How many user research interviews do I need?

For generative discovery, patterns usually start converging around 15 to 30 interviews with the right people. Because every respondent produces a full transcript, you are reading for repeated behaviors and language, not statistical power.

Can an AI really run a discovery interview?

For structured generative interviews, yes. The interviewer is instructed to anchor on real events, ask for stories, and follow the frustration. It will not replace a deep expert interview, but it runs consistent, unbiased sessions at any hour in 7 languages.

How is this different from a survey?

The anchors are open and the follow-ups are unscripted, so it behaves like an interview rather than a questionnaire. You get walked-through stories and workarounds, not checkbox tallies, with full transcripts to quote from.

What do I do with the transcripts?

Read them for recurring behaviors, workarounds, and the words people use, then use the themes with quotes and the plain-language question box to test hunches like "who mentioned building their own workaround?"

Templates / User research interview

Research · Adaptive interview template

User research interview template

Good user research interviews avoid the two questions that ruin them: would you use this, and how much would you pay. Both invite people to imagine a hypothetical version of themselves who is more organized, more rational, and more interested than the real one. Generative discovery is about what people already do, not what they predict they will do.

This template is built around behavioral anchors and hands the moderation to an AI trained to do what a good researcher does: ask for the last time it happened, ask for the story rather than the summary, and stay quiet enough to let the respondent keep talking. When someone says "I usually just deal with it manually," the follow-up asks them to walk through the last time, step by step, because that is where the workaround and the real pain live.

Describe who you are looking for and the problem area in the introduction, then let the interview find the patterns.

When to use it

Early discovery, before you have a solution to test, to map the problem space
Understanding a workflow you plan to build for but do not personally live
Exploring how people currently solve a problem, tools and workarounds included
Scaling qualitative reach beyond the handful of live calls your team can run

Running it well

Ask about the past, not the future. "When did you last" produces evidence; "would you" produces fiction.
Resist pitching. The moment you describe your idea, the respondent starts reacting to it instead of telling you about their life.
Recruit for the behavior, not the demographic. You want people who have the problem, whatever their title or age.
Let the follow-up depth budget run higher here than on a satisfaction survey. Discovery rewards the extra probe, and the transcript is the deliverable.

The questions, and why they're shaped this way

These are the anchor questions. You can edit, reorder, or add to them; the AI interviewer handles the depth between them.

Open text

Walk me through the last time you dealt with this. What happened?

Anchoring to the last real instance beats asking what they usually do. Specific memories carry the details, the workarounds, and the frustration that general answers sand away.

Open text

What's the most frustrating part of doing this today?

Frustration marks where a solution would earn its place. Asked after the story, the answer is grounded in something concrete they just described rather than a generic complaint.

What the interviewer asks next

Follow-ups aren't scripted. The AI reads each answer and probes what a researcher would. Depending on what a respondent says, it might ask:

"You said you usually just handle it manually. Can you walk me through what that looked like the last time, step by step?"

"You mentioned it gets frustrating near a deadline. What specifically goes wrong when you're under time pressure?"

"You said you built a spreadsheet to cope. What does the spreadsheet do that the tools you tried could not?"

Live demo

Play the user research interview interview

This is the real interviewer, not a recording. Answer as yourself and watch the follow-ups adapt to what you write.

Prefer a full window? Open the demo.

Common questions

How many user research interviews do I need?: For generative discovery, patterns usually start converging around 15 to 30 interviews with the right people. Because every respondent produces a full transcript, you are reading for repeated behaviors and language, not statistical power.
Can an AI really run a discovery interview?: For structured generative interviews, yes. The interviewer is instructed to anchor on real events, ask for stories, and follow the frustration. It will not replace a deep expert interview, but it runs consistent, unbiased sessions at any hour in 7 languages.
How is this different from a survey?: The anchors are open and the follow-ups are unscripted, so it behaves like an interview rather than a questionnaire. You get walked-through stories and workarounds, not checkbox tallies, with full transcripts to quote from.
What do I do with the transcripts?: Read them for recurring behaviors, workarounds, and the words people use, then use the themes with quotes and the plain-language question box to test hunches like "who mentioned building their own workaround?"

Related templates

Research

Market research survey template

Interview-quality depth at survey scale, in the respondent's own words.

Research

Concept test template

Find out whether an idea lands before you build it, and hear the objections verbatim.

Product

Feature validation template

Check a feature solves a real problem before you commit the roadmap.

Product

Product feedback survey template

A recurring pulse that digs into the score instead of just plotting it.

Run the user research interview with your own audience

Join the waitlist