# The User Interview Synthesis Prompt That Won't Invent Quotes > Copy an AI prompt that turns a round of user interviews into themes, pains, quotes per participant and contradictions, each quote traced to a timestamp. - URL: https://www.getvoibe.com/resources/user-interview-synthesis-prompt/ - Author: Ayush - Published: 2026-11-07 ## User Interview Synthesis Prompt: What It Produces By interview eight, synthesis starts to drift. You remember the participant who was funny and quotable. You forget the three quieter ones who said the opposite. Ask a chatbot to summarize the round and it can smooth those disagreements into one tidy theme, with a paraphrase dressed up as a quote. The prompt below is built for research synthesis, not meeting summaries. It suits product managers, UX researchers and founders who have just finished a round of interviews and owe the team a readout. It returns five parts: 1. **Themes**, each with the participant codes (P1, P2 and so on) behind it and a count, such as 4 of 6. 2. **Pains and workarounds**, in the participants' own words. 3. **Quotes per participant**, verbatim, each with a timestamp. 4. **Contradictions** between participants, or within one participant's own answers. 5. **Open questions** that the round left unanswered, to test in the next one. The paste version works in ChatGPT, Claude or Gemini with a few pasted summaries. The MCP version is for Voibe Work, the meeting-notes plan we build, which takes notes with no bot in the call. In the MCP version, Claude reads the interviews itself through Voibe Meeting MCP, a Voibe Work feature that lets AI apps read your meetings. MCP, short for Model Context Protocol, is the standard AI apps use to connect to other software. ### Key takeaways | Question | Answer | | --- | --- | | **What does the prompt produce?** | Themes with counts, pains, verbatim quotes per participant, contradictions and open questions | | **What do I paste in?** | The research question and one summary or transcript per participant, labelled P1, P2 and so on | | **Paste version or MCP version?** | Paste a few interviews into any assistant, or connect Voibe Meeting MCP so Claude reads the whole round itself | | **How does it avoid invented quotes?** | It may quote only words in the notes, each with a timestamp, and writes [CHECK] where unsure | | **What do the counts mean?** | A count like 3 of 4 shows a theme exists in the round, not how common it is in your market | | **What does Voibe Work cost?** | $19 a month, $149 a year or $399 once, after a 7-day trial with no credit card | > **Key takeaway:** The prompt turns a round of user interview transcripts or summaries into five parts: themes with the participants behind each, pains and workarounds, verbatim quotes per participant, contradictions, and open questions. It quotes only words that appear in the notes, each with a timestamp, and writes a [CHECK] flag where it is unsure, so you know what to verify. ## The User Interview Synthesis Prompt: The Paste Version The paste version works in any AI assistant. Label each interview with a participant code before you paste it, so the output never carries a real name. The prompt asks for a [CHECK] marker wherever the assistant is unsure. A [CHECK] is a flag in the output that tells you what to verify against the transcript. ``` You are helping me synthesize a round of user interviews. Research question: [what we wanted to learn, e.g. why new users stop before sending their first invoice] Participants: [number] interviews, labelled P1 to P[n]. Each starts with its code and date. Audience: [who reads this, e.g. the product team at Thursday's readout] Length: under 700 words, plus the tables. Write these sections, in order: 1. Themes: a table with theme, participants who raised it (codes), count (e.g. 4 of 6), and the clearest quote with its code and timestamp. 2. Pains and workarounds: one line each, in the participants' words, with codes. 3. Quotes per participant: two or three verbatim quotes each, with timestamps. 4. Contradictions: where participants disagree, or one participant says two different things. Keep both sides. 