Predicting perceived learning effectiveness from generative artificial intelligence in higher education: a...

This study examines predictors of students' perceived learning effectiveness from generative artificial intelligence (GenAI) in higher education by integrating psychological, behavioral, and cognitive perspectives. An explanatory sequential mixed-methods design was employed. Quantitative data were collected through a survey of 658 higher education students and analyzed using hierarchical multiple regression to examine the statistical contribution of GenAI usage patterns, perceived critical-thinking change, trust in GenAI, attitudes toward GenAI, self-efficacy for learning with GenAI, and cognitive offloading, while controlling for GPA, instructor policy toward GenAI, language of study, access type, and frequency of use. The results showed that, in the theoretically specified order of entry, the psychological block-comprising self-efficacy, attitudes, and trust-produced the largest incremental increase in explained variance. Trust remained positively associated with perceived learning effectiveness, whereas cognitive offloading was not statistically significant in the final model. Contextual variables showed limited explanatory power after psychological, behavioral, and cognitive variables were introduced. The final model explained 68.8% of the variance in students' perceived learning effectiveness from GenAI. To explain these findings, qualitative interviews were conducted with 11 students. Thematic analysis indicated that students perceived GenAI as most beneficial when used intentionally and reflectively for explanation, idea generation, and revision. Students with higher self-efficacy described more strategic engagement, including iterative prompting and verification of outputs, while excessive reliance on GenAI was perceived as potentially reducing cognitive engagement. Overall, the findings suggest that students' perceived learning effectiveness from GenAI is associated with psychological readiness, behavioral usage patterns, and perceived cognitive engagement, while contextual control variables exhibit comparatively limited explanatory power. The study provides theoretical and practical implications and directions for future research.