Your AI Doesn't Suck. Your Context Does.

Everybody has been dwelling on Claude Code skills recently (yours included) but there’s an asterisk that most product marketers are ignoring:

Context.

If you installed the skills from last issue, you probably noticed something. Some of you got great output. Some of you got... the same generic AI garbage you were trying to escape 😬

Same skills and Claude Code but wildly different results. Here’s what’s happening.

I had a call last month with a PMM who had done everything right on paper. Customer research synthesis, ICP profiles, competitive intel, voice guide. Six documents. All in a shared Google Drive.

None of it connected to anything.

Every time she opened Claude, she started from scratch. Copy-paste the ICP. Re-explain the product. Remind it of the competitive landscape. Every. Single. Time.

This is just asking for your todo list to get bigger every week.

Denny Hollick wrote about this recently in PMM Camp (if you haven't read it, you can read the issue here).

His core point: before you touch any AI tool, you need to give it something real to work with.

He recommends six core documents. Customer research synthesis, messaging brief, ICP profiles, social proof library, competitive intelligence, voice guide.

That's solid. Start there if you haven't.

1. Product. Your company, product, key differentiators, top 5 features. Not your marketing site copy. The real stuff. What it does, who it's for, what makes it different.

2. ICP with echo language. This is where most context systems fall apart.

What actually matters is the pain stack and the echo language. The exact words your buyers use when they describe their problems.

Not "struggles with cross-functional alignment." The actual quote from the discovery call: "My CEO keeps asking what we're doing with AI and I don't have a good answer."

That language is what makes AI output sound like you talked to the buyer instead of Googling "B2B SaaS pain points." This section alone takes most people from "AI output is generic" to "wait, how does it know my buyer this well"

3. Positioning. Your category, positioning statement, point of view, and what you're against. This keeps every output anchored. Without it, your agent drifts into generic territory by the second run.

4. Competitive landscape. A table with each of your direct competitors: their claim, strengths, weaknesses, your advantage. Not a 40-page battlecard but a structured reference any skill can query when it needs to differentiate.

5. Voice and tone. 3-word brand voice like phrases you use or never use, buzzwords list and tone references. This is the section that makes people say "wait, this sounds like me" instead of "this sounds like AI slop."

6. Proof points. Stats, testimonials, case studies. When your agent generates messaging, it pulls from real evidence, not invented claims.

Here's what I've learned building agents for the past 6 months: context without a system is a fancy Google Doc.

When I build a PMM agent, day 1 is always the same. We build the CLAUDE.md.

Not six documents but one master file with the right context AKA the brain 🧠 that every agent in the system reads before it does anything.

The PMM I mentioned earlier? She took an hour to consolidate her six documents into one CLAUDE.md. Ran message-market-fit again with the same product. The output went from "sounds like every SaaS company" to "this sounds like it came from our last QBR."

Same skill, but a different context.

If you start with understanding your CLAUDE.md file and setting it properly, you are already much ahead of most people.

After getting 191 clones of my repo on free PMM Claude skills, I decided to finally finish the third tier (execution). You now have access to:

→ battlecard: Build a one-page competitive battle card for sales — sourced from win/loss patterns, review data, and pricing intelligence.

→ sales-narrative: Transform a pitch deck into a structured 7-beat story arc

→ feature-announcement: Turn release notes into a multi-channel announcement package that creates demand, not just awareness.

Btw, just by being subscribed here, you’re already doing yourself a favour. I’m betting that marketers only have four months to master Claude Code, so you’re in the right place.

Most of you are here because you want to get better at messaging and AI. And every skill I've built, every agent I'm working on, reads from this one file.

If your CLAUDE.md is weak, everything downstream is weak. When it's dialled in, everything gets better.

I added to my pmm-skillset repo a CLAUDE.md template you can start from. It'll take you roughly 30 minutes to fill out properly. That file will make everything else you build 10x better.

→ If you haven’t got them yet, you can grab them at pressxtomarket.com/launch

Talk soon 👋

- Gab

P.S. If you still feel overwhelmed by AI, Claude Code, and everything else, I’m looking for 5 in-house product marketers to get their own AI agent built by me, according to their business context, in 7 days.

2.5 hours of your time only, a robot that use my own set of PMM skills within a consistent schedule, and real work artefacts you can show to stakeholders on your weekly calls, all for a crazy discount.

First come, first served, wave starting in May.