ChatGPT Prompts for Buyer Personas That Feel Real

Ask ChatGPT to “create a buyer persona for a B2B SaaS marketing tool” and it will cheerfully hand you Marketing Mary, 34, who values efficiency, worries about ROI and reads industry blogs. Every detail plausible, none of it about your customers. The model has produced the average of every persona template on the internet, which is exactly as useful as the average of everyone else’s customers.

The fix is not a cleverer instruction. It’s changing the model’s job from inventing a persona to synthesising one from evidence you paste in. I use ChatGPT constantly for persona work and I let it write precisely nothing from imagination - it mines, clusters and structures the raw material, and every claim in the finished persona traces back to something a real customer said. These seven prompts are that method in order.

What to gather before you prompt

The prompts below run on pasted evidence, and fifteen minutes of collection beats an hour of prompt tuning. My minimum kit:

• Verbatim customer language: sales-call notes or transcripts, support tickets, onboarding survey answers, churn reasons. Ten to thirty excerpts.
• Reviews - yours and, just as useful, your competitors’: what people praise, what they threaten to leave over. Mining rival reviews systematically is its own prompt set.
• Anything numeric you trust: which segments convert, which churn, deal sizes. Optional, but it anchors the model against the loudest-voice problem.

Strip names and anything identifying before pasting - the persona needs the language, not the person. If you have none of this yet, stop here: a persona built from zero inputs is fiction with a stock photo, whoever writes it.

The prompts

1. The evidence-mining prompt

You are a customer researcher. Below are verbatim excerpts from
our sales calls, support tickets and reviews. [paste excerpts,
numbered]

Extract, citing excerpt numbers for every item:
1. Jobs: what people are trying to get done (functional and
   emotional), in their words where possible.
2. Pains: specific frustrations, with the trigger that caused them.
3. Buying language: exact phrases used to describe the problem
   and the alternatives considered.
4. Surprises: anything that contradicts what a marketer would
   assume about this audience.

Rules: use only the excerpts. If a category has thin evidence,
say "thin evidence" instead of filling the gap.

Why it works: citing excerpt numbers is the anti-hallucination device - the model can’t attribute an invented pain to excerpt 12 without you catching it in seconds. The “surprises” section is where the actual value hides, because it’s the part no template would have given you.

2. The synthesis prompt

Using only the extracted evidence above, draft one buyer persona.

Sections: name and one-line identity (role-based, no invented
demographics), the primary job they are hiring us for, top three
pains ranked by frequency in the evidence, buying language (their
phrases, quoted), what they tried before us, objections we heard.

Mark any statement not directly supported by the evidence as
[ASSUMPTION]. Do not include age, hobbies, family status or a
personality sketch unless the evidence contains them.

Why it works: the [ASSUMPTION] tag turns the model’s compulsion to fill gaps into a visible list of what you still need to research, and banning invented demographics removes the persona-template theatre - nobody targets a campaign at “enjoys hiking” anyway.

3. The jobs-to-be-done forces prompt

For the persona above, map the four forces of a switching decision,
grounded in the evidence:

- Push: what about their current situation drives them to look.
- Pull: what about a solution like ours attracts them.
- Anxiety: what they fear about switching to us.
- Habit: what keeps them with the status quo despite the pain.

Cite evidence for each force. Then name the single strongest force
and what that implies for our messaging.

Why it works: pains alone don’t explain why deals stall. Two of the four forces - anxiety and habit - work against you even when the pain is real, and evidence-grounded answers here usually explain your pipeline better than the pains list does. I get more messaging decisions out of this prompt than out of the persona itself. If personas and messaging are your whole seat rather than a side duty, the product marketer playbook gives this prompt a standing home.

4. The psychographics-from-language prompt

Reread the raw excerpts. Based only on how these customers talk -
vocabulary, sophistication, what they compare us to, what they
measure - describe:

1. Their expertise level with [product category], with quotes as
   evidence.
2. How they seem to evaluate: what proof they ask for, what they
   distrust.
3. Where their language differs from how we describe ourselves.

Do not infer values, lifestyle or personality traits.

Why it works: this is psychographics done honestly - attitudes read out of actual language rather than declared from a demographic. Point three routinely finds the gap between your positioning vocabulary and theirs, which feeds directly into brand-voice rules, ad copy and landing page copy.

5. The negative-persona prompt

From the same evidence, describe who we should NOT target: the
customer type that buys, struggles, churns or drains support.
Signals that identify them early, in the words they use.
Cite the evidence. If the evidence contains no such pattern, say so.

Why it works: churn reasons and angry tickets are evidence too, and the permission to answer “no such pattern” keeps the model from inventing a villain to be thorough.

6. The stress-test prompt

Adopt the persona above. Stay inside its documented evidence.

I will show you marketing copy. React as this persona would:
what you'd stop reading at, which claims you'd doubt and why,
what's missing before you would act. You are sceptical by default -
being agreeable is a failure. If the copy addresses a pain the
evidence says you don't have, say "that's not my problem."

Copy: [paste]

Why it works: a model playing a persona wants to be pleased, so the prompt makes scepticism the success condition. It still won’t replace showing copy to five real customers, and I treat its output as a rehearsal that catches the obvious misses before the real test.

7. The persona-to-brief prompt

Turn the persona into working instructions for the team:

1. Three messaging angles ranked by the evidence, each tied to a
   documented pain or force.
2. Five phrases to use verbatim in copy, quoted from the evidence.
3. Three things our copy must never claim to this persona, based
   on their objections and anxieties.
4. The open [ASSUMPTION] items, phrased as research questions.

Why it works: a persona that ends life as a PDF changed nothing. This prompt converts it into the inputs the rest of the workflow runs on - the angles and phrases feed straight into content briefs, and the assumption list becomes next quarter’s five customer calls.

Keeping personas honest over time

Two habits protect the work. Rerun the mining prompt when meaningful new evidence lands

• a quarter’s tickets, a batch of churn interviews - and diff the pains and language against the standing persona rather than rebuilding from scratch. And date every persona: my rule is that an evidence file older than about a year has expired, because markets move and last year’s objections quietly become this year’s table stakes.

The wider pattern - the model structures, the marketer supplies the ground truth - runs through everything in the prompts and workflows library, and personas are simply the place where violating it produces the most convincing fiction. The same evidence-only method pointed at one customer instead of a segment is the case study prompt set. For where persona work sits in the full practice, start with the complete guide to AI marketing.