5. Open questions: what this round did not answer, to test in the next round. Rules: - Quote only words that appear in the notes, exactly as written, with the participant code and timestamp. Never paraphrase inside quotation marks. - Count a participant toward a theme only if they said it. Do not round up or infer. - Do not merge opposing views into one theme. Put them under Contradictions. - Use participant codes only. Leave out names, employers, contact details and anything personal. - Do not recommend features or solutions unless I ask. - If something is unclear, write [CHECK: what is unclear] instead of guessing. Interviews: [paste P1 summary or transcript] [paste P2 summary or transcript] ``` ### Summaries or full transcripts? Full transcripts give the assistant every word, but a whole round may be too long for one chat. Summaries are shorter, and they still work if each keeps three to five verbatim quotes with timestamps. Our roundup of [AI note takers for user interviews](https://www.getvoibe.com/resources/best-ai-note-taker-for-user-interviews/) includes a seven-section summary template built for exactly this. It keeps participant context, pains, quotes with timestamps and surprises for every interview. ### Before you paste Whatever you paste into ChatGPT, Claude or Gemini goes to OpenAI, Anthropic or Google under that company's terms. If your consent form doesn't mention an AI assistant, don't paste the interviews into one. ## The MCP Version: Claude Reads the Whole Round Itself A round of eight interviews means eight summaries to copy. The MCP version skips that step, because Claude reads the interviews from Voibe Work through Voibe Meeting MCP. Meeting MCP comes with Voibe Work and the 7-day trial. Claude Desktop, the Claude app for Mac and Windows, connects in one click: 1. In Voibe Work, open the Meetings page and click MCP. 2. Check that the switch letting AI apps use your meetings is on. 3. On the Claude tab, click Add to Claude. Claude opens and shows Voibe Meeting MCP. 4. Click Install in Claude, and the connection is ready. ### The MCP version for one round Name the round by its dates, so Claude picks the right meetings. In Voibe Work, you rename a speaker by clicking their name in the transcript. If you renamed the speakers to participant codes, say so in the prompt: ``` Read my user interviews in Voibe from [first date] to [last date]. The participants are renamed P1 to P[n] in each transcript. Start by listing the meetings you read, with their dates, so I can confirm the set. Research question: [what we wanted to learn] Audience: [who reads this] Length: under 700 words, plus the tables. Write five sections: themes (a table with theme, participant codes, count such as 4 of 8, and the clearest quote with code and timestamp); pains and workarounds; two or three verbatim quotes per participant with timestamps; contradictions, keeping both sides; open questions for the next round. Quote only words in the transcripts, with code and timestamp. Count a participant toward a theme only if they said it. Use codes, never names. Do not recommend solutions. Write [CHECK: what is unclear] where you are unsure. ``` ### The MCP version for comparing two rounds Cross-meeting work is where the MCP version earns its place. Claude can compare this round with the last one without you pasting every summary from both rounds: ``` Read my user interviews in Voibe from [dates of round 1] and from [dates of round 2]. Participant codes restart each round, so call them R1-P1, R2-P1 and so on. Start by listing the meetings you read, with their dates and round, so I can confirm the set. For each theme, say whether it appeared in round 1, round 2 or both, with participant codes and counts for each round. List the themes that faded and the ones that appeared for the first time. Quote only from the transcripts, with timestamps, and write [CHECK] where you are unsure. ``` Both prompts ask Claude to work only from the transcripts. The list of meetings it read comes first, so you can spot a stray sales call before you trust the themes. ## A Worked Example: Four Onboarding Interviews, Synthesized The example is invented. A small team building invoicing software interviewed four new users who signed up but never sent an invoice. The research question was why. These excerpts come from the four Voibe Work transcripts, with the speakers renamed P1 to P4: ``` P1, 3 Nov [00:06:12] P1: I got stuck at the tax settings. I didn't know which rate to pick, so I closed it. [00:14:40] P1: I'd rather send a PDF from Word than get the tax wrong. P2, 4 Nov [00:04:55] P2: Setup was fine. I just didn't have a client ready to invoice yet. [00:09:30] P2: The tax screen asked for things my accountant normally handles. P3, 5 Nov [00:07:18] P3: I wanted to import my clients from a spreadsheet and couldn't find how. [00:11:02] P3: The tax part was easy. It filled itself in. P4, 6 Nov [00:05:40] P4: Which tax rate? I had no idea. I emailed my accountant and forgot to come back. [00:12:15] P4: If I could import my client list, I'd have