AI companions and subjective well-being: Moderation by social connectedness and loneliness

Conversational AI companions—such as Replika and Character.AI—are increasingly adopted to provide emotional support, yet their psychological effects remain underexplored. This study investigates whether the use of AI companions is associated with enhanced well-being, and whether these associations are moderated by social network/support and loneliness. We analysed cross-sectional data from 14,721 Japanese adults participating in nationwide internet panel surveys conducted in December 2024 and January 2025. Well-being was assessed across three domains: evaluative (life satisfaction), hedonic (happiness), and eudaimonic (purpose and meaning in life). AI use was categorized as either companion or non-companion. Moderators included social network/support (measured via the Lubben Social Network Scale, LSNS-6) and loneliness (UCLA Loneliness Scale). Multivariable linear regression and restricted cubic spline models were used to assess associations and effect modification. Use of AI companions was significantly associated with higher scores across all well-being domains. In contrast, non-companion AI use showed weaker or inconsistent associations. A U-shaped moderation pattern emerged for friend-based social network/support: the benefits of AI companions were most pronounced among those with moderate levels of social connection and attenuated among those with either very high or very low levels. The strongest positive associations were observed among individuals reporting high loneliness. These findings suggest that AI companions may offer emotional and psychological benefits, particularly for individuals with unmet social and emotional needs or moderate social embeddedness. Future research should explore causal mechanisms and develop design strategies that promote well-being without impairing real-world social engagement. • Companion AI use was associated with higher well-being • Associations were strongest among users with moderate friend networks • Lonelier individuals showed more positive associations with AI companions • No clear well-being associations were found for non-companion AI use • AI companions may support well-being, especially for socially vulnerable groups