What an AI customer interview is
You write an interview guide, the same five to eight open questions you would ask on a call. An AI interviewer asks them one at a time, in a chat, to each customer who opens your link. When an answer is thin or interesting, it asks a follow-up before moving on. Each conversation ends with a transcript, and analysis runs across all of them.
The participant is a real person: a customer, a churned user, a prospect. The only thing that is automated is the moderator. That distinction matters, because a growing number of tools sell synthetic users, which are language models pretending to be your customers. Those can help you draft a guide. They cannot tell you what your customers think.
How it works, step by step
The mechanics are simple enough to run your first study in an afternoon.
- Write the guide. Lead with context, then move to specifics. Tell the interviewer what you are trying to learn so its follow-ups point in the right direction.
- Add optional screening questions so people who are not a fit are turned away politely before they start.
- Share one link: in an email, a cancellation flow, a Slack community, or a panel like Prolific. No calendar, no time zones.
- Participants type (or talk, if they prefer) for about 10 to 15 minutes in their own language. The interviewer asks your questions plus follow-ups, then wraps up.
- Read transcripts as they land, and let theme extraction, sentiment by topic, and completion analytics run across the whole set. Every theme cites the interviews and quotes behind it.
What you get back
The deliverable is evidence, not a dashboard. Full transcripts you can edit and quote. Themes grouped across participants, each linked to the exact passages that support it. Sentiment broken out by topic, so you can see that onboarding is loved and billing is not. Drop-off by question, which tells you where your guide is too long. Exports as CSV or a Markdown report you can paste into a doc for stakeholders.
On the Pro plan you can also ask questions of the evidence directly ("what did people who mentioned price say about alternatives?") and get answers with participant citations.
When AI interviews beat the alternatives
AI interviews sit between surveys and moderated calls. They win when you need the depth of a conversation at a volume where scheduling calls is impossible, or with people who will never book a call with you.
- Churned customers. They will answer a five-minute chat from a cancellation email. They will not book a Zoom. See churn interviews.
- Continuous discovery. An always-on link in your product or onboarding sequence yields a steady stream of interviews instead of a quarterly research sprint.
- Concept and prototype feedback. Attach an image, a Figma prototype, or a web link to a question and ask people to react to it.
- Win/loss. Prospects who said no explain why, in their own words, without a sales rep on the line.
Where they fall short
Be honest about the limits. An AI interviewer will not notice a participant hesitating, will not improvise a brand-new line of inquiry, and cannot build the rapport a skilled researcher does over an hour. For foundational research on a topic you barely understand, run a handful of human-moderated sessions first, then use AI interviews to test what you learned at scale.
Also watch the sample. Twenty transcripts tell you which themes exist. They do not tell you how prevalent each theme is in your customer base. Pair interviews with a survey when you need the number.
Writing a guide an AI interviewer can run well
A few habits make a large difference to transcript quality.
- Ask about past behaviour, not hypotheticals. "Tell me about the last time you…" beats "Would you use…".
- One idea per question. Compound questions produce half-answers.
- Put the interviewer's goal in its instructions, e.g. "we want to understand why people stop after the first project", so its follow-ups dig in the right place.
- Cap follow-ups at one or two per question. Participants' patience is the real constraint.
- Pilot it on yourself and two colleagues before sending it to customers.
Getting started
AI User Interviews is built for exactly this workflow and is self-serve: Free plan with 5 active projects and 100 participants a month; Pro is $25 a month, no credit card and no sales call required. Create a project, paste your questions, and share the link. You are reading your first transcript today.