Understanding problematic AI companion use through escapism and resilience

As large language models (LLMs) become more sophisticated and relational, many users now engage with them in ways that resemble friendship, romance, or ongoing emotional support. While these interactions can be positive, emerging evidence suggests that a subset of users experience difficulties regulating their engagement, leading to patterns better characterized as problematic, dysregulated, or maladaptive use. Guided by Compensatory Internet Use Theory and the I-PACE model, this study examines a risk-buffer framework in which escapism is associated with greater problematic AI-companion use and resilience weakens that association. Results indicate that escapism is associated with problematic use of AI characters and resilience significantly attenuates this association. Together, these findings point to a pattern in which escapism is linked to greater problematic use, whereas resilience functions as a protective psychological resource associated with a weaker escapism–problematic use relationship.