sent one that day. ``` The paste version returned this synthesis, shown in full: ``` Themes | Theme | Participants | Count | Clearest quote | | Unsure how to set up tax | P1, P4 (P2: CHECK) | 2 of 4 | "I didn't know which rate to pick, so I closed it." (P1, 00:06:12) | | Wanted to import clients | P3, P4 | 2 of 4 | "If I could import my client list, I'd have sent one that day." (P4, 00:12:15) | | No client ready to invoice | P2 | 1 of 4 | "I just didn't have a client ready to invoice yet." (P2, 00:04:55) | Pains and workarounds - Choosing a tax rate: P1 stopped and closed it; P4 emailed an accountant and forgot to come back. - Workaround: P1 would rather send a PDF from Word than get the tax wrong. - Importing clients: P3 couldn't find how to import from a spreadsheet. Quotes per participant P1 00:06:12 I got stuck at the tax settings. 00:14:40 I'd rather send a PDF from Word than get the tax wrong. P2 00:04:55 Setup was fine. 00:09:30 The tax screen asked for things my accountant normally handles. P3 00:07:18 I wanted to import my clients from a spreadsheet and couldn't find how. 00:11:02 The tax part was easy. It filled itself in. P4 00:05:40 Which tax rate? I had no idea. 00:12:15 If I could import my client list, I'd have sent one that day. Contradictions - Tax setup: P1 and P4 didn't know which rate to pick; P3 said it filled itself in. - P2 said setup was fine, and also that the tax screen asked for things an accountant normally handles. [CHECK: whether P2 found tax setup hard] Open questions - Why did the tax settings fill in for P3 but not for P1 and P4? - Would P2 have sent an invoice with a client ready? - Is client import missing, or hard to find? ``` ### What the prompt did with the four interviews - P2 said setup was fine but also found the tax screen an accountant's job. The prompt left P2 out of the tax count and flagged it. - P3 found the tax part easy, so that answer went under Contradictions instead of vanishing from the theme. - Every quote is a line from the excerpts, with its timestamp. - The output recommends nothing, because the prompt asked for findings, not solutions. - The counts read 2 of 4, not 50%, since four interviews show where to dig rather than how common a problem is. ## How to Adapt the Synthesis Prompt for Other Kinds of Research The rules block is the list under Rules in the paste version. Keep it as it is. Change the five sections and the audience to fit the study, and the same grounding carries over. ### Swap the sections for the study type One term in the table: jobs-to-be-done is a research approach that asks what task a person hires a product to do. The discovery row keeps that task, and its trigger, in the participant's words. | Study type | Replace the five sections with | | --- | --- | | Usability test | Tasks attempted, where each participant got stuck (with timestamp), success or failure per task as stated, and quotes at each sticking point | | Discovery or jobs-to-be-done interviews | The job the participant was trying to get done, the trigger that started the search, the current workaround, and what a better outcome looks like | | Churn or win-loss interviews | Why they left or chose you, the alternatives they named, and the moment they decided | | Concept test | Each participant's reaction to each concept, what they understood it to do, and their questions | ### Change the audience and length - **Executive readout:** set the audience to leadership and the length to under 200 words. Ask for three themes, one quote each, and the biggest open question. - **Affinity mapping:** ask for one observation per line with its participant code, ready to paste onto sticky notes in a whiteboard tool. - **Design team:** ask for the pains and quotes only, grouped by the screen or step where they happened. ### Customer calls instead of research interviews Sales and support calls carry the same signal with less structure. For those, our [feature requests prompt](https://www.getvoibe.com/resources/feature-requests-from-customer-calls-prompt/) pulls requests, frequency and severity from a month of customer calls. ## What to Check Before You Share the Synthesis Synthesis shapes what a team builds next, so a wrong quote or an inflated count costs more than a typo. Check these before the readout. ### Quotes and counts - **Every quote:** find it in the transcript at its timestamp and confirm the words match. Delete any quote you can't find. - **Every count:** open the interviews listed for a theme and confirm each participant said it. - **Contradictions:** make sure the assistant didn't drop a dissenting