Synthesize churn interviews into the real reasons people left
Ranked cancellation reasons backed by verbatim quotes and participant tags — with "it's too expensive" excuses separated from the value gaps hiding underneath them.
Works in any frontier model — ChatGPT, Claude, GeminiFirst
Label each transcript (P1, P2, …) and, if you have more than about five, run them in batches so quotes stay attributable.
The prompt
You are a researcher analyzing cancellation / churn interviews. Below are
transcripts labeled P1, P2, … from customers who left.
Task:
1. Identify why people actually churned. Rank the reasons by how many distinct
participants raised each.
2. For EACH reason include:
- a one-line description
- the participants who raised it (e.g. P2, P5, P9)
- 2–3 VERBATIM quotes, each tagged with the participant
3. PRICE vs VALUE — critical: separate reasons that are genuinely about price
(they'd pay elsewhere / budget was cut) from "too expensive" statements that
are really about not getting enough value (the product didn't do the job, so
any price felt too high). For each churned-on-price participant, judge which
it is and quote the line that tells you.
Hard rules:
- Every reason and every price-vs-value judgment must be supported by a
word-for-word quote from the transcripts, tagged to the participant.
- If a theme or judgment has no real supporting quote, do NOT paraphrase to
manufacture one — write "no direct quote — weak signal" and move on.
Transcripts:
"""
<paste churn interview transcripts, each labeled P1, P2, …>
"""
Built-in guardrail: Every reason and every price-vs-value call must carry a word-for-word attributed quote; themes without a real quote are labeled "weak signal" rather than propped up with paraphrase.
Then
Focus on the price-that's-really-value bucket — that's the churn you can actually fix with product, not discounting — and spot-check two quotes against the source before you take these reasons to your team.
Field-tested Jul 25, 2026. If it stops working, tell us.