Mapping Three and a Half Decades of AI and Education Literature: Trends, Gaps, and Future Directions

Research on artificial intelligence (AI) in education has expanded at unprecedented speed, yet its development remains fragmented, hype-driven, and unevenly distributed across disciplines and regions. This paper presents one of the most comprehensive mappings to date, analysing 16,564 publications (1990–2025) identified through Web of Science and refined to 4,626 core contributions using a hybrid method that combined computational topic modelling, AI-assisted filtering, and human validation. To complement this macro-level mapping, we conducted a qualitative overview of the most cited papers in each major research topic, providing interpretive depth on how influential works have shaped debates on adoption, engagement, personalisation, chatbots, evaluation, risks, monitoring, reviews, and teaching AI. The findings reveal exponential growth since 2020, with major surges during the COVID-19 pandemic and following the release of ChatGPT, but also enduring fragmentation, declining evaluation studies, and a saturation of attitudinal surveys. Promising areas such as personalisation, teaching AI, and engagement have lost visibility despite their long-term importance, while risks and ethical issues remain underexplored and poorly connected to technical research. Citation patterns further show systemic imbalances, with education papers disproportionately cited and engineering outputs under-recognised. We conclude with recommendations to strengthen interdisciplinarity, address neglected areas, include historical perspectives, and address systemic challenges in order to foster a more cumulative, integrative, and critical field.