Large language models accurately identify decision reasons in verbal reports

Significance Understanding why people make the choices they do is central to decision science. We show that large language models can uncover people’s stated reasons from free-text reports, achieving 95% alignment between actual choices and those implied by the identified reasons. Based on this scalable classification approach of decision reasons, we find that the reasons vary primarily with the structure of choice problems, and less so across individuals. Notably, reasons identified from verbal reports provide a more parsimonious account of decision processes than reasons inferred from choices alone, demonstrating the epistemic value of verbal reports. Our findings challenge the field’s reliance on contextual invariance assumptions and establish LLMs as a transformative tool for building more context-sensitive models of human decision making.