How do AI feedback types affect students’ continuance usage intention in AI hallucinations?

AI hallucinations threaten the sustainable use of generative AI in higher education, yet prior research has largely emphasized technical detection rather than students’ psychological responses and socio-emotional repair. This study examined how faithfulness and factuality hallucinations affect university students’ trust and tolerance, and how empathy- and humor-based repair strategies shape trust, tolerance, and continuance usage intention. Using a 2 × 2 × 2 mixed design, 400 university students participated in two educational scenarios. Mixed-effects models and bootstrap mediation analyses showed that factuality hallucinations elicited lower trust and tolerance than faithfulness hallucinations. Across both scenarios, high empathy combined with self-deprecating humor produced the most favorable outcomes. Trust and tolerance independently mediated the relationship between repair strategies and continuance usage intention, whereas serial mediation was not supported. These findings highlight the value of socio-emotional repair in designing educational AI systems that maintain student engagement after hallucination-related failures.