Synthesize a stack of user interviews into themes

Ranked themes across multiple interviews, each backed by verbatim quotes with participant attribution — traceable synthesis, not a summary you can't trust.

Works in any frontier model — ChatGPT, Claude, Gemini

First

Get clean transcripts and label each one (P1, P2, …) so quotes stay attributable; batch in groups of about five if you have many. Since 2026-08-04, a paste over 10,000 characters into ChatGPT becomes a file attachment instead of message text, and the chip takes its name from the first line of what you pasted, so a P1 to P8 batch shows up looking like it holds only P1. That naming is cosmetic, and we re-ran this prompt against the attachment on 2026-08-11 with quote fidelity holding on all eight quotes. Click "Show in text field" if you specifically want the transcripts inline.

The prompt

You are a user researcher synthesizing interviews. Below are <N> transcripts,
each labeled P1, P2, …

Task:
1. Identify the recurring themes across participants (not per-interview summaries).
2. Rank themes by how many distinct participants raised them.
3. For EACH theme, include:
   - a one-line description
   - the participants who raised it (e.g. P1, P4, P7)
   - 2–3 VERBATIM quotes, each tagged with the participant
4. Separately, list surprises — things only one person said that felt important.

Hard rule: every quote must be word-for-word from the transcripts. If you can't
find a real quote for a theme, say so instead of paraphrasing.

Transcripts:
"""
<paste labeled transcripts>
"""

Built-in guardrail: Requires every theme to carry word-for-word quotes with attribution, and to admit when no real quote exists rather than paraphrase — so hallucination is caught instantly.

Then

Spot-check three quotes against the source transcripts — if accurate, trust the pattern and take the top themes straight into your PRD's Problem & context as evidence; if any are paraphrased, re-run with a stronger warning.

Field-tested Aug 11, 2026. If it stops working, tell us.