Study: Book ratings miss well-being, AI recommenders need richer targets

A new study from researchers Aaron Marker, Joel Lehman, and H. Andrew Schwartz challenges the common assumption that user satisfaction metrics like ratings and likes reflect genuine well-being. Analyzing book consumption—where users write nuanced long-form reviews—the team compared satisfaction signals (ratings, sentiment) against psychological well-being dimensions such as meaning, accomplishment, and positive relationships. Their results show that ratings and sentiment only loosely correlate with most facets of well-being and align more with immediate hedonic gratification than enduring eudaimonic expression.

By linking review language to actual book content, the study uncovered specific associations: themes related to religion (r=0.35) and human drives (r=0.26) correlated with higher meaning and accomplishment, while incivility (avg r=-0.15) and past-focused language (avg r=-0.12) linked to lower well-being. The findings suggest that content recommendation systems—currently optimized for short-term satisfaction—could significantly improve user health by incorporating richer, multi-dimensional well-being targets beyond simple ratings or likes.