Hey Juno

What is AI-moderated qualitative research?

By Hey Juno · Published · Updated

AI-moderated qualitative research is a method where an AI interviewer, not a person, runs in-depth interviews with real participants, by voice or text, and the findings are written up from what those people said. The interviewer asks open questions, listens, and follows up on the answers, so you get explanations rather than ticked boxes. The difference from a survey is depth. The difference from traditional interviewing is that hundreds of conversations can happen in days.

Juno, the AI researcher from Hey Juno, is one example. It is built on a simple belief: the job is to listen to people and report their experience accurately, without steering it. This guide explains the method in general terms and uses Juno to show what each step can look like.

What it is, and what it isn't

In-depth interviews are the core tool of qualitative research. A skilled moderator asks open questions, notices what is odd or unfinished in an answer, and asks about it. The aim is to understand why people think and act as they do.

In AI-moderated research, the moderator role is played by an AI system. The participants are real people. The conversation is real. What is automated is the interviewing, and in some tools, the analysis.

The test of any such system is not whether it can chat. It is whether it can draw out what someone means without influencing what they say. An interviewer that nudges answers produces transcripts that read well and are worth little.

It is not a survey with a chat window on it. A conversational survey works through a fixed questionnaire and may add a follow-up such as "tell me more". An AI-moderated interview starts from what the study needs to learn and chooses its questions for each participant. Our guide to AI interviews versus conversational surveys covers the difference in detail.

It is also not simulated research. Some tools use synthetic personas, which are AI-generated stand-ins for customers. That is a different thing. In AI-moderated qualitative research, every finding comes from a real participant. (Juno does run simulated interviews, but only to rehearse a study before launch, never as research data.)

How it works, end to end

The steps are the same as for any qualitative study. What changes is who does the fieldwork.

1. Brief. You describe the decision you need to make and what you need to learn. With Juno this is done in plain language, and you do not need research training: Juno is the researcher, and you are the person who needs the answer. It will also push back if a brief assumes its own answer. Juno turns the brief into research objectives, which are statements of what the study must find out, rather than a list of questions. It shows you the assumptions it made so you can confirm or correct them.

2. Design and check. A good study design is checked for leading wording, built-in assumptions and false either/or choices before anyone is interviewed. Juno does this review, and it can run a rehearsal: a simulated interview that shows where questions confuse or fail to produce useful detail. You can add questions that must be asked word for word.

3. Recruit. Participants can be your own customers, invited by link, or people recruited through a panel. Hey Juno offers its own panel partners, or you can connect a panel provider you already use.

4. Interview. Each participant takes part in their browser, by voice or text. The interviewer follows the study objectives, but not a fixed script. If someone gives a flat answer ("it was fine"), it asks what that means. If someone says something unexpected, it can follow it, while still covering the objectives so interviews stay comparable.

5. Check quality. Not every interview is good. Short, low-effort or incomplete interviews should be flagged rather than quietly counted. In Juno, recruited interviews are marked Qualified, Disqualified or Incomplete, with the reasoning visible.

6. Analyse and debrief. The findings are organised around your objectives. In Juno each claim links back to the participant, the quote and the transcript moment behind it, and you can question the findings in a debrief rather than only reading a fixed report. Juno reports what people said and how far the evidence goes. What to do about it is your decision.

What it is good for

AI moderation suits questions where the reasons matter and you need to hear from more than a handful of people. Typical uses include:

  • Concept and message testing: which ideas land, what people think a message is saying, and why.

  • Churn, onboarding and conversion research: where people get stuck, what tipped a decision, and what the stated reason hides.

  • Pricing and objections: the reasoning behind price sensitivity, in people's own words.

  • Brand and category understanding: perceptions, motivations and barriers across groups, or evidence for an agency pitch.

  • Customer and user research at scale: the same interview across many people, markets or languages.

Interviews can run at once, so fieldwork does not queue behind a moderator's diary. Participants can take part when it suits them, in their own language, with nothing to download. Every interview has a transcript, and voice interviews have audio, so every claim can be checked against what was said.

What the evidence says

Research on AI interviewers is young and mostly small-scale, so treat it as encouraging, not settled.

In one economics study, Chopra and Haaland used an AI interviewer to run 381 interviews about why people stay out of the stock market. The authors argue that AI makes qualitative interviews feasible at scale (Chopra and Haaland, 2023, SSRN). A separate pilot by Wuttke and colleagues (2024), with university students, found that an LLM interviewer produced data comparable to conventional approaches, with the added benefit of scale (Wuttke et al., arXiv).

Both are early studies of specific set-ups. They do not show that every AI interviewer is good. Quality depends on how the interviewer is built, what rules it follows, and how well the study is designed.

Where this doesn't fit

AI moderation is not the right tool for every question. Be cautious in these cases:

  • Observation. If you need to watch people use a product, or see a home or a workplace, an interview of any kind only gives you what people say about it.

  • Very sensitive topics. Some conversations need a trained human who can read a room and take responsibility on the spot. Juno applies a four-tier wellbeing framework in every interview, but some subjects are better handled by a person.

  • Group dynamics. If the point is how people influence each other, as in a focus group, one-to-one interviews will not show it.

  • Numbers. Qualitative research tells you why and how. It does not tell you how many people in a population think something. For that, use a survey, or add structured questions alongside the interview.

How to judge a tool

Ask for specifics, not credentials. A claim of expertise proves nothing. A named check does. Three questions are enough to start:

  • Can I get from a conclusion to the exact interview moment myself?

  • What happens to an interview that went badly?

  • What would change the tool's mind about a finding?

A vendor that cannot answer these has not shown that its research can be trusted. Ask them of Juno too. The fuller set of questions is on the Hey Juno methodology page.

Frequently asked questions

What is AI-moderated qualitative research?

It is qualitative research in which an AI interviewer conducts in-depth interviews with real participants and the findings are based on what they said. It combines the depth of interviewing with the scale usually associated with surveys.

Is AI-moderated research the same as using synthetic respondents?

No. Synthetic respondents are AI-generated stand-ins. In AI-moderated research, real people take part in real interviews. Juno uses simulated interviews only to rehearse a study before launch, never as findings.

How many interviews do I need?

It depends on the question and on how varied your audience is. For a focused topic and a uniform group, empirical studies suggest themes often settle within roughly 9 to 17 interviews, and understanding them fully often takes more. Varied audiences and segment comparisons need more. Our guide to participant recruitment for AI interviews covers sample size in more detail.

Can AI-moderated interviews replace human moderators?

For some studies, yes. For others, no. Our comparison of AI-moderated and human-moderated interviews sets out where each is stronger and how to combine them.