Reappraisal is a linguistic strategy for regulating emotions that has the potential to produce powerful therapeutic effects. Consequently, there is growing interest in using large language models (LLMs) to help people reappraise. However, little is known about the factors that make one reappraisal more effective than another, or how these factors map to different individuals. A key challenge is the lack of a framework to differentiate reappraisals, making it difficult to characterize what is expected to work for whom. We conducted two studies (N = 471) examining values as a person-specific predictor of reappraisal effectiveness. Our findings suggest that LLM-generated reappraisals that aligned with individuals’ more strongly endorsed values (relative to their other values) were more likely to address their specific concerns, be perceived as more effective, and lead to greater emotional change. These results highlight the importance of tailoring reappraisals to individuals’ personal values and provide a foundation for improving the personalization of AI-assisted emotion regulation interventions.
Mehta et al. (Wed,) studied this question.