The numbers tell a striking story. In 2023, the field produced 92 publications, or roughly 33 percent of the total corpus. In 2024, that figure jumped to 187 documents, accounting for just over 67 percent. A doubling of annual output within a year is rare in academia, where research fields typically mature over decades, and it signals both the urgency and the unsettled nature of the debate over AI in teaching. The corpus was dominated by research articles, which made up 214 of the 279 papers, followed by 53 early-access items, 32 conference proceedings, 27 reviews, and a small handful of editorial materials.

What distinguishes this analysis from earlier attempts to chart the ChatGPT-in-education landscape is its methodology. Previous reviews, the authors argue, were often based on small, preliminary datasets, focused narrowly on the tool’s earliest emergence phase, or relied on descriptive statistics such as publication counts and lists of leading authors without deeper science-mapping techniques. By contrast, the new study integrates PRISMA-based screening with advanced bibliometric mapping using VOSviewer and the bibliometrix package in RStudio, producing co-authorship networks, keyword co-occurrence maps, thematic quadrants, and three-field plots that connect authors, countries, institutions, journals, and keywords in a single analytical framework. The combination, the researchers say, makes the study both transparent and reproducible in a way earlier surveys were not.

The screening process itself was demanding. An initial Web of Science search using the keywords ChatGPT and generative AI in combination with education and higher education returned 893 records. After removing duplicates and applying exclusion criteria, including 603 records that fell outside educational research, language, and linguistics fields, ten non-English publications, and one paper outside the 2023 to October 2024 window, the team retained 279 articles. All retrieved records were exported to Microsoft Excel for organization and counting, while Zotero managed citations. The analysis was completed in October 2024, capturing the field’s first twenty months of existence.

The geography of the field is heavily concentrated. Researchers from 68 countries contributed, but the United States led decisively with 67 publications, followed by China with 34, Australia with 25, the United Kingdom with 18, Turkey with 15, and India with 14. Of 791 identified authors, D. Henriksen and P. Mishra topped the productivity rankings with eight publications each, followed by A. Bozkurt and A. Strzelecki with four apiece. Institutional leadership clustered at Arizona State University and Tecnológico de Monterrey, each among the most productive centers in the world. Co-authorship network analysis in VOSviewer revealed 38 of 71 countries meeting a threshold of at least three documents, forming eight distinct collaboration clusters, a sign that despite the field’s youth, international research partnerships have already crystallized.

The output is concentrated in a small set of venues. The 279 papers appeared across 116 journals, but just 15 journals published 130 of them, or nearly 47 percent. Education Sciences led with 21 papers, followed by Education and Information Technologies with 18 and Frontiers in Education with 15. Measured by citations rather than volume, Smart Learning Environments ranked first with 359 citations, ahead of Education Sciences with 315, the International Journal of Management Education with 292, and Contemporary Educational Technology with 265. The fifteen most-cited papers in the corpus collectively accumulated nearly 1,924 citations, an extraordinary rate for a literature barely two years old.

At the level of ideas, the thematic mapping reveals both maturity and immaturity. Using a keyword frequency threshold of 57 occurrences, refined through repeated testing to cut visual noise, the team distilled 849 keywords down to 57 meaningful items grouped into six clusters. These clusters span academic integrity, ethics, assessment, machine learning and generative AI on one side, and technology acceptance, critical thinking, higher education, and teacher education on another. A thematic quadrant analysis classified motor themes such as acceptance, adoption, and information technologies as both central and well developed, while basic themes like artificial intelligence and students were central but conceptually underdeveloped. Niche themes, including pedagogical content knowledge and TPACK, and emerging themes such as plagiarism and framework design, remain on the field’s periphery, suggesting where the next wave of research is likely to land.

The most cited work crystallizes the field’s central tension. Tlili and colleagues’ paper, What if the devil is my guardian angel: ChatGPT as a case study of using chatbots in education, leads with 349 citations, raising concerns about cheating, privacy, and manipulative behavior. Farrokhnia and colleagues’ SWOT analysis of ChatGPT follows with 230 citations, cataloging both the tool’s capacity for plausible, personalized, real-time responses and its tendency to produce incorrect information. Cooper’s exploratory study of generative AI in science education ranks third with 222 citations. Across the top-cited literature, the benefits recur consistently: rapid idea generation, interactive learning environments, immediate feedback, and reduced grading workload, particularly valuable in large classes where individualized feedback is impractical. The drawbacks are equally consistent: factual errors, erosion of critical thinking, and the thorny problem of plagiarism.

That last problem has grown technically harder. The analysis notes that AI-powered paraphrasing tools can reduce plagiarism detection scores to acceptable levels, making traditional plagiarism checkers increasingly unreliable, a finding that will resonate with educators confronting the limits of academic integrity enforcement. The literature also warns of a subtler danger: students and even researchers frequently perceive ChatGPT as an authoritative source of information, risking uncritical reliance on machine-generated content. Studies cited in the review found that ChatGPT can reduce the mental effort required for tasks such as narrative scriptwriting, and evidence from language education suggests possible effects on thinking skills and creativity. At the same time, comparative research has found AI-generated feedback can be more descriptive than peer feedback, complementing rather than replacing it, and the tool has proven effective in designing interactive content and enhancing feedback mechanisms across disciplines.

The field’s intellectual trajectory is itself a finding. In 2023, research emphasized theoretical questions: whether ChatGPT was safe for students, how AI was perceived in educational contexts, and what responsible implementation strategies might look like. By 2024, the focus had shifted decisively toward empirical application, examining ChatGPT as an assessment tool, a source of writing assistance, and a support for tasks including literature reviews. The authors interpret the dominance of adoption and acceptance themes as support for the Technology Acceptance Model, with perceived usefulness and ease of use remaining key drivers of integration. The field also remains overwhelmingly centered on higher education, with education and educational research accounting for 97 percent of output by discipline, leaving K-12 applications, ethical solutions rather than ethical warnings, and non-English contexts as conspicuous gaps. The study’s own limitations, including reliance on a single database and English-language sources, underscore how much of this global transformation remains unmapped, even as the doubling of the literature makes clear that educators, researchers, and institutions can no longer treat generative AI as a passing fad.

Subject of Research: Bibliometric mapping of global research trends on ChatGPT in education

Article Title: Bibliometric analysis of ChatGPT in education

Article References: Sabirova, G., Zharkynbekova, S., Kannazarova, Z., Mamatkulov, K., Goyibova, N., Usmanova, S., & Shosaidova, L. (2026). Bibliometric analysis of ChatGPT in education. Discover Education, 5(1), Article 984. https://doi.org/10.1007/s44217-026-02006-7

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02006-7

Keywords: ChatGPT, generative AI, education, bibliometric analysis, PRISMA, Web of Science, higher education, academic integrity, technology acceptance, VOSviewer, research trends, plagiarism