From functionality to performance: How AI-enabled engagement and collaboration shape mathematics learning outcomes

Understanding the role of conversational artificial intelligence like ChatGPT in enhancing mathematics learning is critical for modern education. Grounded in Bandura's Social Cognitive Theory (SCT), this study investigates how ChatGPT's functionality fosters a transformative environment that influences behavioral and personal outcomes in mathematics learning. A cross-sectional survey design was used in this research among 836 school students in Kathmandu, Nepal. Data were analyzed using structural equation modeling and regularized linear regression to test a SCT-mediated model. The results demonstrate that ChatGPT's functionality, as an environmental factor, has a strong positive influence on collaboration and engagement as behavioral factors, which in turn significantly enhance learning performance as a personal factor, with effect sizes of β = 0.33 for collaboration predicting performance and β = 0.31 for engagement predicting performance. The model explains a substantial 70% of the variance in mathematics performance, confirming the crucial mediating roles of collaboration and engagement as parallel mediators. Furthermore, students' ChatGPT competency, a direct proxy for digital self-efficacy, was a significant predictor of engagement (β = 0.18, p = 0.00). The indirect effects of functionality on performance through engagement and functionality on performance through collaboration were both statistically significant, supporting partial mediation. These findings provide robust empirical support for SCT's triadic reciprocity model in ChatGPT-enabled learning, demonstrating that its impact is not direct but mediated by fostering collaborative mathematical discourse and engaged participation. Theoretically, this study advances SCT by integrating AI literacy as a moderator and extending Human-AI Collaboration Theory to mathematics education. Practically, it recommends that educators prioritize AI literacy training, design collaborative learning activities around ChatGPT-generated outputs, and implement structured usage routines to maximize learning gains while mitigating risks of superficial dependency.