Best AI Prompts for UX Personas (2026)
Prompts for UX personas#
Five starting points#
Act as a senior UX researcher. Below are my raw interview notes from [NUMBER] participants of [DESCRIBE YOUR PRODUCT AND USER GROUP]. Group participants by shared goals, attitudes and pain points, then write one persona per group with name, goals, frustrations, context of use and a one-line summary. Tag every statement as EVIDENCE (with the participant IDs it comes from) or ASSUMPTION (not in the notes). Do not add any detail the notes do not support. <notes>[PASTE YOUR INTERVIEW NOTES]</notes>Act as a product strategist. Draft a proto-persona for [DESCRIBE YOUR TARGET USER] using only the assumptions I list below. Mark the whole document "UNVALIDATED: team assumptions, no research". For each assumption, write the cheapest research question and method (interview, survey, analytics check) that would confirm or kill it. <assumptions>[PASTE YOUR TEAM'S ASSUMPTIONS]</assumptions>Act as a research analyst. Here are [NUMBER] survey responses and the segments we already found in interviews: [PASTE YOUR SURVEY EXPORT AND INTERVIEW SEGMENTS]. Suggest how the responses map onto the segments, count how many responses support each, and list responses that fit no segment. Do not infer personality, age or income from the answers. State where the sample is too small to draw a conclusion.Act as a B2B user researcher. I am preparing to interview [JOB TITLE, e.g. CTOs] about how they decide to adopt [DESCRIBE YOUR PRODUCT CATEGORY]. Write a hypothesis list of likely decision criteria, labelled as hypotheses, and an interview guide with neutral, non-leading questions that test each one. Leave the persona fields blank so I can fill them in after the interviews.Act as a consumer researcher. Below are [NUMBER] real customer reviews and support messages about [DESCRIBE YOUR PRODUCT]. Identify the distinct buyer motivations and decision criteria, quote the exact words that support each, and name a short archetype for each. Flag any archetype supported by fewer than three quotes as weak. <reviews>[PASTE YOUR REVIEWS OR TICKETS]</reviews>What to add to any UX persona prompt#
- Your real research data, pasted in or attached, and an instruction to use nothing else. Anthropic's guide says to supply context and to wrap documents in tags, and OpenAI's guide says to add the data the model needs.
- A rule that every claim is tagged EVIDENCE or ASSUMPTION. Nielsen Norman Group says personas should be based on information about real people.
- A request to quote the notes. Anthropic recommends asking the model to quote relevant parts of long documents before it carries out the task.
- The decision the persona is for. Nielsen Norman Group says each piece of information should have a purpose, so say what the team will design or decide.
- A ban on filler such as hobbies, taglines and stock-photo details unless the research supports them. Nielsen Norman Group warns against irrelevant details.
- A response format, such as one table per persona. Google's Gemini guide lists response format and constraints as things to state.
A worked example#
Act as a senior UX researcher. Below are my raw interview notes from 8 participants of a budgeting app for freelancers. Group participants by shared goals, attitudes and pain points, then write one persona per group with name, goals, frustrations, context of use and a one-line summary. Tag every statement as EVIDENCE (with the participant IDs it comes from) or ASSUMPTION (not in the notes). Do not add any detail the notes do not support.
<notes>
P1: Invoices late clients every month, sets aside tax "when I remember". Uses a spreadsheet.
P2: Hates mixing personal and business accounts. Wants one number: what is safe to spend.
P3: Irregular income; panics in slow months. ...
(P4 to P8 pasted here)
</notes>
- Each persona cites participant IDs for every goal and frustration.
- Assumptions are listed separately and none are hidden inside the narrative.
- The number of participants behind each persona is stated, so a persona built on one person is visible as such.
- Participants who fit no group are listed, not forced in.
- No demographic or personality details appear that the notes do not contain.
The full prompt library is on our prompt library page.
Limits of this guide#
Nielsen Norman Group establishes that personas need research behind them and that synthetic users are hypotheses at best. It does not test any particular prompt, and neither did we. We did not measure how well a model clusters real interview notes compared with a human researcher. The persona-types article gives 5 to 30 interviews as a rough guide for qualitative personas, which is its own range, not a rule. GOV.UK's service manual user research page does not mention personas, so we do not cite it for them.
Sources & Further Reading#
Every source below was fetched on 2026-10-22, except the OpenAI row, which was read on the live page on 2026-10-22 and not through an archive copy.
| Claim | Source | Section or date |
|---|---|---|
| Personas must be based on user research; each detail needs a purpose; avoid irrelevant details; do not let personas replace user testing | Personas Make Users Memorable (Nielsen Norman Group) | Taylor Dykes, October 3, 2025 |
| Synthetic users are idealized, lack prioritization, give no behavioral data; treat as hypotheses; do not present as real-user findings | Synthetic Users: If, When, and How to Use AI-Generated "Research" (Nielsen Norman Group) | Maria Rosala and Kate Moran, June 21, 2024 |
| Proto personas are untested assumptions; qualitative personas from about 5 to 30 interviews; avoid personas built on demographic or analytics correlations | Persona Types (Nielsen Norman Group) | Page Laubheimer, June 21, 2020 |
| Add context, be clear and direct, tag documents, ask for quotes first | Prompting best practices (Anthropic) | Add context; Structure prompts with XML tags; Ground responses in quotes; undated |
| Delimit content with XML or Markdown, include relevant context, be explicit | Prompt engineering (OpenAI) | Message formatting; context; being specific; undated |
| State constraints and response format; supply the context the model needs | Prompt design strategies (Google) | Clear instructions; add context; undated |
| GOV.UK user research page does not mention personas | User research (GOV.UK Service Manual) | Undated |
Need a different task?
Browse all 36 tasks and 180 prompts, with search and level filters.
Frequently asked questions
What are the best AI prompts for UX personas?
Paste your real interview notes, survey answers or support tickets into the prompt and have the model group them into personas, tagging every claim as evidence or assumption. Personas invented from nothing are guesses, and Nielsen Norman Group says they must rest on user research.
Can AI replace user research when building personas?
No. Nielsen Norman Group found synthetic users give idealized accounts, care about everything equally and produce no behavioral data. It allows them for early exploration if you treat the output as hypotheses to test, and says not to present them as real-user findings.
How many people do I need to interview before an AI can help build personas?
Nielsen Norman Group suggests roughly 5 to 30 interviews for qualitative personas, grouped by shared attitudes, goals and pain points. A model can cluster your notes faster, but it cannot tell you what share of your users each persona represents.
By TheBestSuperintelligence Editorial Team. Last reviewed .
How we sourced this. Model names, features and limits change often. The prompts below are our own untested starting points, and we did not measure which wording performs best. Facts are as printed on each source page on the date shown. Methodology