Abstract
Generative artificial intelligence's (GenAI) fast progress has opened up new possibilities, but it has also increased the likelihood of service failure. This study investigates how belief priming affects users' intention to switch following a failure in GenAI services. Based on mental model theory and the associative proposition evaluation model, we conducted scenario surveys and event-related potential (ERP) studies as part of a mixed-method approach to explore the impact of different types of belief priming (AI emotions are real vs. AI emotions are fake) on users' switching intentions, and examined the mediating role of fault attribution and the moderating role of task type (emotional tasks vs. mechanical tasks). The questionnaire results show that priming with the belief "AI emotions are fake" can effectively reduce users' switching intentions after service failures, especially in emotional tasks. Error responsibility attribution plays a mediating role between belief priming and switching intentions. ERP results indicate that, in the event of service failure, the amplitude of the P2 component in the "AI emotions are real" belief priming group was significantly higher than that in the "AI emotions are fake" belief priming group, especially in emotional tasks. This study reveals the significant role of belief priming in shaping users' reactions to failures in GenAI services, providing empirical evidence for GenAI service providers to develop effective service remediation strategies.