Association between perceived usefulness of generative AI for learning and generative AI dependency among university students in Changde, China: a cross-sectional study of statistical indirect effects.

Background: As generative AI becomes increasingly embedded in university learning, whether greater perceived usefulness of generative AI for learning is associated with greater generative AI dependency remains unclear, as do the potential statistical indirect roles of metacognitive self-regulation and academic procrastination. Methods: This cross-sectional questionnaire study surveyed 1,869 university students from three higher education institutions in Changde, Hunan Province, China. Eligible participants were aged 18 years or older and had used generative AI to assist learning within the past month. Pearson correlations were used to examine bivariate associations, and regression and PROCESS Model 6 analyses were conducted after controlling for sex, age, and grade to estimate specific and serial statistical indirect effects. Statistical indirect effects were tested using 5,000 bootstrap resamples and 95% confidence intervals (CIs). Results: Perceived usefulness of generative AI for learning was positively associated with generative AI dependency (total association: Conclusion: Higher perceived usefulness of generative AI for learning was statistically associated with higher generative AI dependency, with statistical indirect effects involving lower metacognitive self-regulation and higher academic procrastination. Given the cross-sectional design, PROCESS Model 6 was used to estimate specific and serial statistical indirect effects, which should not be interpreted as evidence that temporal ordering has been established or that causal mechanisms have been confirmed.