Guide
How to summarize user interviews with AI
ChatGPT or Claude can turn a stack of user interview notes into something the team can act on, if you ask for evidence rather than a summary. Have it break the notes into single observations, each tagged as something a participant did, said or wants (past behaviour is the strongest signal, requests the weakest). Group them into themes by content and count participants per theme ("5 of 8"), with one or two verbatim quotes each. Ask for jobs-to-be-done for the top themes, insights kept apart from what the team already assumed, and recommendations with a confidence level. Replace names with labels like P1 to P8 before pasting. A plain "summarize these interviews" tends to repeat the loudest participant.
Short answer
- Observations tagged did, said or wants.
- Themes with counts and verbatim quotes.
- Insights apart from assumptions.
- Names replaced with labels before pasting.
The brief to copy
Interview synthesis brief
"Synthesize these interview notes.
Break them into single observations, one fact about one participant each, tagged did, said or wants.
Group by content, then name the groups. Count participants per theme ('5 of 8'). A theme needs at least 2 participants; one-person observations go under 'Worth watching'. Say if a theme comes from one segment only.
One or two verbatim quotes per theme, with the participant label; mark paraphrases.
For the top themes, a job: 'When [situation], I want to [motivation], so I can [outcome].'
Separate insights (behaviour from 2 or more participants) from assumptions the data didn't test. List contradicting evidence. Recommendations with High, Medium or Low confidence.
Notes: [paste, with names replaced by P1, P2…]"
A worked example
Illustration: a made-up business. A booking app interviewed eight studio managers about how they handle cancellations. The synthesis, in part:
From the synthesis
Theme: they fill late cancellations by hand (6 of 8). Did: P1, P2, P4, P5, P7 and P8 text regulars when a spot opens. Quote, P4: "I've got a WhatsApp group of twelve people I message at 6am."
Job: When a class spot opens at short notice, I want to offer it to someone who'll actually come, so I can keep the class full without spending my morning on it.
Worth watching (1 of 8): P3 wants a deposit for every booking.
Assumption, not tested: that managers want cancellations fully automatic. Two managers (P2, P7) said they'd want to choose who gets the spot.
Recommendation: an offer-the-spot flow the manager confirms, High confidence (behaviour from 6 of 8).
The researcher checks every quote against the notes before the synthesis goes to the product team, and the "fully automatic" idea stays an open question.
Before you share it
- Check every quote against the notes, word for word.
- Check the counts: a theme with one participant isn't a theme.
- Keep assumptions labelled as assumptions.
- Replace names and contact details before pasting; use your company's account.
- Save the brief for the next round of interviews.
Next step: from themes to a spec
A synthesis usually turns into requirements. Reviewing a product spec with AI before the team builds it covers checking that spec, and writing a PRD from your notes covers drafting it.
Skills for this
- ProductUser interview synthesisSynthesizes interview notes into themes with participant counts, quotes, jobs-to-be-done, insights vs. assumptions, and recommendations with confidence.
- StrategyJobs-to-be-done switch interviewPlans and synthesizes JTBD switch interviews: a timeline guide, a timeline and four-forces summary per interview, and a job statement across interviews with decisions.
- Customer successFeedback themes from NPS and surveysAnalyzes NPS comments, survey answers, or reviews into themes with counts, typical quotes, sentiment, and one suggested action per problem.
- ProductPRD from messy notesTurns notes, Slack threads, and call notes into a PRD: problem with evidence, goals and non-goals, metrics, must/should/could requirements, and flagged gaps.
Questions
Can AI summarize user interviews?
Yes, if you ask for evidence rather than a summary: each observation tagged as something a participant did, said or wants, themes with how many participants support them, verbatim quotes, and insights kept apart from what the team already assumed. A plain summary tends to repeat the loudest participant.
How many participants make a theme?
Set a rule and keep it. A common one is at least two participants; anything from one person goes under "worth watching". Always write the count ("5 of 8"), and say when a theme comes from one segment only.
What shouldn't I paste into ChatGPT from interviews?
Participants' names, contact details and anything they asked you to keep private. Replace names with labels such as P1 to P8 before pasting, and use your company's account; on a personal plan, turn off model training first.
AI skill
Have your AI pull insights out of your user interviews
Put this skill in AI Skill Pod and everyone on your team gets it in their Claude, Cursor or ChatGPT (Pro, Business, Enterprise): one invite link, free to start. With User interview synthesis, each person's AI synthesizes interview notes: observations tagged did, said or wants, themes with participant counts, verbatim quotes, jobs-to-be-done, insights apart from assumptions, and recommendations with a confidence level.
14 days free · No card needed
Sources
- Nielsen Norman Group: Affinity diagramming — group by content first, name the groups afterwards (read 2026-11-04)
- Christensen Institute: Jobs to be done — the progress a person is trying to make in a particular situation (read 2026-11-04)
- OpenAI Help: Data controls — on Free, Plus and Pro personal accounts, model training is on by default and can be turned off; read through search-result text (read 2026-11-05)