AI companions are virtual conversational agents designed to foster parasocial interactions and provide long-term emotional support. Research on user engagement has largely focused on its initiation and maintenance, whereas subsequent user disengagement remains significantly underexplored. This research gap is increasingly critical as mainstream applications face sharp declines in active usage rates and download volumes. This study investigates the determinants of disengagement by considering both technical performance and the unique relational and social-emotional dimensions of AI companions. Using a mixed-methods approach, we first employed computational grounded theory to analyze user concerns on social media, followed by semi-structured interviews to identify the specific triggers of negative experiences and disengagement. Theoretically, this work expands the scope of the engagement process by focusing on the conclusion of human-AI relationships. Practically, it offers design interventions to help developers address user churn and enhance the sustainability of AI companion services.
Ye et al. (Mon,) studied this question.