current directory and present an interactive prompt where you can start typing requests.

Your credentials are stored securely: in macOS Keychain on Mac, or in ~/.claude/.credentials.json on Linux/Windows. You won’t need to log in again unless you explicitly log out (type /logout) or your token expires.

Important note: If you have an ANTHROPIC_API_KEY environment variable set on your system, Claude Code will use that instead of the OAuth flow, which may result in unexpected API charges. You can check your current authentication status at any time by typing /status in a Claude Code session.

Your first conversation with Claude Code

Once Claude Code is running, you interact with it by typing natural language requests at the prompt. Here are the essential commands every marketer should know:

A few beginner tips to keep in mind. First, always navigate to the folder containing your project files before launching Claude Code: type cd followed by the folder path, then claude. This gives Claude Code access to your files.

Second, be specific in your requests. “Create a 1,200-word LinkedIn article about AI in supply chain management, targeting VP-level logistics executives, with a contrarian hook and three data points” will dramatically outperform “write me a LinkedIn post about AI.”

Third, break complex projects into phases rather than asking for everything at once. Research first, then outline, then draft, then refine.

SET-UP INSTRUCTIONS CAVEAT

This could (and let’s be honest, it will likely) change in the next few weeks by the time you’re reading it. I say that because it changed between when I set it up a few weeks ago and now (Feb. 10, 2026). If that’s the case I will try to update this, but if you’re reading this and I have not updated it, the best/easiest thing you can do is take a screenshot of where you’re getting stuck in the terminal, go to Claude, drop your screenshot in and ask Claude what you should do.

Now, for the fun stuff!

Building a LinkedIn thought leadership post generator

This walkthrough creates a reusable system that transforms raw ideas, notes, or transcripts into polished LinkedIn posts matching your brand voice. By the end, you’ll have a working content machine in a folder on your computer.

Set up your project workspace

Open Terminal and create a dedicated folder:

mkdir ~/Documents/linkedin-generator
cd ~/Documents/linkedin-generator
claude

Now tell Claude Code to set up the project structure. Type this at the Claude Code prompt:

Create the following folder structure for a LinkedIn content generation workspace: a /examples folder for sample posts, a /raw-input folder for notes and transcripts, a /output folder for finished posts, and a CLAUDE.md file at the root. In the CLAUDE.md file, include these instructions for yourself: "This workspace generates LinkedIn thought leadership posts for a B2B marketer. Always write in a conversational, authoritative tone. Posts should be 800-1500 words for articles or 150-300 words for feed posts. Use short paragraphs (1-3 sentences max). Start with a hook that challenges conventional wisdom. Include specific data points or examples. End with a question to drive engagement. Never use hashtags in the body text. Add 3-5 relevant hashtags only at the very end."

Claude Code will create all the folders and the CLAUDE.md file. This file is special. Claude Code reads it automatically at the start of every session, so your brand voice and formatting rules persist across conversations without re-explaining them.

Train it on your voice

Drop 3–5 of your best-performing LinkedIn posts into the /examples folder as text files. You can do this manually in Finder/Explorer, or ask Claude Code. The first time, I did this, I just asked Claude to do this for me:

I'm going to paste three of my best LinkedIn posts. Save each one as a separate markdown file in the /examples folder, named example-1.md through example-3.md.

Then paste each post when prompted. Once saved, tell Claude Code:

Analyze all the posts in /examples. Identify patterns in my writing style: sentence length, paragraph structure, hook types, use of data vs. stories, tone shifts, and recurring phrases. Save your analysis as style-guide.md in the root folder.

This creates a machine-readable style guide derived from your actual writing. Claude Code will reference this in future sessions.