participant to make a theme look stronger. - **[CHECK] items:** resolve each one against the transcript, then delete the marker. ### Sample size Five or six interviews can show that a problem exists and how people describe it. They can't tell you how common it is across your market. Report counts as 3 of 4, not as 75%. ### Privacy and consent Use participant codes in everything you share. Leave out employers, job titles that identify someone, and personal details. Rename the speakers to their codes in Voibe Work before you run either version, so the transcripts carry codes too. Your consent form should match what you do with the recordings, so name any AI assistant you send transcripts to. Voibe reminds you to tell everyone you're taking notes when recording starts. Following the recording-consent rules is your responsibility, and our guide to [AI notetaker recording consent laws](https://www.getvoibe.com/resources/ai-notetaker-recording-consent-laws/) covers who has to agree. ## Run the Prompt on User Interviews You Record With Voibe Work A bot in the call can change how a participant talks. Voibe Work captures interviews from your Mac or Windows PC with no bot in the call. Zoom, Microsoft Teams, Google Meet, Webex, Slack huddles and other meeting apps all work. For an in-person usability session, Voibe Work records through your computer's microphone. Our guide to [recording an in-person meeting on your laptop](https://www.getvoibe.com/resources/how-to-record-an-in-person-meeting-on-your-laptop/) covers the setup. ### Set up each interview for synthesis 1. **Create your own template** for user interviews in Voibe Work, such as the seven-section summary template from our note taker roundup. 2. **Rename the speakers** to participant codes. The transcript labels each speaker with a time on every line; click a name to rename it, and the summary updates. 3. **Copy or export** each summary in one click, or as a PDF, for the paste version. 4. **Connect Meeting MCP** for the MCP version, so Claude reads the round itself. One interview can run up to 4 hours. Transcripts and summaries arrive 10 to 20 times faster than before, by Voibe's own measurements (see the [October 2026 meeting update](https://www.getvoibe.com/resources/voibe-free-plan-meeting-notes-windows/)). For a quick question about one interview, chat with your meetings inside Voibe Work. The chat runs on Voibe's zero-retention cloud and answers from one meeting, and nothing you ask is stored. ### Where the interviews go - Meeting audio is processed by open-source models on zero-retention cloud providers and deleted as soon as processing completes. - The transcript and summary are saved on your computer, not on Voibe's servers, and never used to train AI. - Processing runs in the cloud, so meeting notes need an internet connection. ### What an AI app reads through Meeting MCP The Meeting MCP panel opens when you click MCP on the Meetings page. It states the terms in its own words: > “Apps you connect can read your meeting titles, summaries and transcripts. Your dictation is never shared. What an app reads goes to the company behind its AI, for example Anthropic for Claude. None of it goes through Voibe's servers.” The paste version works on the same principle. What you paste into an assistant goes to that AI company under its terms. The panel's switch stops every AI app at once, so you can keep it off between synthesis sessions. ### What Voibe Work costs Voibe Work costs $19 a month, $149 a year or $399 once, with Mac and Windows on one seat. Twelve monthly payments come to $228, so the annual plan saves $79, or 34.6%. Every new account starts with a 7-day trial of every Voibe Work feature. There's no credit card and no automatic charge, and [Meeting MCP](https://www.getvoibe.com/resources/voibe-meeting-mcp/) is part of the trial. New to Voibe Work? Our [first-week setup guide](https://www.getvoibe.com/resources/getting-started-with-voibe-work/) walks you through it. ## User Interview Synthesis Prompt FAQ ### Using the prompt **Can ChatGPT synthesize user interviews?** Yes. Paste the research question and one labelled summary or transcript per participant under the paste version. ChatGPT, Claude and Gemini all return the same five sections: themes, pains, quotes, contradictions and open questions. **Should I paste transcripts or summaries?** Paste transcripts if the round fits in one chat. Otherwise paste summaries that keep three to five verbatim quotes with timestamps each. **How do I stop AI from inventing interview quotes?