AI-Moderated Interviews: If, When, and How to Use Them

Why Read This from an HCI Perspective

This article is highly meaningful for HCI/UX practitioners because it provides evidence-based examples showing how far AI can replace research interviews—and from where the human moderator’s role becomes essential. In particular, by clearly outlining the boundary between structured and semi-structured interviews, it helps readers evaluate both the opportunities and limitations of research automation. It also addresses participant experience concerns—such as trust, nonverbal cues, consent procedures, and privacy awareness—making it immediately useful from a research design perspective.

CIT's Commentary

From a CIT perspective, the core point of this article is that even if an AI interviewer speaks well, it still cannot yet read the context of the interaction. It is quite useful for tasks with clear structure—such as summaries, multilingual support, and screening—but in exploratory research, the moment-to-moment judgments needed, interpreting the meaning of silence, and tracking unexpected clues are still the domain of human moderators. What’s especially interesting is that participants were more sensitive to nonverbal feedback and the process of building trust than to the accuracy of the results. This suggests that the success or failure of AI research tools depends not only on model performance, but also on how carefully research ethics and interaction staging are designed. CIT therefore considers it realistic to view such tools not as ‘replacements,’ but as supportive infrastructure that reduces repetitive, standardized segments.

Questions to Consider While Reading

• Q.What interaction design would be needed to help an AI interviewer—strong with structured questions—perform well with the ‘flexible follow-up questions’ required in semi-structured interviews?
• Q.To make participants trust an AI interviewer, how should the introduction text, consent process, and feedback approach be reconfigured?
• Q.For use cases where effectiveness is clear—such as multilingual support and screening—and for areas where human judgment is needed, such as exploratory research, what criteria are most practical for separating the two?