The Post-ChatGPT Surge: Does AI-Health-Education Research Growth Reflect Evidence or Enthusiasm
TL;DR
Thematic analysis revealed that in 2024, ChatGPT and other large language models started to dominate over machine learning, which previously accounted for the majority of publications, which suggest a possible gap between the quantitative expansion of the literature and the consolidation of the field.
Abstract
Artificial intelligence (AI) research in health education has grown exponentially since the public release of ChatGPT in November 2022. However, questions remain regarding whether this growth represents a mature scholarly field or simply a self-feeding loop of publication growth. This bibliometric study examines the relationship between the increase in publications, citation impacts, and the thematic reorientation towards generative artificial intelligence after the launch of ChatGPT. Bibliographic records were extracted from Web of Science, Scopus, and PubMed and combined into one dataset comprising 4,358 publications after removing duplicates to address the limitation of the previous bibliometric analysis (based on only one bibliographic database). The R packages bibliometrix and biblioshiny were used to visualise and analyse the bibliometric data. Annual publications and citations, distribution of sources and authors, and thematic structure based on Author’s Keywords (with keyword-field cleaning to eliminate indexing inconsistencies and combine related terms) were explored. The number of publications increased from 228 in 2022 to 1,550 in 2025, with an average annual growth of 56.48%. The growth peak was observed roughly one year after the release of ChatGPT, not right away, from 2023 to 2024. The average citation impact per document peaked in 2023, and it dropped steadily in all the following years, both for raw and time-normalised citation impact, with citation-impact concentration remaining locked onto the year immediately following ChatGPT's release. Thematic analysis revealed that in 2024, ChatGPT and other large language models started to dominate over machine learning, which previously accounted for the majority of publications. By employing factorial analysis, another cluster emerged from the documents, which associated particular Generative AI tools with measures of information quality (accuracy, readability). AI in health education research since ChatGPT's appearance may not only exhibit characteristics of either the evidence-generating or the enthusiastic type but also demonstrates tendencies suggesting that the former currently outweighs the latter. The results suggest a possible gap between the quantitative expansion of the literature and the consolidation of the field. The article explores the implications of these findings for research, funding, and health-education practice and outlines several productive directions for future bibliometric analysis.