​Read online​

Welcome to Unpacking Meaning. If you received this from a friend and enjoy it, ​subscribe here.

The copywriter’s guide to not outsourcing your brain to AI

Last week we talked about how AI is stealing our agency, and how the solution is to start using it more like a compass vs a GPS.

I was in the middle of a copy refresh for a client we’d just finished a full positioning and messaging overhaul for, and somewhere in that process I caught myself thinking: this is exactly what compass mode looks like when it’s actually working.

So get your coffee, sit next to me, and let me walk you through it.

The engine behind the output

A few weeks ago I was listening to Lazar Jovanovic on Lenny’s Podcast. He’s a professional vibe coder at Lovable, meaning his full-time job is building production-quality software using AI, without a traditional coding background.

What caught my ear wasn’t really the coding part. Everyone and their grandma are spinning up apps now. It was his system and how he thinks about the craft.

Lazar doesn’t write better prompts. He builds product requirement documents and markdown files that give AI the full picture of what it’s building, why, and within what constraints. The AI doesn’t have to guess. It just executes within a well-designed context.

That’s exactly what I’ve been doing when I use AI to assist with writing copy.

There’s a term for this now: context engineering. It means designing the right information environment so the AI can actually do its job. It’s probably the most important shift in how we work with these tools, and again, it has almost nothing to do with writing better prompts.

Here’s why this matters for anyone writing or managing copy with AI.

When you ask an LLM to write your homepage without context, it pulls from every SaaS homepage it’s ever seen. You get a smoothie made from the entire internet.

With proper context design, you build a small ecosystem of documents that act as the AI’s strategic brain for your specific product and audience. As models get better and evolve, my system also constantly changes. But here’s what that looks like right now:

• ICP Sheets. In-depth persona breakdowns grouped by go-to-market motion. For example, sales-led and PLG get separate sheets, because the roles, pains, and decision-making patterns are different.
• Positioning Canvases. One per product type or ICP segment, defining where you sit, what you’re compared against, and why you’re different.
• Messaging Framework. Key messaging pillars, differentiators, proof points, and strategic narrative. For complex platforms, I expand this into ICP-specific playbooks.
• Value Proposition Canvases. One per ICP. This is our positioning translated into pains, implications, solutions, benefits, jobs-to-be-done, and examples of copy-ready value props.
• Brand Voice Guide. Vocabulary, cadence, tone, rhythm, guardrails, and how the voice changes across channels.
• System Instruction Prompt. The instructions telling the AI which document to reference for which task, so it doesn’t burn context trying to figure that out on its own at every prompt.

That last one is the piece most people skip, and it’s the one that makes the biggest difference. Without it, the AI has to interpret all your documents every time, or worst, you have to ask it to go and look at them based on your task. With it, you’re giving the model a route: for this audience, look here. For this type of page, pull from there etc.

The whole thing works like a messaging operating system.

The documents are the programs, the system prompt is what runs them, and every time you start a new conversation, you’re booting the AI up with the full strategic picture rather than starting from zero.

This is also how I keep things consistent across my team. Everyone writes from the same documents, the same voice, the same research foundation for every client.

So if the copy coming out of your AI workflows feels like slop, the problem probably isn’t the model. It’s the context you’re not giving it.

How the actual work moved

With that context in place, I don’t simply hand AI the whole page to write. I work section by section, because each section has a specific job to do, and you can only evaluate the output if you’re clear on what that job is.

Like every single legend copywriter said, the job of the headline is to get people to read the next sentence and so on. It’s all a series of micro decisions our readers have to make.

So, in short when I prompt for each section this looks like:

• here’s what this section needs to accomplish and for which ICP
• here’s the new strategy it needs to reflect
• give me a first draft

Then I look at what comes back, push it to explore different directions, ask it to go back into the strategy docs for angles it might have missed, get a few variants, and then go in myself and do the final copy-editing work.

That back and forth is the compass approach working its magic. You’re not following turn-by-turn directions, but reading the terrain, adjusting and using your own judgment at every step.

The trap

Here’s where it gets interesting, and honestly a little uncomfortable to admit.

The better the models get, the easier it is to stop thinking.

Because the copy looks good, feels fluent, structured, and so you move on.

But fluency isn’t the same as coherence. And coherence at the section level isn’t the same as coherence across the whole page.

What I’ve started doing, after the first draft is done for a page, is stepping back and asking a completely different set of questions:

• Is the big idea actually coming through here?
• Is the value proposition strong enough?
• How many times have we repeated this word, this idea, this structure, without noticing?

These are questions AI can’t ask for you. That judgment is yours, and the trap is forgetting that.

What this actually means for your work

Compass mode is an orientation toward the work that will always work, no matter the tech behind it.

It means knowing what each piece of copy needs to do before you ask AI to help you do it.

It means designing your context deliberately.

It means evaluating output against strategy.

And it means being the one who holds the full picture, because no one else in the process is doing that.

AI will keep getting better at producing fluent, convincing, well-structured copy. That’s not the question anymore.

The question is whether you’re using it to think more clearly, or using it as a reason to think less.

That’s the difference between compass and GPS. And it shows up in the work and in how your prospects perceive you.

DISCOVERY

Marketing in the Age of AI

Earlier this week I joined Emanuel Rose for a live conversation on Marketing in the Age of AI.​

We talked about something I think a lot of teams are currently wrestling with: how AI changes the way we do marketing without changing the fundamentals of why marketing works.

I'm running my OpenClaw agent 🫨

Ok, I lost it to the FOMO and decided to try installing my own OpenClaw bot. I have to say it's pretty crazy even though quite expensive to justify it, for now.

In 2 days I've connected it to:

• Notion and Google drive for my knowledge base
• Fathom for call recordings
• my calendar and email so it knows about my schedule and routine
• Readwise so it can read all the highlights I save from stuff I read

and a few other bits like allowing it to "read" my voice notes.

It can prep proposals after client calls, give me ideas for this newsletter based on what I consume, and literally anything I can think of. That's the limit for now, thinking of what to delegate. I will have updates soon as I test it for messaging work as well!

If you're curious and brave enough, here's a great guide to getting started.

Best practices vs market research

Here's a great piece by Amanda at Spark Toro, reminding us how "best practices" often fail when they meet the reality of our work. Especially when that reality is changing every week. What should we do then? Always go back to your market.

RESONANCE

Richard Shotton, The Choice Factory

Have a great weekend!

Cheers,

Chris

Chris Silvestri Founder & conversion alchemist

🙌🏻 Let’s be friends (unless you’re a stalker)​

When you're ready, here's a few ways I can help

Not sure where to start? Take our free message-market fit scorecard.​