Generative AI interaction styles and mental health among college students: a latent profile analysis

Abstract

Generative artificial intelligence (GenAI) is now widely used by college students, but the mental health implications of different interaction patterns remain unclear. Because GenAI can support task-focused, cognitive, and emotionally salient forms of engagement, engagement quality may be more informative than use intensity alone. We used latent profile analysis to identify human–AI interaction styles among 7,029 Chinese college students from six universities. Five indicators captured use intensity, perceived interaction quality, trust, psychological closeness, and emotional attachment; depressive and anxiety symptoms were assessed as mental health outcomes. Five profiles emerged: Efficient-Instrumental (27.00%), Exploratory-Casual (31.97%), Deep-Emotional (12.18%), Ambivalent-Attached (9.23%), and Detached-Avoidant (19.62%). Profiles differed significantly in depressive and anxiety symptoms. The Ambivalent-Attached profile, characterized by high emotional attachment but low perceived quality and trust, showed the highest symptom levels, with mean depressive symptoms in the moderate range. Symptom differences aligned more closely with relational quality indicators than with use intensity. Random forest and SHAP analyses identified smartphone addiction, neuroticism, conscientiousness, and cognitive reappraisal as the most informative features for classifying profile membership. These findings suggest that university mental health services should consider how students relate to GenAI, not only how often they use it.