trolls.dev
04 The lab — case study

AI User Interviews

User research at the speed of AI.

AI User Interviews runs realistic user interviews with AI — the moderator asks your questions, follows the interesting threads, and does it a hundred-plus times a week. No calls to schedule, no sessions to sit through, and synthesized findings while the question you asked still matters.

Status
Live
Platform
Web
Field
User research · AI
Cadence
100+ interviews / week
Open the platform ↗ All work
aiuserinterviews.com Live
The room
AI What nearly stopped you signing up?
In parallel 24 live
synthesizing 24 interviews… 3 themes
calls scheduled · 0 run while you sleep
The bottleneck was never the questions — it was the calendar. The bet behind it

Why we built it

The hard part of talking to users is the scheduling.

Everyone agrees you should talk to your users. Almost nobody does it as often as they should — not because the questions are hard, but because the logistics are. Recruiting, screening, finding a time that works for both of you, moderating the call, writing up the notes: the conversation is ten minutes, the overhead is a week.

So research gets rationed. You save it for the big decisions, run six interviews instead of sixty, and hear back long after the moment you needed an answer has passed. The calendar, not curiosity, decides how much you learn.

But the interview itself — ask, listen, follow up, ask again — is something a good enough model can now hold up its end of. If the moderator doesn't have to be a person on a call, the whole queue collapses, and you can talk to a hundred people before lunch.

Take the calendar out of user research, and the only limit left is what you think to ask.

What we made Four moves
  1. 01

    An interviewer, not a survey

    The AI moderator actually interviews. It asks the question on your guide, listens to the answer, and follows the thread — probing the vague reply, chasing the surprising one — the way a good researcher does, not the way a form collects a field.

  2. 02

    A hundred conversations at once

    Because nobody has to be in the room, the interviews run in parallel. Launch a study in the morning and a hundred-plus conversations happen at the same time — no recruiting queue, no calendar Tetris, no reschedules.

  3. 03

    Synthesis while it still matters

    As transcripts come in, the platform reads across all of them and pulls out the recurring themes, the sharp quotes, and where people split — so you're reading findings the same afternoon, not a slide deck three weeks late.

  4. 04

    Your study, your questions

    You write the discussion guide and set who you're talking to. The AI runs it faithfully at scale; you stay the researcher deciding what to ask and what the answers mean.

Choices we sweated

Scale that still tells you something true.

Interviewers, not answer-bots

The whole value of an interview is the follow-up question — the one you didn't script. We spent our effort on interviewers that probe and dig, because a panel that only reads its list is just a survey wearing a headset.

Findings you can trace back

A synthesized theme is only worth trusting if you can see where it came from. Every finding links back to the exact transcripts and quotes behind it, so a claim can be checked against the conversation rather than just believed.

A complement, not a replacement

AI participants are extraordinary for pressure-testing questions, exploring early, and reaching scale no calendar allows — and they are not real customers. We're honest about that line, so the tool sharpens real research instead of pretending to be it.

Further reading from the journal

We build around the moment a tool is actually used — the argument behind this one is in The ten minutes before the meeting, and why we only ship products we run ourselves is in Why we only build what we run.

AI User Interviews conducts interviews with AI participants and AI moderators. It is a tool for exploring questions and reaching scale that a calendar can't — a complement to research with real customers, not a replacement for it.

Colophon End of Case Study 04

Ask a hundred people tonight.

Write the discussion guide, launch the study, and read the synthesized findings before the question goes cold — no calls to schedule, nobody to keep waiting.

Open the platform ↗ More case studies →