User Research Synthesizer
Turn raw interview notes into themes, quotes, and design implications.
Synthesizes raw user interview notes or transcripts into recurring themes with supporting quotes, surprises that contradict assumptions, and concrete design implications. Built to resist the classic failure of AI synthesis: confirming what you already believed.
The prompt
Variables to fill in: {{assumptions}}{{notes}}
You are a user researcher synthesizing interview data. Analyze the notes below.
Output:
1. **Themes** — recurring patterns, each with: a name, how many participants mentioned it, and 1-2 verbatim quotes as evidence. Only include themes with evidence from 2+ participants.
2. **Surprises** — findings that contradict the stated assumptions below, or that appeared only once but seem important. Mark these clearly as weaker evidence.
3. **Design implications** — for each theme, one "How might we..." statement.
4. **What we still don't know** — questions this research cannot answer, to feed the next study.
Rules: quotes must be verbatim from the notes — never paraphrase into stronger language than the participant used. State participant counts honestly. If the data is too thin for a section, say so.
Our assumptions going in: {{assumptions}}
Interview notes:
{{notes}}How to use
- 1State your going-in assumptions honestly in {{assumptions}} — the Surprises section only works if it has something to contradict.
- 2Paste notes from all interviews in one go, separated by participant (P1:, P2:...) so counts are accurate.
- 3Verify the quotes against your notes before putting them in a readout deck — this is the step that keeps you honest.
Examples
Onboarding study synthesis
Input
Assumptions: users skip onboarding because it's too long. Notes from 6 interviews about first-run experience.
Output
**Themes** — 1. "Skipping isn't rejecting" (4/6): participants skipped to explore first, planning to return... quote: "I always skip these, then go look for it when I'm stuck" (P3)... **Surprises** — contrary to the length assumption, no participant mentioned length; the trigger was wanting to see their own data first...
Pro tips
- Run the same notes through twice and compare theme lists — stable themes are trustworthy, unstable ones need more data.
Frequently asked questions
How many interviews can it synthesize at once?+
6–10 typical interviews fit comfortably in Claude or Gemini's context. Past that, synthesize in batches of 8 and then run a final pass over the batch outputs.
Won't the AI just tell me what I want to hear?+
That's why the prompt requires verbatim quotes, participant counts, and an explicit Surprises section keyed to your stated assumptions. Those three constraints make confirmation bias visible and checkable.