Mitigating cognitive offloading in GenAI-supported collaboration: a metacognitive scaffolding approach to enhancing students’ critical thinking

While Generative Artificial Intelligence (GenAI) has transformative potential for collaborative learning, it may also induce cognitive offloading and superficial consensus, leading learners to accept AI-generated outputs uncritically. To address these risks, this study developed metacognitive scaffolds for triadic “Learner–Peer–GenAI” interaction to foster critical thinking. An eight-week quasi-experimental intervention was conducted with 85 sixth-grade students, comparing an experimental group receiving scaffolds (n = 42) with a control group without scaffolding (n = 43). Using a multi-method design involving pre/post tests, SOLO-based cognitive-structure analysis, Epistemic Network Analysis (ENA), and Lag Sequential Analysis (LSA), we examined changes in critical-thinking performance, cognitive structure, and interaction trajectories. Results showed that the experimental group significantly outperformed the control group in overall critical thinking and three dimensions: questioning and verification, logical reasoning, and reflection and regulation, with the largest gains in questioning and verification. High-order cognitive structures accounted for 49.7% of the experimental group’s discourse, indicating a shift toward higher-level thinking. Behavioral analyses identified a dominant sequence—Questioning (QB.AI)→Evidence Retrieval (ER.AI)→Logical Analysis (LA.AI)—and a more cohesive epistemic network, whereas the control group showed surface-level exchange and passive agreement. These findings suggest that metacognitive scaffolding can regulate GenAI-supported collaboration, preserve learner agency, and promote cognitive reciprocity.