Generate-Then-Submit: How Cognitive Load Disrupts College Students’ Self-Regulated Learning Cycle in AI-Assisted Learning

Generative AI in higher education has given rise to a “generate-then-submit” behaviour, whereby students bypass reflection and submit AI-generated answers directly. The cognitive mechanisms underlying this abandonment of self-regulated learning (SRL) reflection remain unclear. Integrating SRL theory and cognitive load theory (CLT), this study constructs and tests an SRL-CLT model using survey data from 1,109 undergraduates across seven Chinese universities, analysed via structural equation modelling. Results reveal two parallel post-performance pathways that explain how cognitive load shapes students’ learning outcomes. A beneficial pathway suggests that germane cognitive load promotes self-reflection, thereby reducing direct submission. A maladaptive pathway shows that extraneous cognitive load increases direct submission without impairing reflection resources. Together, these findings advance CLT by demonstrating that extraneous load disrupts learning primarily through behavioural triggering—direct submission behavior—rather than through cognitive interference. This refined mechanism offers a novel theoretical account of the “generate-then-submit” phenomenon in AI-assisted learning. Based on these insights, a three-level intervention framework (tool optimization, pedagogical scaffolding, and individual cultivation) is proposed to transform the “generate-then-submit” cycle into a deep learning process of “generate-reflect-revise.”