Psychological predictors of continued ChatGPT use among university students: the roles of AI literacy, trust, and academic self-efficacy

Introduction The rapid diffusion of ChatGPT across higher education has outpaced understanding of the psychological factors associated with how far students sustain and intensify their use of it. Drawing on technology acceptance theory, trust-in-automation theory, and Bandura's self-efficacy theory, this study tested a moderated serial mediation model in which AI literacy is associated with continued ChatGPT use through an indirect pathway involving trust in AI and academic self-efficacy, with AI anxiety moderating the literacy-to-trust association. Methods Survey data were collected from 450 Chinese university students who had prior experience of using ChatGPT. Students without such experience were ineligible, so the model concerns variation in continued use among existing users rather than the initial decision to take the tool up. Confirmatory factor analysis supported the measurement model, and a full latent-variable structural equation model with Monte Carlo confidence intervals (20,000 draws) tested the hypotheses. Results AI literacy was associated with continued use both directly and indirectly through trust and self-efficacy. The hypothesized serial indirect path, from AI literacy to trust to self-efficacy to continued use, was significant. AI anxiety attenuated the literacy-to-trust association, and the index of moderated mediation indicated that the serial indirect effect was weaker at higher levels of anxiety. The pattern was unchanged when continued-use intention and current use intensity were modeled as separate outcomes. Discussion The findings position trust and self-efficacy as sequentially ordered correlates rather than parallel ones, and identify anxiety as a boundary condition on that pattern. Because the design is cross-sectional, this ordering is model-specified rather than established temporally; because the sample comprises students who already use ChatGPT, the results speak to the intensity of continued use and not to first-time uptake. Implications for AI literacy education and the design of student-facing generative AI tools are discussed.