["Journal of Computer Assisted Learning, Volume 42, Issue 3, June 2026. ", "\nABSTRACT\n\nBackground\nSince late 2022, the integration of generative AI, particularly ChatGPT, has rapidly expanded in mathematics education, creating a dynamic but heterogeneous research field that requires systematic synthesis regarding its pedagogical value.\n\n\nObjective\nThis review analyses scientific literature on ChatGPT in mathematics education to identify its applications, benefits, limitations and instructional projections.\n\n\nMethods\nFollowing PRISMA guidelines, 54 peer‐reviewed articles from Web of Science, Scopus and SciELO were analysed using deductive‐inductive content analysis across methodological, pedagogical and impact dimensions.\n\n\nResults\nAnalysis revealed methodological diversity, with predominance of qualitative approaches (33.3%) and concentration in higher education (35.2%) and teacher training (24.1%). Main applications included problem‐solving support (46.3%), instructional material design (22.2%) and metacognitive scaffolding (14.8%). Benefits encompassed conceptual understanding enhancement (22.2%), metacognitive development (20.4%) and teacher professional growth (20.4%). Critical concerns centred on accuracy and reliability (46.3%), necessitating explicit verification protocols, AI literacy development (31.5%) and strong teacher mediation (27.8%).\n\n\nConclusion\nChatGPT's educational value is contingent upon instructional design quality rather than tool capabilities. Effective integration demands AI literacy, structured tasks promoting mathematical validation, and pedagogical mediation that preserves disciplinary rigour and student autonomy.\n\n"]