Explanation Transparency and Trust Perception After Generative AI Failure: A Dual-Path Model

AI hallucinations raise questions about how users evaluate generative AI after failures and the explanations that follow. Transparency is regarded as a core principle of responsible AI, but its role in failure scenarios remains unclear. Integrating ISSM, TPB, and Cognitive Appraisal Theory, we examine how users cognitively appraise post-failure explanations and form trust perceptions. A multi-scenario hybrid experimental design (N = 600) showed that explanation transparency was associated with post-failure trust perceptions through two cognitive pathways. Transparency enhanced perceived information quality (PIQ) and perceived behavioral control (PBC). However, the effect of transparency was not linear. Exploratory analyses suggest that task criticality conditions the effects of transparency, while information overload weakens the relationship between PBC and willingness to use. This study reveals that after AI failure, transparency serves not to prove correctness but to support users’ evaluative capacity and sense of control through cognitive evaluation mechanisms.