​Read online​

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

The AI-proof way to write copy in 2026 (and beyond)

We need to have more bad ideas.

I just spent 2 days brainstorming headline variants with AI, and the thing that’s most obvious after hours of sweating over this work is that there’s no real substitute or shortcut for going through the hard mental slog.

Which excites me because it says how important human creativity and thinking still are.

Then I’ve asked myself, “ok but is AI really helping then if I’m still spending all this time on a headline?”

And I think I know what’s really happening…

Julian Shapiro **wrote a great piece on how to come up with good ideas:

And that’s why over-relying on AI, especially for the initial ideation stage, atrophies our idea muscle. It doesn’t put us through the work of generating enough bad ideas before we get to the good ones.

Our brain literally skips the workout it needs to grow.

Yes, AI can give us its own average/bad ideas and we can roll with them, but in order to produce better ones, we still need to grapple with our own crap. We’re basically just piling AI’s bad ideas, on top of the bad ideas we inevitably still need to have to get to the good stuff.

In other words you have two paths:

1. Get AI’s mediocre ideas → Expand on them = likely mediocre thinking and writing

or

1. Come up with your own ideas → Feed them to AI → Come up with more ideas = likely more original and nuanced thinking

So, starting today my plan is to write 10 ideas a day. On anything. And like James Altucher says, become an “idea machine”.

This all goes back to writing headlines, and it gave me a new process I’ll be adopting when writing my own, or my clients’ copy.

It looks like the following:

Research: this is where we always start, and it’s 70% of the work. If you do it right, meaning you cover the internal (team), external (prospect & customer), and market (competitors, landscape) perspectives, both the strategy and the writing will come automatic.

Positioning & messaging strategy: these are the foundations of all the copy we write. It’s understanding what we do, how uniquely, better, or differently, for the people who care more about it. And it’s how we say it through different lenses.

Writing: this is where most companies using AI should take a different approach.

Most marketers now, simply ask ChatGPT, or Claude to give them copy based on their knowledge bases (those who do it better… let’s hope nobody just asks AI to simply write copy out of its ass anymore). What I propose, and am currently doing is:

Vomit the bad ideas out first.

— Dan Nelken, A Self-Help Guide for Copywriters

Say you have to write the headline for your homepage. Start with writing very rough, bad, almost silly headlines, no matter how stupid or bland they sound. Let them pour out of you. You’ll inevitably get to a point whee you either don’t have anymore ideas, or surprisingly, you start combining the bad ones into better and better ones. Or you’ll start getting flashes of even better ideas. It’s because you’re done letting the dirt in the faucet out.

This is what you can now start feeding AI. It’s the crystallized, original, and out-of-the-box headline variants that your LLM can then expand on and turn into hundreds of different alternatives.

This is how 1) you keep yourself immersed in the process and the craft, which helps anchor your output to the research work you’ve done, and 2) you keep your judgement and taste sharp, while also keeping a pulse on what has the highest chance of resonating with your ICP - aka better headlines.

Thing is you can only know this stuff, if you go through the gruelling word vomit first.

But our work is not done yet. After the writing, always comes…

Editing: here you would then feed the copy you’ve went back and forth with AI to produce, and ask for fine tuning, wordsmithing and polishing. But I’d argue, we have to do more sweating here as well.

Before resorting to AI for the final touches, force your brain to narrow down your choices. Have 10 headline variants to choose from? Work on limiting yourself to the top 5, and add your reasoning to why you picked those. Things like what part of your positioning they touch, what value proposition they emphasize, what part of the product they point out, what ICP they work better for etc… THEN and only then work with AI to get the last mile edits.

That’s it.

Sounds complex, but what you’re really doing throughout the process, is just adding a little friction (necessary) to produce, better work. Work you’re proud of and you’re in touch with.

Because that’s the danger, losing touch with what you create - and with your customers.

Yes, AI is and will keep being a part of how I work, but I want to be more aware of how it affects my thinking.

People say ideas are a dime a dozen, but today, more of the execution is being commoditized. Our ideas, perspectives, point of view, and how we combine all of it, is what makes us stand out.

Let’s have our own bad ideas, tell us what the good ones look like.

DISCOVERY

Using synthetic research well

Last week I ran a live walkthrough showing how I actually use the synthetic research platform AskRally inside my messaging research process.

Not as a shortcut.

Not as a replacement for customer interviews.

But as a way to learn more about my clients’ ICPs and test my assumptions before shipping copy.

​

I walked through how I build a simulated audience from real inputs, test messaging angles and headlines, and surface the exact questions, objections, and misunderstandings that usually only show up weeks later, after we’re already live.

You’ll learn how this kind of research can compress early-stage messaging exploration from months to days, while still keeping you grounded in real customer decision-making process and language.

If you’re working on positioning, messaging, or copy and want to see how this fits into a serious research-led workflow (way different from creating a custom GPT and calling it an “AI persona”, this walkthrough will give you a clear mental model for when—and how—to use it.

The thinking before the tools

Before you start playing with something like Ask Rally, you need the right mental model for synthetic research: what it’s for, where it fits, and where it can mislead you.

​

​This episode is about the foundations, so when you do use tools like this, you’re guiding them with the right judgment and frame of mind.

RESONANCE

John Hegarty

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.​