** Tell it to quote only words that appear in the notes, with a participant code and timestamp. Then check every quote against the transcript before you share it. ### Research method **How many user interviews do I need before synthesis?** Synthesis works on any round, but five or six interviews show that a problem exists, not how common it is. Report counts as 3 of 5 rather than as percentages. **What is a contradiction in interview synthesis?** A contradiction is a place where participants disagree, or where one participant says two different things. Keeping both sides stops a theme from looking stronger than it is. ### Voibe Work, MCP and privacy **Can Claude read all my user interviews at once?** Yes, with Voibe Meeting MCP connected to Claude Desktop. Meeting MCP comes with Voibe Work and the 7-day trial, and what Claude reads goes to Anthropic under its terms. **Can I paste interviews into an AI assistant?** Only if your consent form covers it, because what you paste goes to that AI company under its terms. Use participant codes and leave out personal details first. **Where does Voibe Work keep interview notes?** Voibe Work saves the transcript and summary on your computer, not on Voibe's servers. Meeting audio is deleted as soon as processing completes, and nothing is used to train AI. ## Synthesis You Can Defend in the Readout Good synthesis shows where participants agreed, where they didn't, and the exact words behind both. Label every interview with a code and keep the rules block intact. Then check every quote and count before the readout. To choose the note taker itself, see our roundup of [AI note takers for user interviews](https://www.getvoibe.com/resources/best-ai-note-taker-for-user-interviews/). The [AI prompts for meeting notes](https://www.getvoibe.com/resources/ai-prompts-for-meeting-notes/) library has more prompts that turn meetings into finished work. ## Frequently asked questions ### Can ChatGPT synthesize user interviews? Yes. Paste the research question and one labelled summary or transcript per participant under the paste version. ChatGPT, Claude and Gemini all return the same five sections: themes, pains, quotes, contradictions and open questions. ### Should I paste transcripts or summaries? Paste transcripts if the round fits in one chat. Otherwise paste summaries that keep three to five verbatim quotes with timestamps each. ### How do I stop AI from inventing interview quotes? Tell it to quote only words that appear in the notes, with a participant code and timestamp. Then check every quote against the transcript before you share it. ### How many user interviews do I need before synthesis? Synthesis works on any round, but five or six interviews show that a problem exists, not how common it is. Report counts as 3 of 5 rather than as percentages. ### What is a contradiction in interview synthesis? A contradiction is a place where participants disagree, or where one participant says two different things. Keeping both sides stops a theme from looking stronger than it is. ### Can Claude read all my user interviews at once? Yes, with Voibe Meeting MCP connected to Claude Desktop. Meeting MCP comes with Voibe Work and the 7-day trial, and what Claude reads goes to Anthropic under its terms. ### Can I paste interviews into an AI assistant? Only if your consent form covers it, because what you paste goes to that AI company under its terms. Use participant codes and leave out personal details first. ### Where does Voibe Work keep interview notes? Voibe Work saves the transcript and summary on your computer, not on Voibe's servers. Meeting audio is deleted as soon as processing completes, and nothing is used to train AI. ## Related - [Best AI Note Takers for User Interviews](https://www.getvoibe.com/resources/best-ai-note-taker-for-user-interviews/): Nine note takers for research, with or without a bot. - [Feature Requests From Customer Calls Prompt](https://www.getvoibe.com/resources/feature-requests-from-customer-calls-prompt/): A month of customer calls, turned into a request table. - [AI Prompts for Meeting Notes](https://www.getvoibe.com/resources/ai-prompts-for-meeting-notes/): Copy-paste prompts that turn meetings into finished work. - [Getting Started With Voibe Work](https://www.getvoibe.com/resources/getting-started-with-voibe-work/): Your first week of meeting notes with no bot in the call. - [Voibe Meeting MCP](https://www.getvoibe.com/resources/voibe-meeting-mcp/): Let Claude and ChatGPT read your meetings. - [AI Notetaker Recording Consent Laws](https://www.getvoibe.com/resources/ai-notetaker-recording-consent-laws/): Who has to agree before you record.