Generate posts from raw input

Now the system is ready. Drop a raw input into /raw-input. This could be meeting notes, a podcast transcript, bullet points from a conference talk, or even a rough idea scribbled in a text file. Then ask:

Read the file in /raw-input called conference-notes.md. Using my style guide in style-guide.md and the examples in /examples, generate 3 different LinkedIn posts from this material. Each post should take a different angle: (1) a contrarian take that challenges industry assumptions, (2) a "lessons learned" narrative with a personal story hook, and (3) a data-driven insight post. Save all three to /output with descriptive filenames.

Claude Code will read your notes, cross-reference your style guide, and produce three distinct posts saved as files you can review and edit. The quality improves dramatically when you provide real examples rather than generic instructions: one good example of your writing beats twenty adjectives describing your desired tone.

Create a reusable slash command

To make this repeatable without re-typing long prompts, create a custom slash command:

Create a folder called .claude/commands in the project root. Inside it, create a file called linkedin-post.md with this content: "Read the most recent file in /raw-input. Using the style guide in style-guide.md and examples in /examples, generate 3 LinkedIn posts taking different angles (contrarian, narrative, data-driven). Save to /output with today's date in the filename. After generating, list the 3 posts with their hooks so I can quickly choose which to refine."

Now, in any future session, you simply type /linkedin-post and the entire workflow runs automatically. This is the real power of Claude Code for marketers: building reusable systems!

Three more marketing use cases worth building

Campaign performance analyzer

Create a folder, drop in your campaign CSVs from Google Ads, LinkedIn Ads, or HubSpot, and ask Claude Code to cross-reference performance across channels. Anthropic’s own growth marketing team uses this exact approach. They feed Claude Code hundreds of existing ad variations with performance metrics, and it identifies underperforming ads, generates new headline and description variations meeting strict character limits (30 characters for headlines, 90 for descriptions), and outputs them in a ready-to-upload format.

The key prompt pattern is this: provide the data file path, specify which metrics matter most (CTR, CPA, ROAS), define your target audience, and ask for specific, actionable recommendations rather than general observations. For large datasets, Claude Code can spin up multiple sub-agents working in parallel, one analyzing headlines, another analyzing descriptions, producing hundreds of variations in minutes rather than hours.

Content audit and gap analysis engine

Export your blog content from your CMS as a CSV or collection of markdown files, along with Google Analytics data and Search Console exports. Place everything in organized subfolders. Then ask Claude Code to analyze publishing cadence over the past six months, map content against customer journey stages, identify topics your competitors cover that you don’t, and cross-reference against objections mentioned in sales call transcripts.

The output can be an interactive HTML presentation with clickable tabs, expandable sections, and data visualizations: a single file you open in any browser and share with stakeholders. Content marketing agency Animalz built exactly this workflow and published installable plugins for it. What previously required a consultant and two weeks of spreadsheet work happens in an afternoon.

Competitive intelligence tracker

Competitive intelligence tracker

Point Claude Code at competitor websites and ask it to extract pricing tiers, feature lists, positioning language, and messaging frameworks. It browses each site, saves structured notes as markdown files, and generates comparison tables. The real power comes from creating a custom slash command like /update-competitors that re-runs this analysis monthly, comparing what changed. Combine this with an MCP server connection to Google Sheets, and Claude Code can automatically update a shared competitive tracking spreadsheet your entire team references.

Supercharging Claude Code with MCP integrations

MCP (Model Context Protocol) can turn Claude Code into a connected marketing command center. You can think of MCP as a universal adapter that lets Claude Code talk directly to your marketing tools using natural language. Instead of exporting a CSV from HubSpot, downloading it, and dropping it in a folder, you connect the HubSpot MCP server and simply ask Claude Code to “pull this quarter’s lead data from HubSpot and analyze conversion rates by source.”

Setting up an MCP server takes one terminal command. For example, to connect Notion:

claude mcp add --transport http notion https://mcp.notion.com/mcp

To connect a tool that requires an API key, like Airtable:

claude mcp add --transport stdio airtable --env AIRTABLE_API_KEY=your_key_here -- npx -y airtable-mcp-server

The MCP servers most relevant for B2B marketers include Google Sheets (46 tools for reading and writing spreadsheet data), Slack (search messages, post updates), Google Drive (access and process documents), HubSpot (CRM data and pipeline analysis), Google Analytics (200+ dimensions and metrics via natural language), Google Search Console (SEO performance data), and Notion (workspace and database management).

