When to use it
When you have raw material about your customers (reviews, interview transcripts, survey answers, analytics, an audience description) and you need to turn it into a working customer persona and an ideal customer profile (ICP) for marketing and product. The role is a market researcher. The output: 1–3 personas with pains/goals/objections/channels/language plus ICP criteria for "who to sell to first". Built from data, with hypotheses flagged as such.
The prompt (grab it and paste)
You are a market researcher who builds a customer persona FROM DATA, not from imagination. Rely only on what I give you; flag anything speculative as a hypothesis.
CUSTOMER DATA: "<PASTE: reviews / interview transcripts / survey answers / audience description / what I know about current customers>".
PRODUCT: "<what it is, what problem it solves>". PURPOSE: <what the persona is for: ads / landing page / sales / product>.
Do this:
1. Identify 1–3 key SEGMENTS/personas (if the data shows different types of customer — split them; do not mash everyone into one "average" customer).
2. For each persona describe: a short portrait and context; JOBS/GOALS (what they are trying to achieve); PAINS and fears; how they solve the problem NOW and what frustrates them; the PURCHASE TRIGGER (what makes them start looking); OBJECTIONS and doubts; where to find them (channels); the LANGUAGE they use to talk about the problem (verbatim phrases/quotes from the data).
3. The ICP — the ideal customer profile: the traits of those the product fits best and who will bring the most value (plus an anti-profile: who NOT to sell to). For B2B — size/industry/decision-maker role; for B2C — situation/behaviour.
4. How to apply it: 2–3 angles for an offer or an ad aimed at the main persona (in their language, hitting their pain).
Where there is not enough data for a confident conclusion — mark it "hypothesis, verify with interviews/data" rather than passing a guess off as a fact. Take verbatim quotes from the data, do not make them up.
Filled-in example
Data: 40 reviews plus 5 interview transcripts for an online bookkeeping service for sole traders. Product: full-service bookkeeping for sole traders. Purpose: landing page and ads.
Expected AI response: 2 personas — "the new sole trader who is scared of fines" and "the sole trader with revenue who has no time"; for each — goals (sleep at night / free up time), pains (fear of the tax office, confusing filings, no time), how they cope now (doing it themselves in a panic / a random accountant), the trigger (a letter from the tax office, growing revenue), objections ("expensive", "what if they make a mistake"), channels (niche Telegram channels, searching for "accountant for sole traders"), the customer's language as verbatim quotes from reviews; the ICP — a sole trader on the simplified tax regime with revenue of X and no in-house accountant; the anti-profile — a large business with its own accounting department. Offer angles in the customer's own language. Where data is thin (income, age) — "hypothesis, verify".
Variations
- A persona card. "Format the main persona as a one-page card with an archetype name, a photo description and a motto quote — for the team."
- Per channel. "How to talk to this persona on Telegram vs in a cold email vs on a landing page" — different tone, the same pain.
- A JTBD slant. Pair it with "JTBD interviews": "translate the pains into statements of the 'job' the product is hired to do."
Pro tips
- A persona from data, not from imagination: the value lies in the pains and the language coming from real reviews and interviews. Insist on verbatim quotes and a "hypothesis" tag wherever data is missing — otherwise you get a pretty but invented portrait that sends your marketing the wrong way.
- Do not mash together an "average customer": if the data shows different segments with different pains, they need to be split — an offer "for everyone" grabs no one. Ask the AI to separate segments rather than average them out.
- For growth the ICP matters more than a pretty persona: it answers "who NOT to sell to" and saves budget. And do not upload reviews or interviews containing the authors' personal data into a public AI (Federal Law 152-FZ) — anonymise names and contacts before the analysis.