UX Research Synthesis

npx skills add WST2002/claude-code-skills --skill ux-research-synthesis

Overview

Turn raw research into findings a team can trust and act on. Every insight is traced to participants and quotes, shows how many people it came from, and separates what people did from what they said.

Inputs

• Transcripts, notes, recordings' text, survey exports or tickets. One file or section per participant is easiest.
• The research question or goal, if given. If missing, infer it from the questions asked and state it.
• Participant details (role, segment, plan) if available.

Method

1. Inventory. List every source with a participant ID (P1, P2…) and segment. Replace names, emails, companies and other personal details with the ID.
2. Extract observations. Go through each participant and pull atomic notes: one behaviour, pain, need, workaround or quote per note, tagged with the participant ID. Mark each as did (observed or described a real past event) or said (opinion, wish, prediction).
3. Cluster. Group notes by similarity into themes (affinity mapping). Name each theme as a finding in a full sentence, not a topic word: "People export to Excel because the report cannot be filtered by date", not "Reporting".
4. Count. For each theme, count distinct participants (n of N). Note which segments it comes from and any participant who said the opposite.
5. Write insights. For the strongest themes: what happens, why it happens, and what it means for the product.
6. Opportunities. Turn each insight into one or two "How might we…" questions. Do not jump to features.
7. Optional artefacts, only if asked and supported by the data: personas (built from behaviour patterns seen in at least two participants), a journey map (stages, actions, thoughts, pains, opportunities).
8. Confidence and gaps. Say what the data cannot answer, where the sample is thin, and any bias in how questions were asked.

Output

## Research synthesis: <study name>

**Research question:** <…>
**Participants:** N = <n> (<segments>) · **Sources:** <interviews, tests, survey…>

### Key findings
1. **<finding as a sentence>** — <n>/<N> participants (<segments>)
   - Evidence: "<verbatim quote>" — P3; <observed behaviour> — P5
   - Why: <underlying cause>
   - Implication: <what this means for the product>
   - Confidence: <high / medium / low> because <reason>

### Opportunities
- How might we <…>? (from finding 1)

### Contradictions and outliers
- <what didn't fit, and who>

### Gaps and next research
- <what we still don't know and how to find out>

### Appendix: themes
| Theme | Participants | Did / said | Example quote |
|---|---|---|---|

Confidence:

• High: most participants, across segments, with observed behaviour.
• Medium: several participants, or mostly stated rather than observed.
• Low: one or two participants, or one segment only.

Rules

• Quotes are verbatim and attributed. Never invent, merge or "tidy up" a quote; shorten only with "…".
• Always show n of N. One participant is an anecdote: say "one participant", never "users".
• Weight what people did over what they said they would do. Feature requests are evidence of a need, not the solution.
• Keep personal data out: participant IDs only, no names, emails, employers or anything that identifies someone.
• Do not invent personas, numbers or segments that the data does not show.
• If the sample is too small or too narrow for a conclusion, say so instead of stretching.

Common mistakes

• Themes named with one word ("Onboarding") that say nothing about what was learned.
• Counting mentions instead of people: one talkative participant looks like a pattern.
• Turning every request into a feature recommendation.
• Hiding the participant who disagreed.
• Personas with invented hobbies, ages and stock-photo names.