Examining the influence of expectation disconfirmation on AI agent switching intention in mental health

AI agents have emerged as prominent alternatives in mental health service delivery. However, users consistently express a preference for human agents over their AI counterparts. This study examined the factors influencing users' intentions to switch from AI agents to human agents in mental health contexts. Grounded in the expectation disconfirmation theory, we developed a theoretical model examining the effects of user expectations and AI agent performance on disconfirmation, satisfaction, and switching intention. The model was empirically validated through a 2 × 2 (user expectations: high vs. low × AI agent performance: high vs. low) between-subjects experimental design (N = 200). The results suggested that AI agent performance emerged as the primary determinant of user satisfaction and switching intention in mental health contexts. In contrast, user expectations, while influencing disconfirmation, exhibited a relatively limited direct influence on satisfaction and switching intention. Intriguingly, the impact of AI performance on satisfaction was moderately strengthened under conditions of high user expectations. These findings both validate the applicability of expectation disconfirmation theory in human-AI interaction research and offer practical recommendations for developers of AI-based mental health interventions.