When Engagement Does Not Translate Into Achievement: Multilevel Evidence From AI-Supported Higher Education in Vietnam

Artificial intelligence has become an integral component of higher education, creating new opportunities for learning while raising important questions about how student engagement contributes to academic performance across different educational contexts. This study examines relationships among AI self-efficacy, cognitive offloading, student engagement, teacher support, and academic performance using a multilevel structural model based on data collected from 1,392 undergraduate students nested within 54 classes in Vietnam. The findings indicate that student engagement is positively associated with academic performance at the individual level, whereas the corresponding relationship is not supported at the classroom level. AI self-efficacy is associated with stronger engagement and lower reliance on cognitive offloading, while teacher support demonstrates stronger associations with classroom engagement and academic performance than shared engagement. A multilevel perspective reveals that individual learning processes and instructional context contribute to academic performance through different pathways rather than through a single mechanism. This study advances current understanding of AI-supported learning by demonstrating that engagement operates differently across analytical levels and by highlighting the importance of integrating cognitive regulation with instructional support when designing AI-supported higher education environments.