Exploring the Integration of Generative AI and Game-Based Learning in Mathematics Education: A Narrative Literature Review

This narrative literature review synthesises emerging evidence on the integration of generative artificial intelligence, particularly large language models such as ChatGPT, with game-based and gamified approaches in mathematics education. Nine empirical studies published in 2024-2025 were analysed to examine usage modalities, pedagogical affordances, and implications for learner engagement and educational equity. Three modalities of LLM integration emerged from the corpus: authoring tools for content generation, pedagogical agents providing scaffolding, and problem-solving environments enabling interactive simulations. The corpus reports statistically significant reductions in mathematics anxiety in two quasi-experimental studies and increased participation among students who tend to be less engaged in class, alongside persistent challenges that include scaffolding difficulties for students with knowledge gaps (with one study reporting zero task completion among low-performing learners), low authenticity rates in AI-generated problems (6.8% passing all quality criteria in the one study that systematically assessed them), and demographic skew in generated content. From the synthesis, three interpretive frameworks are developed as working hypotheses rather than validated models: a socio-emotional mediation framework positioning psychological safety as a candidate mechanism supporting learning, a dimensional authenticity framework distinguishing dimensions that may carry differential pedagogical weight, and a dual alignment framework articulating architectural directions for educational AI simulation. The review distinguishes access equity from outcomes equity, noting that the available evidence is more compatible with improvements in access than with assured outcomes for students with foundational knowledge gaps. Priority directions for future research include larger studies with adequately powered comparison groups, intervention work on adaptive scaffolding and interaction competencies, systematic bias auditing, and longitudinal designs examining learning persistence and transfer.