The impact of AI hallucination on the chatbot service quality in e-commerce: the moderating role of human agent based service recovery

Generative artificial intelligence has been widely adopted in e-commerce, demonstrating significant advantages in enhancing information processing efficiency and human-computer interaction. However, its inherent hallucination characteristic often results in the generation of false or misleading content that does not align with reality. In e-commerce, such AI hallucination is likely to trigger a crisis of trust among users and lead to negative evaluations of service quality. Clarifying the mechanisms through which AI hallucination affects chatbot service quality and exploring effective strategies to mitigate its negative effects have become critical issues. The study conducted two scenario-based experiments using a scripted chatbot with pre-written responses. No large language model (LLM) generated the experimental responses. Both experiments employed a 2 × 2 between-subjects design to examine the effects of simulated hallucination level and human agent-based service recovery on perceived service quality. Study 1 included 236 participants (129 female and 107 male, mean age = 30.10, SD = 6.76), and Study 2 included 364 participants (172 female and 192 male, mean age = 33.98, SD = 9.81). Mediation effect analysis supported indirect associations between higher perceived AI hallucination and lower perceived chatbot service quality through cognitive dissonance and technical frustration. The moderation effect further indicated that a human agent based service recovery strategy effectively mitigates technical frustration but struggles to reverse cognitive dissonance. This work extends research on AI hallucination in e-commerce and offers practical guidance for platforms on integrating human-AI collaboration to mitigate technological limitations.