Large language models accurately identify decision reasons in verbal reports

Understanding the reasons behind human choices under risk is a central goal of the decision sciences, yet traditional methods relying on behavioral data are limited by strict invariance assumptions. Here, we introduce a scalable method using large language models (LLMs) to analyze verbal reports and identify the articulated reasons for choices between monetary lotteries. We show that a validated LLM accurately identifies predefined decision reasons in participants' free-text reports, aligning with their actual choices in over 92\% of trials. Our analysis reveals that reason usage varies systematically and is driven more by the choice problem's structure than by individual differences. A predictive model based on these problem-specific reason profiles outperforms prospect theory in out-of-sample prediction. This work demonstrates that verbal reports are a rich data source and that LLMs can unlock their potential, challenging foundational invariance assumptions and paving the way for more context-aware models of human decision-making.