How do I run customer interviews with AI?
AI helps most at the bookends of customer research: designing non-leading questions before interviews and synthesizing transcripts after. Use it to generate neutral, open-ended questions that uncover real problems instead of confirming your assumptions, then to cluster transcripts into themes, pain quotes, and jobs-to-be-done. The interviews themselves still work best human-to-human, where you can probe and read nuance. Let AI prep and analyze; you do the listening.
Good interviews start with good questions. Ask AI to generate 15 to 20 open-ended, non-leading questions about a specific problem for a specific customer. The discipline is avoiding questions that telegraph the answer ('Wouldn't it be great if...'); instead aim for past-behavior questions ('Tell me about the last time you dealt with X'). Frameworks like The Mom Test focus on what people actually did, not what they say they'd do, and AI can draft questions in that style if you ask for them.
After interviews, AI shines at synthesis. Paste your transcripts and ask it to extract recurring themes, pull direct pain quotes, and write jobs-to-be-done statements ('When I [situation], I want to [motivation], so I can [outcome]'). Combine multiple sources, interviews plus surveys, reviews, and support tickets, for a fuller picture and to confirm that a pain you heard in one call shows up elsewhere. You can also use AI insights to design a quick unmoderated test (Maze, Userbrain) to validate a specific idea with more people.
Keep humans central to the conversation itself. AI can technically run scripted chat surveys, but live interviews let you follow surprising answers and sense hesitation, which is where the real insight hides. Watch for the model over-summarizing and flattening contradictions; read the raw quotes for anything important, and treat its themes as a starting hypothesis to confirm, not a finished conclusion.
Prompts to try
Copy these into ChatGPT or Claude to go deeper.
Generate 20 non-leading customer interview questions to uncover [problem] for my [target customer].
Synthesize these interview transcripts [paste] into themes, pain quotes, and JTBD statements.
Design a 5-question Maze/Userbrain test using AI insights to validate [feature/idea].
Build an AI customer research workflow combining interviews, surveys, reviews, and support tickets.