AI Conversational Interviewing Scales Deep Interviews to 571 Respondents — And That Changes Everything About Public Opinion Research

A team of researchers from LMU Munich and the University of Mannheim has published a paper on arXiv introducing AI Conversational Interviewing, a method that leverages large language models to conduct semi-structured, in-depth interviews at scale. The study, which involved 571 respondents recruited via Prolific and a Payback Panel, compared three interview modes: voice-based, chat-based, and a free-choice option. Each participant experienced an AI-led conversation about migration policy, followed by a standardized survey. The AI system used a predefined interview guide but adapted its probing based on respondent answers, mimicking human interviewers.

The findings demonstrate that AI conversational interviews surface considerations and reasoning that even comprehensive closed-ended surveys miss—such as distinctly different mental models of migration among subgroups with similar surface-level attitudes. For example, respondents with comparable policy preferences revealed contrasting underlying logics about economic impacts or cultural integration. Crucially, evaluations of the AI interview were at or above those of the standardized survey across all modes, though completion rates were highest in the chat mode. The authors release open data and open-source pipeline materials, arguing this method can help researchers scale qualitative depth without sacrificing the richness of human conversation. This work contributes to a growing field using AI to augment traditional social science methodologies, offering a practical tool for opinion measurement, market research, and user experience studies.