Boon or a bane: Different usage of LLM correlates with mental health

While large language models (LLMs) are increasingly integrated into daily life, the relationship between purpose-specific usage and mental health, as well as the factors that moderate these links, remains poorly understood.

This study examined how LLM usage for different purposes relate to mental health outcomes (i.e., depressive symptoms, insomnia symptoms, and suicidal ideation) and whether these associations are moderated by individuals' propensity to use putatively adaptive or maladaptive emotion regulation (ER) strategies.

A cross-sectional survey was conducted among 6710 Chinese college students, including 3731 active LLM users. Participants reported their purpose-specific frequency of LLM usage, ER strategies, depressive symptoms, insomnia symptoms, and suicidal ideation. Regression and moderation analyses were performed.

Purpose-specific analyses showed that higher LLM usage for relationship advice was linked to lower levels of depressive symptoms, insomnia symptoms, and suicidal ideation, while usage for work assistance showed the opposite pattern. Greater usage for life and entertainment was associated with more insomnia symptoms, and usage for daily chatting was linked to increased suicidal ideation. Critically, adaptive ER strategies buffered risks and enhanced protective associations, whereas expressive suppression tendency amplified negative outcomes.

Individuals' mental health is differently associated with LLM usage for different purposes, and these associations further vary according to their ER tendencies, underscoring the need for nuanced approaches in digital mental health research.