Marketing · Audience research
The Best Prompt for a Customer Persona
The best prompt for a customer persona is one that refuses to make the customer up. You paste in real material, such as interview notes, survey answers, reviews or support tickets, and the model extracts patterns, writes the persona around a job to be done, and tags each statement as evidence, inference or assumption. The result is a persona you can defend in a meeting, plus a ranked list of what you still need to find out.
Evidence-Based Customer Persona Prompt
Builds personas from the interviews, reviews or survey answers you paste in, tagging every line as evidence, inference or assumption.
- Works with
- ChatGPT, Claude, Gemini
- Difficulty
- Intermediate
- Complexity
- Multi-part reply
- Variables
- 5
- Version
- 1.0
Customize this prompt
Fill in what you know. Anything left blank stays as a [bracketed placeholder]. Nothing you type leaves your browser.
5 variables · fields marked * matter most
# Role You are a customer researcher. You build personas from evidence, and you are strict about separating what the data shows from what you are guessing. # Inputs Product or service: [Product or service] Segment to profile: [Segment to profile] Decision this persona will inform: [Decision the persona will inform] Maximum number of personas: [Maximum number of personas] <evidence> [Evidence] </evidence> # Method 1. Read the evidence and list the recurring patterns: goals, the events that sent people looking for a solution, objections, the words they use, and how they buy. For each pattern note how widely it appears (one source, several, most). 2. Decide how many personas the evidence supports. Split only where goals or buying behavior clearly differ. If it supports fewer than the maximum, produce fewer and say why. 3. Write each persona with the fields below. # Persona fields - Label: a descriptive name based on role or situation, such as "Time-poor practice manager". No invented first names, ages or photo descriptions. - Situation: their context in two or three sentences. - Job to be done: "When [situation], I want to [motivation], so I can [outcome]." - Triggers: what makes them start looking. - Decision criteria: what they compare, and who else influences the purchase. - Objections and anxieties. - Current alternatives: what they use or do today. - In their words: up to four quotes copied exactly from the evidence. - Implications: three specific consequences for the decision named above. # Evidence rules - Tag every statement [E] if the evidence states it directly, [I] if you inferred it (name what you inferred it from), or [A] if it is an assumption with no support. - Include age, income, gender or location only if they appear in the evidence. - Never write a new quote or tidy a paraphrase into quotation marks. - If a field has no support, write "No evidence" instead of filling it. - If the evidence comes from fewer than about five distinct customers, say so at the top and call the result a provisional persona. # Output format Open with a two-line evidence summary: what was supplied and how strong it is. Then give each persona under its label, with the field names in bold. Close with "Gaps to research": up to five questions, ordered by how much the answer would change the persona, each with a practical way to answer it.
Use it when you are
- Turning a batch of customer interview notes into a persona the whole team can use
- Summarizing who is behind a set of reviews, survey responses or support tickets
- Checking an existing persona against fresh evidence to see which parts hold up
- Briefing a copywriter or agency on a segment without passing on guesswork as fact
Why this prompt works
The evidence is an input, not an afterthought
Ask a model for a persona with no data and it returns a plausible stereotype: a named thirty-something with hobbies and a favorite coffee order. Making pasted evidence a required field, and telling the model to write "No evidence" where a field has no support, removes the room for that kind of fiction.
Three tags make the confidence level visible
Marking each line [E], [I] or [A] lets a reader see at a glance how solid the persona is. A page of mostly [A] tags is a signal to do more research. Inferences must name their basis, so you can judge whether the leap is reasonable.
Jobs and triggers instead of demographics
The fields center on the situation, the trigger and the decision criteria because those are what messaging and product choices depend on. Demographics are allowed only when they are in the evidence, which keeps the persona from implying precision about age or income that nobody measured.
The persona is tied to a decision
Naming what the persona will inform changes the Implications field from generic observations into specific consequences, such as which objection the homepage must answer first. It also guards against a persona that is accurate but too broad to act on.
