Differentiated Impacts of GenAI-supported Self-Regulated Game-Based Science Learning: a Behavioural and Epistemic Network Analysis of High- and Low-Achieving Students

Integrating Generative Artificial Intelligence (GenAI) into digital game-based learning (DGBL) environments presents new opportunities for enhancing student learning through personalised support, immediate feedback, and adaptive interaction. While previous research has explored the effectiveness of GenAI in various educational contexts, limited studies have examined how GenAI can support students’ self-regulated learning (SRL) within game-based science learning environments. Addressing this gap, the present study aimed to explore the impact of GenAI-supported self-regulated digital game-based learning (GenAI-SRDGBL) on junior high school students with different academic achievement levels, focusing on their learning motivation, behavioural patterns, perceptions, and prompt usage in a physics course. Using a quasi-experimental design, 48 students were categorised into high and low achievers based on their academic achievement. Data were collected through achievement tests, learning motivation questionnaires, in-game activity logs, student illustrations, and student-GenAI prompt interactions, and were analysed using statistical methods and Epistemic Network Analysis. High-achieving students reported higher post-intervention intrinsic motivation and used GenAI more strategically to seek advanced information and connect learning resources, whereas low-achieving students showed more repetitive GenAI access and relied heavily on basic information and formula prompts. The groups did not differ significantly in extrinsic motivation or illustrated perceptions. These findings highlight the need for differentiated, scaffolded GenAI support to foster effective self-regulation and to prevent metacognitive overreliance in game-based learning.