Cognitive Outsourcing in Generative AI-Supported Learning: A Behavioural-Sequence Analysis of Metacognitive Differentiation

Metacognitive ability is a core component of learners’ cognitive capacity, directly shaping knowledge construction, problem-solving, and intellectual development. Its cultivation requires learners to reflect on, adjust, and evaluate their own learning strategies. Yet generative artificial intelligence (GAI) increasingly performs tasks once undertaken by learners themselves—reflection, error detection, and correction—potentially diminishing motivation for deep thinking, weakening self-regulated learning, and ultimately constraining long-term metacognitive growth. Drawing on a one-year teaching experiment with 90 graduate students in education (organised into 18 groups and analysed at the group level), this study examined how GAI-assisted learning affects learners’ metacognitive abilities. Behavioural-sequence analysis revealed that, although overall metacognitive engagement improved significantly across two semesters, performance-group trajectories diverged sharply: high-performance learners sustained metacognitive optimization, whereas low-performance learners exhibited patterns consistent with metacognitive abandonment and cognitive outsourcing. These findings indicate that the benefits of GAI-assisted learning are contingent on pedagogically balanced implementation and on learners’ pre-existing metacognitive capacity.