Prompt Engineering
AI prompts for customer discovery interviews: questions, synthesis, and next actions
A practical prompt workflow for customer interviews: prep, question design, synthesis, and decision-ready summaries—without letting AI hallucinate research.

Foundation
What AI can (and can't) do in discovery
Discovery fails when teams treat interviews like marketing. The goal is not to hear "this is amazing". The goal is to uncover concrete details: what people do today, what they tried, what it cost them, what broke, and what finally made them change.
AI is great at structure (turn a messy goal into a plan), language (turn vague questions into neutral ones), and synthesis (cluster repeated themes). AI is bad at inventing reality. If you ask it to guess what customers want, it will produce plausible nonsense.
So your prompts should constrain the model to inputs you provide, require quote-level evidence where possible, and ask it to label assumptions explicitly.
Prep
Interview plan prompt: define the decision and audience
Start by naming the decision the interviews must support. Examples: which customer segment to target first, which workflow step is most painful, or whether a paid solution is justified.
Use the prompt below to produce an interview plan you can actually run: who to recruit, what to learn, how many interviews to schedule, and what a "good interview" looks like.
Copy-ready prompt
Act as a customer research lead. Build a customer discovery interview plan for this product idea: [IDEA]. Context: - Target audience (if known): [AUDIENCE] - Decision this research must support: [DECISION] - Constraints: [TIME/BUDGET/TEAM] Output: 1) Research goal (1 sentence) 2) Ideal participant criteria + who NOT to recruit 3) Recruiting channels + screener questions 4) Interview guide outline (sections, not full questions yet) 5) Sample size recommendation and why 6) Risks and bias traps to avoid 7) What evidence would change our mind Rules: Do not invent findings. Label assumptions vs confirmed inputs. Keep questions neutral and behavior-focused.
Question design
Question bank prompt: get neutral, behavior-first questions
A useful question bank covers past behavior, current workflow, constraints, workarounds, and decision triggers. Avoid hypotheticals like "Would you use this?" because they invite politeness, not truth.
Ask AI for multiple options per question so you can choose the most neutral phrasing. Then run a bias check to remove leading language.
Copy-ready prompt
Act as a senior qualitative researcher. Create a customer discovery interview question bank for this audience and problem:
- Audience: [AUDIENCE]
- Problem area: [PROBLEM]
- Proposed solution (for context only): [SOLUTION]
Requirements:
- 5 warm-up questions (role, context, tools)
- 10 behavior questions ("what did you do last time", steps, frequency)
- 8 pain questions (where it breaks, cost of failure, emotional/friction costs)
- 6 workaround questions (tools, hacks, substitutions)
- 6 decision questions (budget owner, approval, switching triggers)
- 4 comparison questions (alternatives and why they win/lose)
- 4 wrap-up questions (priorities, referrals, artifacts to share)
For each question, provide 2 neutral phrasings and 1 follow-up probe. Avoid leading questions and avoid asking for feature requests. Keep everything grounded in real past behavior.During interviews
A simple rule for AI: summarize after, not during
If you use AI during the call, keep it to logistics (agenda reminders, time checks) and do not let it steer the conversation. The highest risk is the model nudging you toward your preferred story.
Instead, record the call (with consent) or take raw notes. After the interview, paste the transcript or notes into the synthesis prompts below.
If you only have bullet notes, tell the model that the notes are incomplete and that it must not infer missing details.
Synthesis
Transcript synthesis prompt: themes, tensions, and proof
Good synthesis separates what was said from what it might mean. Ask for themes, tensions, and triggers—but require evidence. A strong output includes quotes (or exact note excerpts) and tags each theme with confidence.
Use one prompt per interview first, then a cross-interview synthesis. This prevents the model from blending speakers into a fictional "average customer".
