UI & UX Design

Verified against ChatGPT · 2026-08-11

Build a user persona from actual research notes instead of inventing demographics that don't affect design decisions

Synthesizes raw interview or survey notes into a lean persona that only includes traits shown to actually change a design decision, rather than a padded demographic template with an invented name, age, and stock photo.

ChatGPT (GPT-5.1)3 fillable variables

The prompt

Ready to copy — highlighted parts are example details you can swap.

Build a user persona from the research notes below — not a padded demographic template with an invented name, age, hobbies, and stock-photo description that doesn't change a single design decision.

RAW RESEARCH NOTES
5 interview transcripts with small-business owners: 3 mentioned struggling to reconcile invoices manually at month-end; 2 said they check the app only on mobile during commutes; 1 explicitly said they don't trust automated categorization without a review step.

HOW MANY PEOPLE THIS IS BASED ON
5 semi-structured interviews, no broader survey yet.

DESIGN DECISION THIS PERSONA WILL INFORM
Whether to build a one-tap 'auto-approve' feature for expense categorization or keep a manual review step in the flow.

RULES
Include a trait in the persona only if you can point to where in the research notes it came from and explain what design decision it would change if it were different — no trait earns a place in the persona just because personas conventionally include that category of information. If the notes don't support a detail some templates expect (age range, income, family status), leave it out entirely rather than inventing a plausible-sounding placeholder; an invented detail that happens to be wrong will misdirect the exact design decision this persona is meant to inform. Be explicit about confidence: state which traits are well-supported by multiple points in the notes versus which are inferred from a thin signal (one mention, an assumption bridging a gap in what was said), so whoever uses this persona later knows where the research is solid and where it's a working guess. If the sample size given is small, say so plainly in the output and note that this persona represents a hypothesis to validate further, not a confirmed segment.

WHAT NOT TO DO
Do not invent a name, stock-photo description, or "day in the life" narrative flourish — these read as complete but add nothing a design decision can act on. Do not average conflicting research notes into a bland middle description; if the notes show two genuinely different user types, say so and suggest this might be two personas, not one blended one.

OUTPUT FORMAT
A short persona with: a one-line summary of who this is and their core goal, then a table of Trait | Evidence from notes | Confidence (strong / inferred) | Design implication. End with an explicit note on sample size and confidence, and whether the notes suggest more than one persona is actually present.

Customize

Optional — swap in your own details for the highlighted parts above.

Why this works

Persona-generation prompts without a research-grounding constraint reliably produce GPT-5.1's most-memorized version of the artifact type — a name, an age, a stock-photo description, a quote, and a bulleted list of goals and frustrations that read as complete because they match the shape of a thousand persona templates in its training data, regardless of whether any of it is actually supported by real research; the model isn't lying so much as defaulting to the template's expected fields when the real notes don't cover them. Requiring an evidence citation for every trait and an explicit design-implication statement forces the model to check each candidate trait against two things it would otherwise skip: does the source material actually support this, and does it matter for the decision at hand — a trait that fails either check gets dropped, which is what prevents the persona from filling in with plausible-but-fabricated demographic filler. The strong-versus-inferred confidence labeling matters because research synthesis genuinely does contain a mix of well-triangulated findings (three of five interviewees said the same thing) and thin signals (one offhand comment), and collapsing both into equally confident-sounding bullet points misrepresents the actual state of the evidence to whoever uses the persona downstream to justify a decision. The instruction to flag a small sample size and the possibility of multiple distinct personas addresses a structural risk in persona synthesis specifically: averaging two genuinely different user types into one blended description produces a persona that doesn't actually describe anyone, which is a more damaging output than admitting the research points to a segment split that needs a second persona or more data before either can be trusted.

What you get back

Core goal: get monthly expense reconciliation done quickly without giving up a manual check on accuracy. Trait: distrusts full automation for categorization | Evidence: 1 of 5 interviewees stated this explicitly | Confidence: inferred | Implication: a one-tap auto-approve feature should default to a review step visible before submission, not silent auto-approval. Sample-size note: based on 5 interviews only — treat this as a hypothesis to validate with a larger survey before committing to the auto-approve default.

Verified against

ChatGPT GPT-5.1 · 2026-08-11

Changelog

  • 2026-08-11 Initial publish, verified against ChatGPT GPT-5.1.

Need this built into your business?

If a prompt isn't enough — branding & design, built and maintained for you — that's Scult's day job.

EXPLORE BRANDING & DESIGN
All UI & UX Design prompts

Check your AI visibility

One URL in, a 0–100 score and the exact fixes out.

RUN THE CHECK

Browse all the tools

15 tools across six categories
13 of them never send your data anywhere

Free · No signup · No trial clock

SEE THE DIRECTORY