Companion-oriented AI chatbot use and loneliness among Chinese emerging adults: the mediating role of self-attributed offline social displacement and the moderating role of relational-interdependent self-construal

Objective Companion-oriented AI chatbots are diffusing rapidly among Chinese emerging adults, yet their relationship with loneliness remains unclear. Adopting a media-effects perspective that treats companion AI use as a form of media use, this exploratory study examined whether companion AI use is associated with loneliness through self-attributed offline social displacement, and whether relational-interdependent self-construal moderates this association. Methods Using a cross-sectional survey, we administered measures of companion AI use, self-attributed offline social displacement, loneliness, and relational-interdependent self-construal to 843 Chinese emerging adults (aged 18–29 who had used a companion AI within the preceding 3 months). Because no established Chinese measures were available for the two companion-AI-specific constructs, study-specific preliminary item sets were used and their internal structure was examined with split-half EFA/CFA before the moderated mediation analyses. Results Self-attributed offline social displacement partially mediated the relationship between companion AI use and loneliness (the indirect effect accounted for roughly one-third of the total effect). Relational-interdependent self-construal moderated the second stage of this mediation: the indirect effect through displacement was significant only among individuals with moderate-to-high interdependence and was non-significant among those with low interdependence. Conclusion The findings are consistent with a self-reported displacement-related association in the context of generative companion AI and point to a possible boundary condition on that association tied to relational-interdependent self-construal. The results should be interpreted as preliminary and correlational because the displacement measure asked respondents to attribute reduced offline interaction to AI use and was not validated in an independent sample.