When loneliness amplifies the risk of problematic emotional dependence: perceived anthropomorphism and basic...

Companion-oriented conversational AI can be experienced as socially responsive, yet evidence about problematic reliance requires construct clarity and noncausal interpretation. This study examined whether perceived anthropomorphism was associated with problematic AI-companion dependence directly and indirectly through basic psychological need satisfaction.

A cross-sectional online survey was completed by 666 adults in China who reported using a listed AI companion or conversational AI product during the preceding month. The nonclinical outcome was operationalized with six indicators spanning impaired control, distress, continued use, perceived understanding, emotion-regulation reliance, and displacement of offline communication. Measurement evidence was examined using reliability estimates, one-factor confirmatory models, and HTMT. Standardized composite-score regressions used HC3 robust standard errors, and the indirect association was evaluated with 5,000 bootstrap samples.

Perceived anthropomorphism was positively associated with need satisfaction (β = 0.569, p < 0.001). In the outcome model, need satisfaction (β = 0.317, p < 0.001) and perceived anthropomorphism (β = 0.385, p < 0.001) were both positively associated with problematic dependence. The standardized indirect association was 0.180 (95% bootstrap CI [0.137, 0.229]); R ² was 0.323 for need satisfaction and 0.389 for problematic dependence.

The pattern is consistent with a need-satisfaction account of companion-oriented AI use, but the cross-sectional design does not establish temporal ordering, substitution, or causality. Design and policy implications should therefore focus on transparency, user agency, and safeguards against escalating reliance rather than treating anthropomorphism itself as inherently harmful.