Conducting Qualitative Interviews with AI

Qualitative interviews offer rich insights into human behavior, but their lack of scalability has limited their use in economics. This paper introduces and evaluates AI-led interviews as a novel method for collecting qualitative data. Across different economic contexts, a comprehensive set of quality metrics demonstrates the high quality of AI-led interviews. We then demonstrate that AI-led interviews generate thematically richer responses than other survey methods for collecting qualitative data at scale. A decomposition exercise shows that the benefits of AI-led interviews mainly arise from asking dynamic probing questions. Importantly, AI-led interviews make themes visible that would otherwise be unobserved. These themes matter: We show that responses in AI-led interviews are internally valid and predict future behavior six months later. Finally, we demonstrate that AI-led interviews allow hypothesis generation about important economic phenomena, reaching thematic saturation in small samples comparable to studies using human-led interviews.