Positioning is the one decision every other decision copies from. Your website, your deck, your sales call — they all inherit from it. It's also the thing founders most often skip, because it's abstract, there's no artifact to point at, and it feels like procrastination dressed up as strategy.
AI changes the economics. You can now draft and pressure-test your positioning in an afternoon instead of avoiding it for a quarter. What AI can't do is decide — and deciding is the whole point.
Start with the frame
The old frame still works because it forces the hard choices: For [audience] who [need], [product] is a [category] that [benefit], unlike [alternative]. Fill it in, then say it out loud to a stranger. If they don't immediately get what you do, it isn't done.
Write it three ways — three audiences, or three angles — and compare. AI is genuinely good at generating the variations fast. You're looking for the one that's specific enough to exclude people. Positioning that tries to be for everyone is for no one.
Use AI as an interrogator, not an author
The highest-leverage move isn't asking AI to write your positioning. It's asking it to interrogate you. Give it your rough answers and have it pressure-test them: Where's the proof? Who exactly is this for? What are you quietly avoiding? The model is a tireless, slightly annoying analyst — use it that way, and it surfaces the gaps you've been stepping around.
This is the same discipline as everything else: AI drafts, you ship. It generates and challenges; you decide and commit.
Three pillars, each with a proof point
Once you have the statement, back it with three messaging pillars — the three reasons someone should care. The rule that keeps this honest: every pillar gets a real proof point, or you mark it as [NEEDS: proof] and go get one. A pillar with no evidence is a wish, and readers can tell.
AI will happily invent impressive-sounding proof. Don't let it. A flagged gap is honest; a fabricated claim is the fastest way to lose trust.
Decide — the part AI won't do
At some point you have to choose one positioning and let everything inherit it. That's not a prompt; it's a judgment call about what you're actually building and who you're building it for. The model can lay out the options cleanly. It cannot care, and it cannot commit. You do that.
Get it right once and the rest of the launch stack gets dramatically easier — because every piece is now copying from a spine instead of improvising. If you want to run this with a group and real review, it's the most important module in our course.