Abstract: [Purpose/Significance] Artificial Intelligence Generated Content（AIGC）is reshaping the fundamental paradigm of contemporary information production and dissemination. When encountering AIGC, users often find it difficult to detect the hidden hallucinatory information due to the superficial rationality of the content. Therefore, effective intervention mechanisms are urgently needed to be explored to improve users' ability to identify AI hallucinations. [Design/Methodology] This study adopted a between-subjects experimental design: 2（text type: with vs. without hallucinations）× 2（intervention type: risk warning vs. explanatory information）. Popular science summaries generated by DeepSeek were used as experimental materials, and 651 participants were recruited to engage in a scenario-based experiment. Non-parametric tests and generalized linear models were employed for data analysis. [Findings/Conclusion] In the absence of intervention, users demonstrated limited ability to identify AI hallucinations, and intuitive processing alone was insufficient for effectively distinguishing hallucinated information. The effects of the interventions was jointly constrained by both the characteristics of the content and user traits: explanatory information can significantly improve the identification of texts containing hallucinations, whereas risk warnings had a relatively limited effect; users' AI literacy and prior knowledge negatively moderated the intervention effects, reflecting a substitution effect of cognitive resources and indicating individual differences in intervention effectiveness. [Originality/Value] The findings confirm the explanatory power of Dual-System Theory in the context of AI hallucination recognition , providing a new perspective for understanding cognitive processes involved in human-AI interaction. The conclusions offer implications for optimizing human-AI interaction interfaces and improving the governance of AI hallucinations.

Keywords: Generative artificial intelligence, AI hallucination, Risk warning, Explanatory information, Between-subjects design