Abstract:

[Purpose/Significance] Artificial intelligence literacy is increasingly recognized as a critical competency in the GenAI era. Generative AI tools, such as ChatGPT and similar applications, are rapidly reshaping how students learn, write, search for information, and solve problems in higher education. Although these tools offer considerable potential for coursework learning, they also raise concerns about hallucinated content, overreliance, academic integrity, and uncritical acceptance of machine-generated outputs. Few existing technology acceptance models incorporate AI literacy as a core predictor, and there is little empirical evidence on its role in shaping usage behaviour. Some studies report a positive association between AI literacy and GenAI acceptance, whereas others suggest that lower literacy may predict greater receptivity and more frequent use. These inconsistencies highlight the need to examine AI literacy within a coherent technology acceptance framework. [Method/Process] This study addresses this gap by extending the UTAUT2 model with AI literacy and testing it with a large sample of 1 492 Chinese university students from Henan Province. The predictors included effort expectancy, performance expectancy, social influence, facilitating conditions, hedonic motivation, price value, habit, and AI literacy. The outcome variables were behavioral intention and usage behavior. AI literacy was measured with four items adapted from a validated AI literacy scale, covering awareness, use, evaluation, and ethics. Other constructs were adapted from established UTAUT2 scales and measured on a seven-point Likert-type scale. Multiple linear regression analysis was employed to examine the factors influencing GenAI usage behavior in coursework learning. [Results/Conclusions] The results show that the extended model demonstrates good explanatory power for behavioral intention (adjusted R²=0.698). Habit (β=0.417) and social influence (β=0.215) replace performance expectancy as the core drivers, while effort expectancy and price value are non-significant due to the characteristics of digital natives and the free-ecosystem context. Facilitating conditions exhibit a "split effect"—negatively predicting intention but positively predicting actual use. Notably, AI literacy shows opposing associations with behavioral intention and use behavior: it positively predicts behavioral intention (β=0.200) but negatively predicts use behavior (β=-0.139). We tentatively interpret this finding as reflecting a "rational gatekeeper" effect, which suggests that highly literate students may be more cautious about using GenAI. This study contributes to the literature by integrating AI literacy into the UTAUT2 framework. This extends the theoretical boundaries of GenAI education and reveals the paradoxical role of AI literacy in GenAI usage. These findings provide empirical evidence for universities to formulate informed GenAI educational policies. This study also has limitations. The sample was drawn only from Henan Province, so generalizability is limited. AI literacy was measured with four items, which may omit deeper dimensions such as critical thinking and ethical judgment. The study only examined direct effects and did not test for mediating or moderating mechanisms. The explanatory power for self-reported usage frequency was modest, indicating that other factors may matter. Future research should employ more comprehensive AI literacy scales, examine mediation and moderation, conduct qualitative interviews, collect behavioral logs, and incorporate variables such as self-efficacy, technology anxiety, trust, and perceived credibility. Cross-regional and longitudinal designs would further clarify how AI literacy shapes responsible GenAI use in higher education.

Key words: generative artificial intelligence, UTAUT2, college students, artificial intelligence literacy, usage behavior