Learning-oriented generative AI use, self-directed learning, and learning anxiety in higher education: a cross-sectional study

Generative artificial intelligence (GenAI) has become an increasingly common learning resource in higher education, yet existing research has focused mainly on adoption, acceptance, and perceived usefulness rather than on the learning processes and emotional experiences associated with students’ everyday academic use of GenAI. Drawing on self-directed/self-regulated learning theory and a control-related interpretation of learning anxiety, this study examined the relationships among learning-oriented GenAI use, self-directed learning, and learning anxiety among university students. A cross-sectional survey was conducted with 742 students from six higher education institutions in China. Learning-oriented GenAI use, self-directed learning, and learning anxiety were each measured using six-item scales rated on a five-point Likert format. Data were analyzed using descriptive statistics, Pearson correlations, confirmatory factor analysis, structural equation modeling, and bootstrapped mediation analysis. The three-factor measurement model showed good fit to the data ( χ 2 = 287.46, df = 132, χ 2 /df = 2.18, CFI = .962, TLI = .955, RMSEA = .040, SRMR = .039). Learning-oriented GenAI use was positively associated with self-directed learning ( β = .47, p < .001), self-directed learning was negatively associated with learning anxiety ( β = −.41, p < .001), and GenAI use was also directly and negatively associated with learning anxiety ( β = −.10, p = .018). Bootstrapping indicated a significant indirect effect of GenAI use on learning anxiety through self-directed learning (indirect effect = −.19, 95% CI [−.25, −.13]), suggesting partial mediation. These findings indicate that the educational significance of GenAI depends not only on access or acceptance, but also on whether students use GenAI in ways that support planning, monitoring, reflection, and independent academic judgment.