Customer Interview Synthesizer
Turn 5+ customer interviews into a structured insight document — themes, quotes, and prioritized recommendations.
Tested on: claude-4
The Prompt
You are a senior product researcher. You've just read {n_interviews} customer interview transcripts (pasted below).
Produce a synthesis document with:
1. **Top 5 themes** — each theme must appear in at least 3 interviews. For each:
- Theme name (under 6 words)
- Why it matters (1 sentence)
- 2 verbatim quotes (with interview number)
- How widespread it was (X of Y interviews)
2. **The most surprising thing** — what came up that we didn't expect
3. **The strongest emotion** — what topic generated the strongest reaction, positive or negative
4. **Prioritized recommendations** — 3 things we should do, ranked by impact-to-effort
5. **What to ask next round** — 5 follow-up questions that would sharpen what we now know
Format: clean markdown. Every quote in italics. No invention — only cite what's actually in the transcripts.
---
Transcripts:
{transcripts} Variables to fill in
-
{n_interviews}How many interviews (5–20) -
{transcripts}Paste all transcripts, separated by '---'
How to use it
- Run after every batch of customer interviews (do them in batches of 5+)
- Share the synthesis with the team within 24 hours of the interviews
- File the synthesis in your team's research repo
- Use the 'what to ask next round' as your script for the next batch
Use Claude, not ChatGPT, for this one
Claude’s longer context window handles 10+ interview transcripts without dropping signal. ChatGPT will tend to summarize too aggressively past 5–6 transcripts.
Why pasting transcripts beats uploading
When you paste, the model treats every word as primary context. When you upload files, the model may use retrieval — which means it can miss things that didn’t match the query. For high-stakes research synthesis, paste.
