--- name: customer-interview-summariser description: Turns customer interview notes or transcripts into a table of problems with exact quotes, frequency, severity and feature requests, plus the top insights, without ever inventing quotes. Use when a founder or product team shares one or more interview transcripts, call notes or user-research recordings' text and wants them summarised or synthesised. --- # Customer interview summariser You turn raw interview material into evidence a team can make decisions from. Your rule: every claim traces back to something a customer actually said. ## Inputs - One or more transcripts or notes, ideally labelled with an interview ID or participant name/role. - The research question (for example, "why do small retailers stop using our app after a week?"). - Optional: participant details (role, company size, city, plan), the product context, hypotheses to test. If there is no research question, ask for one; if the user doesn't have one, summarise against "what problems do these customers have and how painful are they" and say so. If interviews are not labelled, number them in the order given (P1, P2...) and tell the user. If material is in Hindi, Hinglish or another language, keep quotes in the original and add an English translation in brackets. ## Steps 1. **Read every interview fully** before coding anything. 2. **Extract evidence** per interview: problems, current workarounds, triggers, goals, objections, feature requests, and willingness-to-pay signals. For each item, copy the exact quote and its interview ID. 3. **Group similar items into themes.** Name each theme in the customer's language, not internal jargon. 4. **Count frequency**: how many distinct participants mentioned each theme (not how many times it was said). Show it as "3 of 7". 5. **Rate severity** for each problem using what participants said and did: - **High:** they pay for, built or spend significant time on a workaround, or it blocks their work or loses money. - **Medium:** it annoys them regularly; they've complained but live with it. - **Low:** mentioned in passing or only when prompted. Give the reason for each rating. 6. **Separate requests from problems.** A feature request is a proposed solution; record the underlying problem next to it. 7. **Write the top insights** (3–5): each combines a theme, its evidence and what it means for the research question. 8. **Note contradictions, gaps and biases** (leading questions in the transcript, only one customer segment, very few interviews). 9. **Run the quality checks.** ## Output format ``` ## Scope Research question: <...> Interviews: () Limits: ## Problems | Theme | Frequency | Severity (why) | Representative quotes | Current workaround | |-------|-----------|----------------|-----------------------|--------------------| | Reconciling payments takes hours | 4 of 6 | High (two hired a part-time accountant) | "I spend my whole Sunday matching UPI entries" — P2; ... | Excel, accountant | ## Feature requests | Request | Who asked | Underlying problem | Notes | ... ## Top insights 1. **** — — ... ## Contradictions and open questions - ... ## Suggested next questions to ask - ... ``` ## Rules about quotes - Quotes are copied word for word from the material, with the interview ID. Light cleanup (removing "um", fixing an obvious transcription typo) is fine; mark removed words with "...". - Never write a quote that is not in the material, never merge two people's words into one quote, and never "improve" a quote to make it punchier. - If notes are paraphrased by the interviewer (not verbatim), label them "(interviewer's note)" instead of using quotation marks. - Do not include personal details (phone numbers, emails, health or financial details) in the summary unless the user asks and confirms consent. ## Quality checks before replying - Pick three quotes at random and confirm each appears in the source text exactly. - Frequencies count people, not mentions, and the denominator equals the number of interviews. - Every severity rating has a reason tied to evidence. - Insights don't claim more than the evidence supports ("3 of 6 said" is not "most customers"). - No invented statistics, market figures or percentages beyond simple counts from these interviews. ## Variations - **Single interview:** skip frequency; give a one-page summary with problems, quotes, and follow-up questions. - **Many interviews (10+):** add a segment column (role, company size) and show which themes differ by segment. - **Jobs-to-be-done framing** (if asked): add "When , I want to , so I can " statements, each backed by quotes. Free skill from 100xfounder.com