Modeling learning engagement in AI-supported education: a crossover enabler–inhibitor framework

Artificial intelligence is reshaping how learners interact with content, feedback, and motivation. Yet the psychological processes that support or disrupt engagement in AI-powered environments remain underexplored. This study develops a crossover model to explain how enabling beliefs and emotional strain jointly influence learning engagement. Drawing from self-determination, cognitive load, and social cognitive theories, the model tests how performance expectancy and technology self-efficacy function alongside feedback overload and AI learning anxiety. Learning motivation and AI fatigue are positioned as mediators linking these factors to behavioral engagement. Data were collected from 251 learners with experience using AI-driven educational tools. Structural equation modeling revealed that learning motivation is the most powerful predictor of engagement. Performance expectancy increases motivation, while feedback overload weakens it. Technology self-efficacy has no significant effect on either motivation or fatigue. Fatigue, although linked to anxiety and overload, did not directly reduce engagement. These results suggest that motivation, not depletion alone, drives sustained learning behavior. This study contributes by reframing inhibitors as threats to motivation rather than isolated stressors. It offers practical insights for designing AI systems that preserve psychological energy, minimize cognitive strain, and reinforce learning purpose. Engagement is not automatic. It must be carefully engineered within the emotional architecture of intelligent systems. For AI-system designers, this means embedding motivation-aware dashboards, adaptive feedback controls, and anxiety monitoring features to protect and sustain learner engagement.