Generative AI in Mathematics Learning: A Bibliometric Analysis and Narrative Review

Since late 2022, generative artificial intelligence (GenAI) has expanded rapidly in mathematics education, yet the research field remains insufficiently mapped, and evidence of its effects on learning is inconsistent. This study mapped the development, structure, and emerging topics of research on GenAI in mathematics learning and synthesized evidence on its uses, effects, and challenges. Bibliometric analysis was combined with a narrative review. Searches for records published between 2022 and 2026 in OpenAlex, ERIC, ScienceDirect, Google Scholar, and Garuda yielded 1,480 records, with 1,206 unique records retained after cleaning. Descriptive and text-mining analyses examined the full corpus, while citation and keyword co-occurrence analyses were conducted in VOSviewer using 758 OpenAlex records. The findings show that 95.2% of records were published from 2024 onward, 48.6% mentioned ChatGPT, and 29.5% explicitly addressed primary or secondary education. The keyword network comprised five clusters, with AI literacy, academic integrity, and self-regulation emerging as the newest topics. The narrative synthesis indicates that GenAI can support learning when integrated into scaffolded instruction with safeguards, whereas unrestricted use as an answer source can undermine learning. These findings highlight instructional design as central to the educational value of GenAI and inform the extension of scaffolding perspectives to AI-mediated mathematics learning. The study identifies school-level experimental research and teacher preparation as priorities for advancing evidence and guiding GenAI integration.