Analysing metacognitive differences in AI-assisted maths learning: an ENA study of reflection and scaffolding

Generative artificial intelligence offers new opportunities for adaptive scaffolding in mathematics learning, yet its role in students’ metacognitive processes remains underexplored, particularly across mathematics anxiety levels. This study used Epistemic Network Analysis to examine reflections from 24 high school students before and after a 4-week AI-supported inquiry-based scaffolding intervention. Results showed reconfigured reflective discourse involving positive metacognitive experience, strategy monitoring/evaluation, and AI support. Metacognitive knowledge was expressed less independently and appeared embedded in monitoring and evaluation, while strategy planning remained stable, indicating a reflection–action gap. Exploratory patterns suggested that low-anxiety students linked knowledge more closely with monitoring, whereas high-anxiety students initially used AI support as an affective resource. These findings support anxiety-aware and planning-oriented AI scaffolding in mathematics.