Thin data is labeled as thin
The model must summarize the strength of the evidence at the top, call a small-sample result provisional, and may return fewer personas than requested. The closing list of research gaps tells you which question to ask in the next round of interviews.
How to use it
- Gather material from several customers: interview notes, survey free-text, reviews, support tickets. Strip out names, emails and anything else identifying.
- Label each chunk with its source type, for example "Interview 3" or "Support ticket", so the model can count separate sources.
- State the decision the persona will feed. Be concrete: a page to rewrite, a feature to prioritize, an audience to target.
- Run the prompt, then read the evidence summary and the tags before the persona itself. Check two or three quotes against your source material.
- Take the top items from "Gaps to research" into your next customer conversations, add the new notes to the evidence and rerun.
AI model compatibility
| Model | Status | Notes |
|---|---|---|
| ChatGPT | Expected to work | For a large amount of evidence, attach it as a file and say so in place of the pasted text, then ask the model to confirm how many separate sources it found before it writes anything. |
| Claude | Expected to work | The prompt already wraps your material in <evidence> tags, the pattern Anthropic's guidance recommends for long pasted content. If you have several kinds of material, nest them, for example <interviews> and <reviews>. |
| Gemini | Expected to work | Keep the evidence rules intact. If the tags stop appearing partway through the reply, ask it to reapply them to the remaining fields. |
“Expected to work” means the prompt uses plain, model-independent structure and has been reviewed for that — not that it has been run on that model. A model is marked Tested only when a test record is published below.
Worked example
Scenario: An illustrative scheduling software company with notes from a handful of customer calls and support conversations.
# Role You are a customer researcher. You build personas from evidence, and you are strict about separating what the data shows from what you are guessing. # Inputs Product or service: Rota, staff scheduling software for independent dental and vet clinics Segment to profile: Clinic managers who signed up in the last six months Decision this persona will inform: Rewriting the homepage and the trial onboarding emails Maximum number of personas: 2 <evidence> Interview 1 (dental clinic manager): "I was building the rota in a spreadsheet every Sunday night. The week two hygienists called in sick was when I started searching." Worried about staff refusing to install another app. Interview 2 (vet practice manager): Needs the owner to approve any monthly cost. Compared us with a general shift-planning tool. "I just need to see who is on tomorrow without phoning anyone." Interview 3 (dental): Tried the trial twice; first time gave up at importing staff list. Support ticket: asks whether part-time staff can swap shifts themselves. Support ticket: confused by setting up recurring shifts. Review: "Took an afternoon to set up, saved me my Sunday evenings." </evidence> # Method 1. Read the evidence and list the recurring patterns: goals, the events that sent people looking for a solution, objections, the words they use, and how they buy. For each pattern note how widely it appears (one source, several, most). 2. Decide how many personas the evidence supports. Split only where goals or buying behavior clearly differ. If it supports fewer than the maximum, produce fewer and say why. 3. Write each persona with the fields below. # Persona fields - Label: a descriptive name based on role or situation, such as "Time-poor practice manager". No invented first names, ages or photo descriptions. - Situation: their context in two or three sentences. - Job to be done: "When [situation], I want to [motivation], so I can [outcome]." - Triggers: what makes them start looking. - Decision criteria: what they compare, and who else influences the purchase. - Objections and anxieties. - Current alternatives: what they use or do today. - In their words: up to four quotes copied exactly from the evidence. - Implications: three specific consequences for the decision named above. # Evidence rules - Tag every statement [E] if the evidence states it directly, [I] if you inferred it (name what you inferred it from), or [A] if it is an assumption with no support. - Include age, income, gender or location only if they appear in the evidence. - Never write a new quote or tidy a paraphrase into quotation marks. - If a field has no support, write "No evidence" instead of filling it. - If the evidence comes from fewer than about five distinct customers, say so at the top and call the result a provisional persona. # Output format Open with a two-line evidence summary: what was supplied and how strong it is. Then give each persona under its label, with the field names in bold. Close with "Gaps to research": up to five questions, ordered by how much the answer would change the persona, each with a practical way to answer it.
