ChatGPT Prompts for Customer Interviews (Scripts, Notes, Follow-Ups)

Most customer interviews go wrong in the same way: you describe your idea, the other person is polite, and you leave thinking you validated something. You didn't. You got encouragement. Rob Fitzpatrick's The Mom Test is named for that trap: even your mom will lie to protect your feelings if you ask whether your idea is good.

This guide is for solo founders who want ChatGPT to help with the structure of discovery interviews, not to invent insights. You'll get three copy-paste prompts: a non-leading 30-minute script grounded in past behavior, a synthesis pass that turns messy notes into themes, quotes, and risks, and a follow-up message that asks for a referral or a win-loss detail without pitching. Where we lean on research, there's a link. There are no invented conversion rates.

What ChatGPT is good at (and what it must not do)

ChatGPT has never spoken to your buyers. It cannot tell you whether a problem is real. Used carelessly, it will happily invent "insights" from thin air, or turn your interview guide into a product pitch dressed up as questions.

Where it is useful:

• Drafting neutral, open-ended questions so you don't lead people toward the answer you want (Nielsen Norman Group on leading questions).
• Organizing messy notes into themes, verbatim quotes, and open risks after the call.
• Writing short follow-ups that stay in learning mode: thank you, one clarifying question, a soft ask for a referral.

Y Combinator's public guidance on talking to users pushes the same discipline: talk about their life, ask about specific past events, and listen more than you talk (YC Startup Library: Talking to users). ChatGPT can help you prepare for that. It cannot replace the calls.

What to gather before you open ChatGPT

Ten minutes of prep is enough:

• Who you're talking to. One sentence: role, context, and why they might have the problem (e.g. "solo consultants who invoice clients monthly").
• The problem area, not your product. Name the job or friction in their world, not your feature list.
• What you already know. Two or three facts from prior chats, support tickets, or your own experience. Mark guesses as guesses.
• What would change your mind. One sentence: evidence that would make you pause or kill the idea.
• Logistics. Call length (aim for 25–30 minutes), whether you're recording with consent, and how you'll take notes.

Keep the product off the table until the end of the call, or leave it off entirely. Early pitches bias every answer that follows.

If you still need people to talk to, use warm intros, buyer communities, and short cold outreach that asks for a learning call — not a demo. See the first-10-customers and cold outreach guides for that. This piece assumes the calls are booked and you need better questions, notes, and follow-ups.

Prompt 1: Non-leading 30-minute discovery script

This prompt builds a guide you can actually run. It favors open stems ("Tell me about…", "Walk me through…") over yes/no and "would you…?" questions — the pattern Nielsen Norman Group recommends for user interviews (User Interviews 101).

You are helping me prepare a 30-minute customer discovery interview. Do not pitch my product. Do not invent customer quotes or statistics.

Who I'm talking to: [ROLE / CONTEXT]
Problem area (not my product): [THE JOB OR FRICTION IN THEIR WORLD]
What I already know (facts only): [2–3 FACTS]
What would change my mind: [ONE SENTENCE]
Call length: 30 minutes. I may record with consent.

Write a script I can run live:
1) 2-sentence opener that explains I'm learning about how they handle [PROBLEM AREA], not selling anything.
2) A consent line for notes/recording.
3) 2 warm-up questions that are easy and non-personal.
4) 6–8 core questions about SPECIFIC PAST behavior (last time, most recent week, a concrete example). Use open stems: "Tell me about…", "Walk me through…", "What happened next…". No "would you use…", no "do you struggle with…", no questions that name my solution.
5) For each core question, add 1–2 short probe options ("Can you say more about that?", "What did you try next?").
6) A 3-minute closing: thank them, ask if I missed anything important, and ONE soft ask: either a referral to someone else with the same problem, or permission for a short follow-up email. No product pitch.
7) A one-page note template: Context | Timeline of the story | Workarounds tried | Money/time cost (only if they volunteered it) | Verbatim quotes | Open questions / risks.

Rules: every question must be answerable from their past, not their imagination. Flag any line that still sounds leading and rewrite it.

What good output looks like: questions about the last time something happened, probes that dig into specifics, and a closing that does not sneak in a demo. If the script mentions your product name before the closing, delete that part and regenerate.

