Unpacking GenAI-enabled deep learning engagement: role perceptions, human–GenAI synergy strategies, and underlying mechanisms

Background: Generative AI is rapidly transforming how students learn, yet its impact on deep learning varies significantly among individuals. This study investigates the mechanisms linking students' role perception of GenAI, their human-GenAI synergy strategies, and deep learning engagement. Methods: A mixed-methods design was employed to analyze data from 181 undergraduate students in a blended course, integrating qualitative coding of reflection reports with quantitative path analysis and latent profile analysis. Results: Human-GenAI synergy strategies fully mediate the relationship between role perception and deep learning engagement. A critical cognitive threshold was identified where high-order metacognitive engagement emerges primarily when students perceive GenAI as a thinking collaborator rather than a basic efficiency tool. Latent profile analysis revealed six heterogeneous learner profiles, exposing a skill-cognition decoupling phenomenon where complex tool manipulation does not necessarily translate to deep cognitive engagement. Furthermore, an assessment paradox was uncovered, where highly engaged exploratory students often received lower traditional academic scores compared to pragmatic, tool-dependent users. Discussion: It is concluded that fostering genuine deep learning engagement in the digital era requires shifting pedagogical focus from technical skill training to reshaping students' GenAI role perceptions and updating assessment paradigms to recognize and reward process-oriented human-GenAI metacognitive synergy.