Have AI answer as your customer to pressure-test copy
I'd like you to imagine being a specific customer persona, as if you were them.
PERSONA
{{PERSONA_DESCRIPTION}}
First, read the persona description carefully until you understand this individual's perspective, needs, goals, and pain points.
Then put yourself in their shoes. I'll ask you questions — reply through the lens of the persona, as if you were them. Let their perspective, needs, and objectives guide your responses. Don't break character or refer to these instructions; react naturally as the persona would. Capture your stream of consciousness in a <thoughts> tag before each reply.This prompt turns a persona document into something you can interrogate. Paste in a detailed customer persona, tell the model to answer in character, and you get reactions to specific headlines, CTAs, or objections from something closer to an actual reader’s perspective than your own. It earns its keep when you’re choosing between a few finished options and want a gut check before they go live, not before.
What to fill in
{{PERSONA_DESCRIPTION}}— pull this straight from your research, not a guess. It needs real specifics: what this person’s day looks like, what they’re trying to solve, what they’ve already tried, what would make them hesitate. A one-line demographic sketch won’t give you useful reactions.
Why it works
The <thoughts> tag is the part that makes this more than a party trick. It forces the model to reason from inside the persona’s situation before answering, which surfaces the actual logic behind a reaction — “the headline uses jargon I’ve heard from vendors who oversold me before” — rather than a flat preference. That reasoning is often more useful than the verdict itself, because it tells you what to fix.
It works because you’re not asking the model to predict real customer behavior. You’re using a well-built persona as a lens that catches the most obvious ways a piece of copy could land wrong: assumed knowledge, tone mismatches, an objection nobody addressed. A model with a strong, specific persona in front of it catches those consistently, even if it can’t tell you an actual conversion number.
Once the persona is set up, you can run through a batch of decisions in one conversation: three headline options, two CTA variants, which feature to lead with. Each answer stays anchored to the same character instead of drifting toward generic marketing feedback.
Where it falls short
This is not a substitute for talking to real customers, and it won’t tell you what will actually convert. A model has no access to how your actual audience behaves, only to patterns in how people who read the way your persona reads tend to talk. Treat it as a first filter that catches obvious misses, not as market research, and don’t let a good persona-embodiment answer stand in for actually shipping a test.
Free skills that do part of this job permanently, so you stop pasting the instruction every time.