The Zapier MCP server deserves special mention because it connects to over 8,000 apps and offers 300 free tool calls per month, making it the fastest way for non-technical teams to experience MCP’s power.

You can browse available servers at directories like mcpservers.org, mcp.so, and mcpmarket.com.

To see which servers you’ve connected, type claude mcp list in your terminal or /mcp during a Claude Code session.

Essential tips for marketers new to Claude Code

Start with a CLAUDE.md file in every project. This is your persistent brand brief. Include your company’s tone of voice, target audience descriptions, content guidelines, formatting preferences, and any terminology to use or avoid. Claude Code reads this file at the start of every session, so you never have to re-explain your brand.

Break large analyses into focused requests. The larger the volume and number of bundled requests, the higher the chance of hallucination. Instead of “analyze everything in this folder and give me all insights,” try “analyze the publishing cadence trends in blog-export.csv” followed by “now analyze author contributions in the same file.” Directed questions produce dramatically more reliable results.

Use the “feel free to use multiple agents” instruction. Adding this phrase to complex prompts tells Claude Code it can spawn parallel sub-agents, one scanning transcripts for quotes on topic A while another handles topic B, with a third synthesizing everything into a draft. This is particularly powerful for content repurposing workflows where you need a blog post, five LinkedIn posts, an email, and a carousel from the same source material.

Remember that Claude Code auto-updates. If you used the native installer, updates happen automatically in the background. You can also manually update by typing claude update. Run claude doctor periodically to check your setup health, verify your installation type and version, and get suggestions for any configuration issues.

Conclusion

Claude Code represents a fundamental shift in what marketers can accomplish without developer support. The combination of direct file system access, natural language interaction, reusable slash commands, and MCP integrations creates something genuinely new. It’s like an autonomous marketing operations layer that lives on your computer.

The LinkedIn generator walkthrough is a good example of the core pattern: set up a project folder, teach Claude Code your voice through examples and a CLAUDE.md file, build reusable workflows with slash commands, and iterate. That same pattern scales to campaign analysis, competitive intelligence, content audits, and dozens of other workflows.

Start with one workflow, get comfortable with the terminal, and expand from there. The five-minute installation is the only technical barrier; everything after that is just describing what you want in plain English.

Prompt of the Week: AI-to-Human Content Regeneration

We’re all using AI to draft content, but the real skill now is making sure it doesn’t sound like AI. If your content reads like it was written by a ChatGPT, it won’t convert and you’ll lose the audience quickly.

This prompt systematically removes AI patterns and injects human tone and feel without losing the substance.

