😘 The agreeable machine

When I started working on my positioning, one thing genuinely delighted me: after eight years of freelance work, I finally had enough material to do proper evidence-based positioning. Old proposals, client email threads, mentoring reviews, my own published content—I could use all of that to base decisions on, rather than vibes.

I had a whole research plan lined up, with nine sources. I imagined we’d go through it all and then synthesize the learnings into structured positioning options.

Claude had other plans. It wanted to jump to conclusions from step one.

I first handed it my existing website, purely to introduce how I’d been presenting myself so far. I said explicitly: this copy is stale, the positioning is outdated, treat it as background. The moment it finished reading, it started handing me positioning suggestions built on the exact copy I’d told it to disregard.

I believe my exact words were “hold your horses, mate!”

The uncomfortable part is that it sounded good.

Agreeable by design

To be clear, AI absolutely belongs in this work. It solves the problem that kept my repositioning (and probably yours) postponed for years: research time. Wading through years of won and lost deals, mining hundreds of reviews for the customers’ actual words, connecting the patterns across all of it. Weeks of work, done in days.

So I understand why we’re all told to “just use AI for it.” Every marketer I know is getting the same nudge right now, from the AI demos in their feed to a leadership team asking why research still takes weeks.

But the catch is built into the tool itself. A language model is agreeable by design. It converges the moment it can, and it converges on whatever you hinted at. When did a model last tell you your idea was mediocre?

This makes it much harder to spot flaws in the reasoning.

A wrong answer you would catch. An agreeable one sails right past you, wearing your own words.

The rules that kept it honest (mostly)

What saved my project was a small set of standing rules, all versions of the same idea: the model generates, you judge.

Choose your canon before it reads a word. I told Claude which positioning frameworks we would work inside: April Dunford’s seminal work, the Fletch PMM homepage messaging canvas, Louis Grenier’s Stand The F*ck Out. Prime the model with the ideas you trust, then let it range wide inside that frame. Skip this, and it assembles “positioning” from the averaged opinions of the entire internet.

Hold every hypothesis neutral until all the sources are read. This rule went in the minute that “website incident” happened, and it carried the whole project. The model had to finish every source on the list before proposing anything. Left to its own devices, it will try to converge early every single time. Your job is to make it wait. It can be as easy as adding one sentence in your research prompt, no matter how much info the analysis is based on.

Ask for structure before polish. Have it propose the pillars and hypotheses first, and put your checkpoint there, while everything is still rough. Polished-but-wrong wording is far harder to fix than rough-but-right, because a polished draft looks finished, and finished things get waved through on autopilot.

Make it attack its own recommendation. Security teams call this red-teaming: assigning someone to attack your plan on purpose, so the cracks show before reality finds them. It works just as well on positioning. Once the model proposes something, flip it: is this actually differentiating, or does it just sound differentiating? Would a competent competitor put the same sentence on their homepage? Then have it grade the proposal against the expert standards you set on day one.

The one that got through

Deep into synthesis, the model proposed “sameness” as my brand’s enemy. The beige. The force my whole practice supposedly fights.

My red-team prompt existed to kill exactly this kind of idea, but it passed through. I actually liked it enough to sleep on it.

When I looked at it with fresh eyes, though, the problem was obvious: in a market where every expert preaches standing out, declaring war on sameness is the most generic move available. It’s the category’s default sermon. My shiny differentiator was the one claim everyone in my field already makes. Fighting sameness is, ironically, pretty damn samey.

I cut it by hand. No prompt caught it, and I’ve stopped expecting them to.

The process gets you 80% of the way: it turns a mountain of research into a handful of decisions worth examining yourself. The last stretch runs on being human: your understanding of your audience, your market, and the culture around them should dictate what makes it into the brand strategy.

***

AI did the volume I’d probably never have had time to do: 89 proposals, 288 reviews, and eight years of my own content, read and coded in days.

I made every call that mattered, including the one the red-team missed.

This is what it’s all about now: the discipline to make the unflattering call about an idea while a very fluent machine tells you your first draft was great.

And that type of judgment is miserable to practice alone. I know, because I did: no deadline, no colleague, nobody across the table to work with.

That gap is exactly what The Positioning Setup is for. Over two weeks starting August 24, you can run this same process for your business (or to understand how to apply it for clients). The learning is async-first, and the judgment calls happen in a room: live feedback sessions where we discuss the specifics of your project. You leave with a v1 positioning and messaging framework you can trust. If you like it, I’d love to see you there. If it’s not for you, that’s OK. Just remember:

The pushback matters,

👌 Handpicked stories for you

Here are some interesting things I stumbled upon this week while reading, writing, researching, and being a generally curious marketing person on the internet:

💼 Three for work

Ownable Ideas: Why the Wealthiest Experts Own Ideas That Travel Without Them | Katelyn Bourgoin’s in-depth primer on ownable ideas, the kind that build authority while traveling without you. She maps six forms an idea can take, from coined problems to identity labels, with named examples for each.

14 Red Flags That Make Your Writing Sound Like AI (+How to Fix It) | A list of patterns that make copy read machine-made, each with a fix. Read it once, then hand the list to your AI and have it fold these patterns into your own AI cliché blacklist, so the check runs before you ever see a draft.

If I Were Starting My Marketing Career Today, Here’s What I’d Learn First | Ross Simmonds on what he’d learn first if he were starting over: distribution before creation, one channel mastered deeply, the business side of marketing. AI-native workflows make the list too, but as one skill among six. I’d agree with that weighting.

👀 Two for fun

This scroll/stroll through pixel-art cities | Urban backgrounds from old video games, one moody skyline at a time. My kind of doomscrolling.

This game of onboarding horrors | I laughed, then sent it to my UX friends. It counts your rage clicks, and it hits a bit too close to home.

🐈 Weekly cat

Looking up is the right call.