Prompt 2: Interview notes → themes, quotes, and risks

After three to five calls, paste your raw notes. Ask for patterns and contradictions, not a victory narrative.

I completed customer discovery interviews. Synthesize ONLY from the notes below. Do not invent quotes, numbers, or themes that aren't supported.

Product I'm exploring (for your context only — do not rewrite history around it): [ONE SENTENCE]
Interviewee profiles: [SHORT LIST]
Raw notes (paste all calls):
---
[PASTE NOTES]
---

Output:
1) Themes: 4–7 themes. For each: name, 1–2 sentence summary, which interviewees it showed up in, and 1–2 verbatim quotes (or "no clean quote").
2) Contradictions: where people disagreed, or where the same person said two incompatible things.
3) Workarounds and current alternatives people already use.
4) Risks to my idea: list only risks implied by the notes. Label each as EVIDENCE or INFERENCE.
5) Questions I still need answered before I build more.
6) A "do not claim yet" list: statements I might want to put on a landing page that these notes do NOT support.

Rules: if a theme only appears once, mark it as weak. Prefer under-claiming. Never invent a statistic.

Run this after every few calls, not once at the end. Early themes that don't repeat are often noise. Strong themes that keep showing up — with quotes — are what you build and write copy from.

Keep contradictory quotes side by side instead of averaging them away — averages often hide two buyer segments. And check your landing page against the "do not claim yet" list before you publish. If the page says something the notes don't support, rewrite the page, not the notes.

Prompt 3: Follow-up email and referral (or win-loss) ask

A good interview ends with permission to follow up. Use ChatGPT to draft a short note that stays in learning mode. For people who already evaluated or rejected something like yours, the same prompt works as a gentle win-loss ask.

Draft a short follow-up email after a customer discovery call. Keep it under 120 words. No hype, no product pitch, no fake urgency.

Context:
- Their name / how I'll greet them: [NAME OR "Hi"]
- What we talked about (facts from the call): [2–3 BULLETS]
- One clarifying question I still have: [QUESTION ABOUT THEIR PAST BEHAVIOR OR WORKAROUND]
- Soft ask (pick one): [REFERRAL to someone else with the same problem] OR [WIN-LOSS: why they chose / didn't choose an alternative they mentioned]
- My name and one-line who I am: [NAME + ROLE]

Write:
1) Subject line (neutral, specific to their situation — not "quick question" spam).
2) Email body: thank them, reflect one concrete detail so they know I listened, ask the clarifying question, then the soft ask. Make it easy to say no.
3) A 2-sentence LinkedIn/DM variant of the same message.

Rules: do not invent details I didn't provide. Do not attach a calendar link unless I ask. Do not say "excited" or "synergy."

Send it within a day while the conversation is fresh. If they introduce you to someone else, treat that person as a new interview — same non-leading script, not a warmer pitch.

Common traps

• Pitching in the first five minutes. You train people to evaluate your idea instead of describing their life.
• Asking "would you use this?" Polite people say yes. Ask what they did last time instead.
• Leading with your preferred answer. "Was that frustrating because of the export step?" puts the cause in their mouth. Ask what happened, then why it mattered (NN/g on leading questions).
• Letting ChatGPT invent themes. If a quote isn't in your notes, it isn't evidence.
• Interviewing only friends. They're kinder and less representative. Use them to practice the script, then talk to strangers in the segment.
• Skipping the note template. Memory smooths contradictions. Write them down the same day.
• One call = validated. Themes that appear once are anecdotes. Look for repeats across people who don't know each other.

The short version

1. Prep: who, problem area, what would change your mind — product stays offstage.
2. Use Prompt 1 for a 30-minute, past-behavior script with probes and a note template.
3. Run the calls. Listen more than you talk. Capture verbatim lines.
4. Use Prompt 2 after every few interviews to extract themes, contradictions, and unsupported claims.
5. Use Prompt 3 for a short follow-up that asks for a referral or a win-loss detail without pitching.

ChatGPT can draft the questions and tidy the notes in minutes. The truth still only shows up when real people talk about what they already do.