You are a content authenticity specialist who transforms AI-generated drafts into genuinely human-feeling content. Your goal is to identify and eliminate AI patterns while preserving valuable ideas.
## Input Content
**Original Draft:** [PASTE_AI_GENERATED_CONTENT]
**Content Type:** [SOCIAL_POST/BLOG/AD_COPY/EMAIL]
**Target Audience:** [SPECIFIC_DEMOGRAPHIC]
**Brand Voice:** [CASUAL/FORMAL/PLAYFUL/EXPERT]
**Platform:** [WHERE_IT_WILL_BE_PUBLISHED]
## AI Pattern Detection & Removal
### Phase 1: Identify AI Markers
Scan for and flag:
- Em dashes (—) usage
- Overused transitions ("Moreover," "Furthermore," "In today's world")
- Perfect grammar and punctuation
- Formulaic structures (intro-3 points-conclusion)
- Generic phrases ("dive into," "unlock," "elevate," "transform")
- Excessive use of colons and semicolons
- Too-perfect parallel structure
- Hedge words ("arguably," "potentially," "seemingly")
### Phase 2: Human Imperfection Injection
Add strategic imperfections:
- **Intentional casualness**: Contract words, start sentences with "And" or "But"
- **Rhythm variation**: Mix short punchy sentences. With longer, more meandering thoughts that feel like natural speech patterns
- **Voice inconsistencies**: Occasionally break your own rules
- **Specific details**: Replace generic examples with ultra-specific ones
- **Time markers**: "Last Tuesday" instead of "recently"
- **Personal asides**: Brief tangential thoughts in parentheses
- **Emotional tells**: "Honestly," "Look," "Here's the thing"
### Phase 3: Structural Humanization
- Break conventional formatting
- Use unexpected paragraph breaks
- Include sentence fragments. On purpose.
- Add conversational callbacks ("Remember that thing I mentioned?")
- Include real-world specifics (actual brands, places, cultural references)
## Output Format
### Version A: Light Touch
[Minimal changes - just remove obvious AI patterns]
- Key changes made: [LIST]
- AI markers removed: [COUNT]
- Readability impact: [ASSESSMENT]
### Version B: Medium Humanization
[Balance of AI improvement and human authenticity]
- Strategic imperfections added: [LIST]
- Voice adjustments: [WHAT_CHANGED]
- Anticipated performance: [PREDICTION]
### Version C: Full Human Mode
[Maximum authenticity, might sacrifice some clarity]
- Radical departures: [WHAT_YOU_DID]
- Risk assessment: [WHAT_MIGHT_NOT_WORK]
- Best use case: [WHEN_TO_USE_THIS]
## Humanization Techniques Applied
### Style Adjustments:
- Removed: [X] em dashes → replaced with: [ALTERNATIVES]
- Removed: [X] perfect transitions → replaced with: [NATURAL_FLOW]
- Added: [X] intentional typos/casual language
- Added: [X] culture-specific references
### Content Enrichment:
- Generic example: [ORIGINAL] → Specific example: [REPLACEMENT]
- Vague claim: [ORIGINAL] → Concrete detail: [REPLACEMENT]
- AI phrase: [ORIGINAL] → Human alternative: [REPLACEMENT]
### Final Authenticity Check:
□ Would a human actually say this?
□ Does it sound like someone typing, not generating?
□ Are there enough imperfections to feel real?
□ Does it match how the brand actually talks?
Provide all three versions with clear documentation of changes, allowing for selection based on platform requirements and risk tolerance.

I’ve added this to my The Ultimate ChatGPT Prompt Library for B2B Marketing Leaders notion doc. Check it out for 65+ more prompts.

Know someone who might find this prompt useful? Share it with them!

Share Stack & Scale

Tool of the Week: Krea – Turn Yourself Into Shrek (Or Anything Else)

I’m sure you’ve seen those side-by-side videos on social where one is the person talking, the other mirrors their movement perfectly… but as Shrek.

That’s tools like Krea. Krea lets you transform live or recorded video into stylized, AI-generated visuals that track your exact expressions and movement. production team. For marketers, this means faster creative testing and engaging content without a big budget or production team.

Try it: https://www.krea.ai

AI Resource Roundup

A GTM Guide to AI Models (RevEngine): I thought this was a great practical breakdown of which AI models to use for which GTM tasks. With more of us using more than one LLM, it’s going to be key to learn how and when to route work to the right model.

$20 in AI Credits → Repeatable UGC System (LinkedIn post from Dan Ahmadi): A tactical walkthrough of building a repeatable AI-driven UGC engine on a shoestring budget.

The AI-Enabled Future of Sales (The Transaction Podcast): A grounded conversation on how AI is reshaping modern sales teams and GTM motions.

