AI Conversational Interviewing: Scaling Up Semi-Structured and In-depth Interviews

Research Context and Approach

Public opinion research has long struggled with a fundamental trade-off: standardized surveys offer high scalability and comparability but lack depth, while traditional in-depth interviews provide rich qualitative insights but are prohibitively expensive to scale. This study introduces AI Conversational Interviewing, which leverages Large Language Models (LLMs) to conduct semi-structured, open-ended interviews at scale. The researchers tested this method by surveying 571 participants on migration policy, comparing voice-based, chat-based, and choice-based interaction modes against a standard survey battery.

Analytical Value

The primary advantage of the AI-led approach is its ability to uncover the 'why' behind public opinion. While standardized surveys can measure the intensity of an attitude (e.g., on a Likert scale), they often miss the underlying mental models. The study demonstrates that respondents with identical survey scores often hold vastly different, sometimes contradictory, justifications for their views. By allowing for neutral, adaptive probing, the AI interviewer surfaces these distinct belief systems, providing researchers with a deeper understanding of how citizens construct their political opinions.

Practical Viability and Implementation

Participants generally evaluated the AI-led interviews as favorably as, or better than, traditional standardized surveys. The study provides experimental evidence on implementation, showing that while voice and chat modes are both feasible, they require careful tuning of turn-taking and prompting to ensure a natural flow. The authors emphasize that this method is not a replacement for human-led qualitative research or standardized surveys, but rather a powerful, complementary tool that expands the social science toolkit to address questions previously considered out of reach for large-scale quantitative studies.