Abstract:

Artificial intelligence (AI) has undergone a significant technological shift with the invention of Large Language Models (LLMs). In education, ChatGPT has been met with both optimism and skepticism. Initial research mostly took the form of discussion and observation, with empirical studies following later. Some of the most effective uses identified were in problem-based learning (PBL) and intelligent tutoring of moderate duration. Meta-analyses found a major positive impact on learning performance and cognitive engagement, with moderate increases in learning perception, higher-order thinking, and behavioural and emotional engagement. Negative effects included the dangers of user over-reliance, system bias, and threats to academic integrity, such as plagiarism.

The aim of our study was to determine the extent and nature of the most recent research on the use of ChatGPT in educational contexts. Due to the large volume of publications, we chose a bibliometric approach, beginning with a search of the Web of Science database for ChatGPT in titles and education-related terms within titles or keywords, limited to journal articles and conference papers.

The resulting 2,798 documents were analysed using Microsoft Excel, Bibliometrix/Biblioshiny, and VOSviewer in terms of:
1) performance – production and citations (by year, document, reference, source, reference source, author, institution, country, and keyword); and
2) science mapping – network analyses (co-word on keywords, co-citation on reference sources, and co-authorship on countries).

The results show that while the number of publications in 2024 tripled compared to 2023, growth was only moderate in 2025, with documents from 2023 being the most cited. Among document types, only 12% were conference papers. The most cited documents and references address AI adoption in general education and medical education. The most articles were published in the journal Education and Information Technologies, while Education Sciences had the highest citations per document. The most productive authors from Hong Kong, Taiwan, and Turkey study the use of AI in English, STEM, and medical education. The Education University of Hong Kong produced the most documents. Among countries, the United States has the most publications and collaborations, with the strongest ties to China. Meanwhile, Australia has the highest citations per document. Two thirds of documents included ChatGPT as a keyword. Among education-related keywords, the most frequent were education, higher education, medical education, patient education, and students.

Co-word analysis on keywords revealed four clusters by education areas and research topics:
1) general, language, STEM, K12 (study of performance, perceptions, critical thinking, motivation);
2) medicine, computing (assessment, prompt engineering, accuracy, ethics, decision-making, AI literacy);
3) higher education, technology (technology acceptance, academic integrity, attitudes, satisfaction);
4) patient, health (readability, quality, health literacy).

Our research updates and complements existing analyses of ChatGPT use in education. We found some research topics to be underrepresented in current literature, indicating a need for further study, such as multimodal inputs, pre-college levels. We also recommend greater focus on assessment, anti-plagiarism, and AI literacy.

Keywords:

Generative AI, ChatGPT, education, students, teaching, bibliometric analysis.