Hot AI Jobs

Growth Marketer at Krea (yea, the same Krea from the tool of the week)

• Location: San Francisco, CA (On-site)

• Pay Range: Not Listed

Growth Marketing Automation Manager at LeanData

• Location: Santa Clara, CA

• Pay Range: $90,000 - $120,000 + a 10% variable

AI ABM Manager at Scribe

• Location: San Francisco, CA (Remote-friendly)

• Pay Range: $105K–$150K + equity

Even if you’re not on the hunt, these job specs are a great pulse check on where AI marketing is headed.

If you haven’t dug into Claude Code yet, don’t be afraid. Honestly, that was me a few months ago. Now, I’m loving it. Once you get past the intimidation factor, it’s less “terminal hacker energy” and more superpower for builders.

Next week, we’re diving into the other thing that’s taken over our feeds: OpenClaw. I’m bringing on one of my old buddies and an OGs, and Koka Sexton breaks down exactly how he’s using it and how to set it up the right way.

This week was one of those weeks. Deep in client work, head down, and juggling just enough things that the long-form guide I wanted to finish never quite crossed the line. I’ve been spending a lot of time lately inside Claude Code, and I had plans to break down how marketers can actually use it in practice.

I didn’t get that guide done, but I still wanted to ship something genuinely useful. Instead, I’m sharing a podcast conversation I recently participated in on the Agents of Insight podcast with . We got into AI hype, GTM engineering, and how the marketer’s role is changing in real time.

This week’s Stack is a little different but worth your time.

Watch on YouTube

Here’s what we cover in this conversation:

• How to build “always-on” GTM systems that quietly fill the funnel without relying on spammy outbound or rented lists

• The look-alike playbook: how to start from one closed-won deal and systematically find hundreds of high-propensity prospects

• How to turn competitor weaknesses into pipeline using G2 reviews, sentiment analysis, and smart positioning

• Why de-anonymizing website visitors is becoming table stakes — and how to separate real buying intent from casual research

• The real framework behind GTM Engineering (and why most teams get it wrong by focusing on tools instead of systems)

• How to navigate AI hype vs. real leverage — what’s noise, what’s useful, and where marketers should actually invest time

• The “speed of learning” advantage that separates teams that win from teams that stall

• A clear north star for CMOs: how to think about markets, messaging, and execution as AI reshapes the role

  Share Stack & Scale

Hope you enjoyed that conversation. It was a fun one, and honestly a good reminder that beneath all the AI hype, the real leverage still comes from systems, clarity, and speed of learning.

I’m heads-down this week finishing up a Claude Code guide for marketers, so stay tuned for that next week.

– Brandon

Hope everyone’s staying warm, and if you’re in the U.S., that you made it through the winter storm without getting stranded. I avoided the travel chaos (prayers to those who didn’t), and we only lost one school day with the little kiddos, which feels like a small miracle.

For week’s build, we’re going back to the Clay well, and it’s a bit calmer but just as powerful. Yes, it’s simpler than some of the recent ones, but incredibly useful. In fact, this is another one that I think every marketing team can and should use. It shows how to listen for brand mentions on social, analyze sentiment, and send clean summaries straight to Slack for your team.

This week’s Stack:

• 1 video: Build a Clay-powered social listening system that sends clean brand signals to Slack

• 1 prompt: Write hyper-personalized snippets

• 1 tool: Find and reverse-engineer high-performing content

• 3 resources: AI orchestration, prospecting prompts, and attribution shifts

• 3 jobs: Senior GTM and marketing leadership roles at some cool AI companies

Let’s go!

Workflow Walkthrough: Turn Social Noise Into GTM Signal Without Expensive Tools

This episode is a full walkthrough of how I use Clay to run social listening that delivers clean signals straight to Slack, ready for your team to take action on.

You’ll see how I monitor brand mentions across Reddit, LinkedIn, and Twitter/X, score sentiment, summarize long posts, and alert the right people in real time. I also include a few other potential builds you can layer on top of these foundational social listening tables. Again, this is on the simpler side of theClay workbooks I’ve built, but one of the more impactful ones that every team can and should use.

