An AI-moderated interview, or AIMI, is a research interview with real participants in which software asks a researcher-defined set of questions and generates follow-up questions from each participant's answers. The studies reviewed here describe a semi-structured method: the researcher controls the study objective, core questions, order and probing instructions, while the AI personalizes some follow-ups in real time. AIMIs can use text, voice or both. They can be run as self-administered interviews or in a controlled setting. They should not be confused with synthetic respondents, because the answers come from recruited people.
AIMI stands for AI-moderated interview. In the five Glaut Research studies, AIMIs sit between a static questionnaire and a human-moderated depth interview. Like a survey, an AIMI can give every participant the same core questions in the same order. Like an interview, it can react to an answer and ask a relevant follow-up. Curtin University describes this as a semi-structured process: predefined questions are asked consistently, then answer-specific probes vary by participant.
This makes AIMI a useful label for the collection method. It does not imply that every research activity, from recruitment to reporting, is automated.
The common workflow across the studies is:
The University of Nottingham study used six researcher-written open questions and allowed one AI-generated follow-up after each. Curtin University used predefined, pre-ordered core questions with answer-specific probes. Mannheim compared dynamic AI probes with two predefined survey follow-ups. These designs show that the AI does not independently decide the research agenda.
Follow-ups are generated from two inputs: the participant's latest answer and the researcher's instructions. Nottingham provides the clearest evidence of how those inputs interact. When an initial question was broad and the AI was instructed to explore an underdeveloped area, the probe often added useful information. When the initial question already asked for reasons and the AI was also told to ask for reasons, the follow-up could become tautological. The practical implication is that a probe is only as useful as the relationship between the static question, the answer and the probing instruction. Dynamic generation does not guarantee novelty.
The studied implementations automated the delivery of questions and the generation of contextual probes. Voice implementations also used speech interfaces or transcription.
Researchers retained control over:
Responsive Research adds an important caution. Automated themes and summaries can compress contradictions or edge cases. Human review remains necessary when the research decision depends on nuance.
The evidence supports a hybrid use case, but the label depends on the design.
An AIMI produces open-ended, conversational material associated with qualitative research. At the same time, it can preserve standardized core questions across larger samples. Mannheim used two groups of 100 participants. Human Highway compared 503 traditional questionnaire cases with 500 conversational AI cases. Nottingham recruited a demographically representative UK sample of 296 respondents.
These studies show that conversational open ends can be placed inside a structured sample and questionnaire. They do not make every AIMI statistically representative. Representativeness still depends on recruitment, quotas and analysis.
Most Glaut studies used self-administered sessions that participants completed in their own time.
AIMI therefore describes who moderates the interview, not one fixed timing model.
Text and voice were tested. Human Highway allowed participants to select text, voice or a combination, while Responsive Research also enabled participant choice. Mannheim intentionally restricted responses to text so that the AI and survey conditions were comparable. Curtin used oral answers in both the AI and human conditions.
The studies analyzed transcripts using several complementary methods:
No single metric is enough. Longer answers may still be repetitive. Theme counts can miss the richness of how a theme is expressed. Responsive Research recommends treating automated synthesis as an analytical starting point and returning to the underlying responses when meaning may have been compressed.
A generic chatbot describes an interface. An AIMI describes a research method with a defined sample, a discussion guide and controlled probing objectives. The difference is methodological discipline. The core questions should cover the research objective consistently, while the adaptive probes respond to individual answers. A chatbot can be conversational without meeting those requirements.
The terms overlap. Mannheim and Human Highway describe AIMIs as conversational alternatives to static questionnaires. A conversational survey may be the broader category, while AIMI emphasizes that the AI takes a moderator role by generating contextual probes. The five studies do not establish a universal naming standard. Researchers should define the method in the report instead of relying on the label alone.
The reviewed studies test discrete interview sessions with a predefined guide. They do not test repeated participation over time, participant-to-participant interaction or community management. An online community or diary study is defined by longitudinal or social participation. An AIMI could potentially be used within those designs, but that combination was not evaluated in these papers.
An AI interviewer asks questions to a real person. A synthetic respondent generates the answer itself. All five studies used recruited human participants. The data analyzed by Mannheim, Human Highway, Curtin, Nottingham and Responsive Research came from people, not simulated personas. This distinction is essential because evidence about AIMIs does not validate synthetic samples.
Yes. The research involved panel participants, qualitative recruits, university students and staff, and a UK public sample. AI was the moderator or probing mechanism. The respondents were human.
Across the studies, AIMI is best understood as a semi-structured interviewing method that combines standardized researcher control with adaptive questioning. Its strongest documented contribution is adding contextual follow-up to open-ended research at a larger operational scale than conventional human moderation.
The studies also set clear boundaries. AIMI performance depends on question design and participant input. It does not remove the need for researchers to design the study or interpret the evidence.
No. In the reviewed studies, researchers defined the topic, core questions and probing goals.
That is a design choice. Nottingham used one probe per static question. Mannheim allowed more but analyzed only the first two for comparability.
Yes. Mannheim, Human Highway and Nottingham combined open-ended interviewing with closed or demographic questions.
No. The method concerns data collection. The studies used both automated metrics and human interpretation, and Responsive Research warns against accepting summaries without review.
Collect, analyze, and report research from any source with more depth, speed, and control.
Schedule a free demo