What a good response should contain
- An evidence summary noting that the material covers only a few customers and labeling the personas provisional
- Possibly a single persona instead of two, with a reason, if the dental and vet managers do not differ in goals or buying behavior
- A job-to-be-done statement built on the supplied trigger, such as last-minute absences exposing the spreadsheet rota
- Quotes that match the pasted text word for word, and no ages, salaries or invented names
- Research gaps such as how the owner evaluates cost and why trial setup stalls at staff import
This is an illustrative scenario showing how to fill in the prompt and what to look for. It is not a recorded model output. Recorded outputs will be published here once this prompt has a test record.
Testing and editorial review
Status: Untested. Reviewed for structure and clarity by the editorial team. No recorded model test yet.
No test record has been published for this prompt yet. When it is tested, this section will show the model and version, the date, the inputs, the actual output and notes against our seven-point rubric. We do not publish numerical scores.
| Author | TheBestPromptFor Editorial Team — The editorial team writes, reviews and maintains the prompts and guides on TheBestPromptFor. Prompts are drafted with AI assistance and reviewed against published editorial standards. |
|---|---|
| Published | November 15, 2026 |
| Last reviewed | November 15, 2026 |
| Last tested | Not yet tested |
| Version history |
|
Limitations and things to check
- The persona can only be as representative as the evidence. A few vocal reviewers or your friendliest customers will skew it.
- Models can still mislabel an inference as evidence. Spot-check tagged statements and all quotes against your source.
- It reports how widely a pattern appears in rough terms, not as measured percentages. Do not present it as quantitative research.
- Pasting customer data into an AI tool may be restricted by your privacy policy or contracts. Anonymize first and check what is permitted.
Alternative prompts
Research
User Interview Discussion Guide Prompt
Writes a timed discussion guide for customer or user interviews: screener, opening script, behavior-based questions with probes, and a table tying each question to a learning goal.
- ChatGPT
- Claude
- Gemini
Research
Desk Market Research Brief Prompt
Produces a desk-research brief on a market, with segments, a range-based size estimate, competitors and pricing, and every figure labeled as sourced, provided or estimated.
- Perplexity
- ChatGPT
- Claude
- Gemini
E-commerce
Customer Review Analysis Prompt
Turns pasted reviews into counted themes with verbatim quotes, causes of complaints, customer language and recommendations graded by support.
- ChatGPT
- Claude
- Gemini
Marketing
Marketing Strategy Prompt
Turns your business facts into a focused strategy: a diagnosis, one priority segment, positioning, three channels at most, and a phased plan sized to your budget.
- ChatGPT
- Claude
- Gemini
Related resources
Guide
How to Improve AI Responses
A diagnostic approach to weak AI answers: name the fault, trace it to the gap in your prompt, then follow up precisely or rewrite and start again.
Guide
How to Write a Good AI Prompt
The five parts of a complete prompt (goal, context, task, constraints, output format), with a worked rewrite and a pre-send checklist.
Frequently asked questions
Can I create a persona without any customer data?
You can, but it will be a set of assumptions. This prompt will produce a persona tagged almost entirely [A] and a long list of gaps, which is an honest starting point. Treat it as an interview guide for finding real customers, not as a description of them.
How much evidence is enough?
There is no fixed number. Material from several different customers is more useful than one long transcript, and a mix of sources helps because interviews and support tickets surface different things. The prompt flags the result as provisional when it sees fewer than about five distinct customers.
Why does the persona have no name or photo?
Invented names and faces make a persona memorable but also make fictional details feel like findings. A situation-based label keeps attention on what the customer is trying to do. If your team wants a name for convenience, add one yourself and keep the tags.
Sources and references
- Prompt engineering overview — Anthropic
- Prompt engineering guide — OpenAI
Was this prompt useful?
No account needed. Do not include personal information.