What you’ll see:

• How to source brand mentions across Reddit, LinkedIn, and Twitter/X inside Clay

• My sentiment scoring setup (negative, neutral, or positive) for instant gut checks

• How I summarize messy posts into clean, actionable Slack alerts

• When to use Claygent vs lightweight models to control credit usage

• How to extend listening into competitor monitoring, prospecting, and creator discovery

If you’re just getting started with clay and need some extra credits to play around with, sign up with my referral code and you’ll get 3,000 free credits.

Here’s the Clay Workbook Template you can copy into your own Clay instance.

Get the Clay Template

Prompt of the Week: Write hyper-personalized snippets

Most “personalized” outreach isn’t actually personal. You can spot it from a mile away. It’s just a first-name merge with a company mention sprinkled in. But this prompt (hopefully) fixes that by forcing the model to anchor every opener in a real, verifiable prospect detail and explicitly tie it to your value prop, so your message feels earned, not automated.

It’s especially useful when you’re doing LinkedIn outreach at scale but still want each opener to sound like you actually did the homework.

System / Role

You are an expert B2B sales copywriter specializing in hyper-personalized LinkedIn outreach.
Context
• Prospect LinkedIn profile: <PROFILE_LINK>
• Our value proposition document: <VP_DOC>
Task

Generate 3 distinct LinkedIn message openers (≈40 words each) for cold outreach.
Requirements

Each opener must:
1. Reference one specific, verifiable prospect detail from LinkedIn (e.g., recent role change, post they shared, award, company initiative).
2. Explicitly connect that detail to a relevant benefit from our value proposition.
3. End with a curiosity-driven question that invites a reply (not a yes/no question).
4. Sound natural, confident, and human—avoid generic praise or sales clichés.
5. Be materially different from the others (no reused angles or phrasing).
Output Format

Return only:
1. Opener 1
2. Opener 2
3. Opener 3
Do not include explanations, headers, or commentary.

I’ve added this to my The Ultimate ChatGPT Prompt Library for B2B Marketing Leaders notion doc. Check it out for 60+ more prompts.

Know someone who might find this prompt useful? Share it with them!

Share

Tool of the Week: Snowball, Borrow What’s Working

I’ve got a friend who borrows (read: steals) ideas from high-performing content in his niche, recreates them, and keeps shipping. He’s been blowing up on YouTube, Shorts, and TikTok by doing one thing consistently: not guessing.

This is the tool he uses. Snowball breaks down what’s actually performing, why it’s working, and hands you a blueprint to recreate it.

Try it:

https://www.snowballapp.ai/

AI Resource Roundup

Your AI Teammates Aren’t a Team (Yet) by Liza Adams: A sharp take on why stacking AI tools isn’t enough and why orchestration is the real unlock. If you’re experimenting with agents, copilots, or “AI teammates,” this reframes the problem in a way most teams miss.

What’s Working in AI Prospecting: Prompts to Start Using Today: Scale has been running some awesome workshops/webinars recently, and here’s a link to on I attended a few months back. It’s practical rundown of prompts and workflows SDR teams are actually using right now to book meetings with AI.

How AI Is Changing Marketing Attribution (Carilu Dietrich with Alex Bauer and Mada Seghete.This is a thoughtful look at how AI is reshaping attribution models as buyer journeys get messier and harder to track. Like Carilu, I’ve been following Alex and Mada’s journey building Upside from the beginning because it’s the best attribution software I’ve seen.

Hot AI Jobs

VP of Product Marketing at PolyAI

• Location: United States (Remote)

• Pay Range: $175K–$200K

Head of Demand Generation at Ashby

• Location: United States (Remote)

• Pay Range: $230K–$300K

Vice President of Marketing at Pano AI

• Location: United States (Remote)

• Pay Range: $270K–$300K

I’m spending more time lately thinking about signal vs. noise. And not just in GTM systems, but in how we all work.

More soon… and if you’re building workflows like this, I’d love to hear what you’re experimenting with next.

– Brandon