How I Turn Claude Code (Desktop) Into a Full Marketing Team
How to build and automate real marketing work without writing code.
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!
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
Well done. Very thoughtfully composed