One-on-one interviews give you depth: a person's own words, their reasons, the story behind a choice. They have always been slow to run, so most studies stop at fifteen or twenty. AI-moderated interviews change that. The same kind of conversation can run with two hundred people in a few days, at whatever hour suits each participant.
The method
A senior researcher writes the discussion guide: the questions, the order, the stimulus to show and the rules for follow-up. An AI moderator then conducts each interview one on one, by text or by voice, whichever the participant prefers. When an answer is vague, it asks for a specific example. When an answer opens something new, it asks why. Every participant is a real, screened person, recruited the same way we recruit for any study.
The question it answers
What do many people think, in their own words, and why? The method sits between a survey and a set of depth interviews. You get open, probed answers from enough people to count themes and compare groups.
How we run it
The researcher sets limits for the moderator: how many follow-ups per question, which topics to hold back until later, and when to move on. We pilot the guide with a small group, read every pilot transcript and adjust before full launch. During fieldwork, the researcher reads transcripts as they arrive, removes interviews that fail quality checks and audits a sample of the moderator's probes for leading wording. A researcher checks the theme coding and writes the readout.
What you get
The final guide, every transcript, a theme summary with counts by group, quotes linked to each theme, and a written readout with recommendations. Because the counts come from probed conversations, you can see how often a theme comes up and hear how people describe it.
Where it goes wrong
A weak guide produces weak interviews at scale, so the guide deserves the most senior time on the project. An automated moderator can probe too often and tire people out, so we cap follow-ups. Hands-on product use, group reactions and emotional topics still call for a human moderator, and the sample shows how we choose. And theme counts from open conversation are directional. When a decision needs a precise number, we follow up with a survey.
The sample below follows a fictional reusable water bottle brand that interviewed 180 owners about why bottles get left at home, and tested a new lid idea. Click the tabs to see the design, the guide, a transcript excerpt with probes, the themes and when to choose AI or human moderation. Every brand, person, quote and number is invented.
Illustrative only. All brands, products, people, quotes and figures are fictional. Open the sample full screen.