Product/UX how-to

How to do customer interviews with AI.

Customer interviews are one of the highest-leverage product and marketing activities — and one of the easiest to do badly. Claude accelerates the prep and synthesis phases. The actual conversation stays human. Here is the workflow.

The premise

Why customer interviews matter

Direct customer interviews produce insights nothing else does. Surveys give you what people say they think. Interviews give you what they actually believe, struggle with, and want.

Most teams do too few interviews because the prep and synthesis labor is heavy. AI compresses both phases, which makes weekly customer interviews feasible.

The discussion guide prompt

Phase 1

I am preparing a customer interview for our [PRODUCT/PROJECT].

Who I am interviewing: [PERSONA / ROLE]
What I am trying to learn: [SPECIFIC RESEARCH QUESTION]
My current hypothesis: [WHAT I THINK]
What I already know from prior conversations: [PASTE]
Duration: [30 / 45 / 60 minutes]

Generate a discussion guide:
1. Opening (5 min): rapport + context
2. Core questions (~70% of time): 6-8 open-ended questions designed to learn what I do not know
3. Specific probes: for each core question, 2-3 follow-up probes if needed
4. Closing (5 min): "what did I forget to ask"

Question design rules:
- Open-ended (NOT yes/no)
- Past behavior, not hypothetical (not "would you" but "have you")
- Specific, not abstract
- Avoid leading questions
- The 6-8 questions should test specific assumptions, not gather generic feedback
The synthesis prompt

Phase 3 (after the interview)

Here is the interview transcript: [PASTE]

My hypothesis going in: [PRIOR]
What I was trying to learn: [RESEARCH QUESTION]

Synthesize:
1. Top 3-5 things I learned (with direct quotes as evidence)
2. What confirmed my hypothesis
3. What CHALLENGED my hypothesis (this is the most valuable section)
4. Specific things the customer said in their own words that I should preserve verbatim
5. New questions this raised
6. The single most important takeaway for the team

Be specific. "They liked the feature" is not useful synthesis. "When asked about [feature], they said [quote] and the context suggests [interpretation]" is useful.
Related

Related how-tos

Want customer research workflows built?
Implementation includes research workflow design.
See Implementation → Book the AI Audit