Copy-ready prompt
Act as a research synthesis analyst. Using ONLY the transcript/notes below, create a structured interview summary. Input (transcript or notes): [PASTE TRANSCRIPT OR NOTES] Output sections: 1) Participant context (role, company type, constraints) 2) Current workflow (steps as described) 3) Pain points (with direct quotes/excerpts) 4) Workarounds and tools used 5) Decision triggers and blockers 6) Non-goals and boundaries (what they will NOT do/pay for) 7) Top insights (3-7), each with: claim + evidence quote + confidence (High/Med/Low) 8) Open questions to validate next Rules: Do not add facts that aren't in the input. If evidence is missing, write "Unknown" and ask a follow-up question.
Cross-interview
Pattern-finding prompt: what repeats across people
Once you have 5+ interviews summarized in a consistent structure, AI becomes useful for clustering. The key is to keep traceability: every theme should link back to the interview(s) and supporting excerpt(s).
Copy-ready prompt
Act as a qualitative research lead. Synthesize the interview summaries below into themes and a decision-ready insight memo. Input (multiple interview summaries): [PASTE SUMMARIES] Output: 1) Theme clusters (5-10): name + description + who it applies to 2) Evidence table: for each theme, list interview IDs and supporting excerpts 3) Tensions/tradeoffs: what customers balance (time vs cost, control vs convenience, etc.) 4) Segment notes: which themes vary by role, company size, or maturity 5) Risks if we build the wrong thing 6) 3 recommended next-step experiments (lean tests), each with: hypothesis, method, success metric, and timeline Rules: No invented quotes. If evidence is weak, say so. Keep outputs concise and executive-readable.
Next actions
Experiment design prompt: turn insights into validation tests
Discovery is only useful if it changes what you do next. Use AI to propose small tests: landing page tests, manual concierge pilots, clickable prototypes, or pricing conversations—based on the actual pains you heard.
The prompt below converts themes into a two-week validation plan with specific assets and measurable signals.
Copy-ready prompt
Act as a product validation coach. Based on these customer discovery themes, propose a 14-day validation plan. Themes + evidence (paste): [PASTE THEMES] Output: 1) Top 3 hypotheses to validate (with why they matter) 2) Smallest test for each hypothesis 3) Assets needed (copy, prototype, script, checklist) 4) Outreach plan (who to contact and what to say) 5) Success metrics and decision thresholds 6) Risks, confounders, and how to reduce bias Rules: Do not assume we can code anything. Prefer tests that can be run with docs, spreadsheets, and calls.
Internal links
Where this fits in the NEOA workflow
If you are building an AI-assisted business workflow, customer discovery should feed your briefs, your prompts, and your skill design. The strongest systems connect research → prompts → repeatable workflows → review.
Sources
Further reading
If you want deeper guidance on interviewing without bias and turning conversations into decisions, these are good starting points.
Related agent skill
Business Idea Validator Agent
Analyze a business idea across audience, problem intensity, monetization, competition, MVP scope, and launch risks.
Free prompt pack
Get the prompt pack behind practical AI workflows.
Download 50 prompts for SEO, content, research, and business automation, then use them with this guide to make the workflow repeatable.
Free download
Get the prompt pack.
Choose your main interest and unlock the Markdown download.
Free during NEOA beta. You can download after submitting the form.
FAQ
Common questions
Can AI replace customer discovery interviews?
No. AI can help you plan interviews and synthesize what you collected, but it cannot generate real customer evidence. Use it to structure your process and summarize transcripts you provide, not to invent what customers think.
How many interviews do I need before patterns matter?
It depends on how narrow the segment is, but you can often hear repeated themes after 5–8 interviews in a focused audience. Use interviews to find hypotheses, then validate with small tests and measurable signals.
What should I paste into the synthesis prompt?
Paste a transcript or your raw notes. If notes are incomplete, tell the model they are incomplete and require it to label unknowns rather than filling gaps.
What's the biggest prompt mistake in discovery?
Asking for conclusions without providing evidence. Your prompts should force traceability: themes must point back to specific excerpts, and anything uncertain should be labeled as an assumption.
Final recommendation
Make the workflow repeatable before you scale it.
Use AI to improve the process, not to fabricate the outcome. If your prompts force the model to cite what was actually said, your discovery becomes easier to review, share, and act on—without turning into confident fiction.