Based on the results from the Web of Science database, an overview of the trends in annual publication output in GAI-related medical education over the past 2 years is presented in Figure 1. In total, 281 relevant articles were found based on the above search method.

These results reveal a clear upward trajectory in accordance with the output of GAI-related medical education research over time. On the whole, the observed trends and general upward trajectory suggest that there will be ongoing growth in the GAI-based medical education technology in the future.

In previous studies (Lu, 2024), CiteSpace was employed to generate diagrams for different disciplines associated with the intelligent medical engineering field.

Based on the subject categories exported from the analysis of CiteSpace, there are 71 disciplines associated with the research of GAI-related medical education space, with 8 of these disciplines being associated with more than 10 publications, including the Health Care Sciences and Services, Education and Scientific Disciplines, General and Internal Medicine, Surgery, Medical Informatics, Education and Educational Research, Multidisciplinary Sciences, and Computer Science and Information Systems. The disciplines also include Nursing, Pharmacology and Pharmacy, Telecommunications, Emergency Medicine, Medical Ethics, and Transportation Science and Technology.

Based on the above analysis, the results underline the predominant role that medicine, education, and computer sciences play in this field, suggesting that computer science and medical education will continue to shape advances in GAI-related medical education research output in the future. On the whole, these results emphasize the multidisciplinary characteristics of the GAI-related medical education research.

Figure 2 shows the cluster analysis of categories based on title words. This cluster analysis reveals that the main directions included ChatGPT efficacy, remote lab, large language model performance, family medicine, broad-style examination questions, and generative artificial intelligence. These results emphasize the specific directions within the GAI-related medical education space, while also underscoring the importance of collaboration among various directions to propel the field forward.

Analyzing international cooperation in a given research field can clarify how relationships among countries have evolved and how they have shaped the research in the GAI-related medical education field. Based on the analyzed studies retrieved above, 65 countries/regions were found to contribute to GAI-related medical education research over the 2 years.

Based on the geographic distributions for the retrieved studies, the United States was identified as the most prominent contributor to the field with 105 publications. Other important contributors to this research field include China, England, Germany, and Canada, with 61, 18, 18, and 15 publications, respectively. These results emphasized the global nature of the GAI-related medical education field, suggesting that, while the United States and China remain the leaders in this field, there is also a broad global interest in ongoing, extensive collaborative research efforts.

Over the past 2 years, 106 institutions have been identified as having contributed to the GAI-related medical education field. The leading institutions include Harvard University, Harvard University Medical Affiliates, Harvard Medical School, the National University of Singapore, Sichuan University, Stanford University, the University of California System, the University of Texas System, the State University System of Florida, and Chongqing Medical University. According to the QS World University Rankings 2026, half of the top 10 universities have contributed to the GAI-related medical education field, and they include the Massachusetts Institute of Technology, Stanford University, Harvard University, ETH Zurich, and the National University of Singapore. The results further indicate that the GAI-related medical education is a research hotspot. Nodes that form the resultant network exhibit close connections, emphasizing the high degree of collaboration and interactivity among various institutions. A cluster analysis of these institutions is additionally performed based on subject disciplines (Figure 3).

This cluster analysis reveals that the main research directions included model evaluation study, scientific literature searches, clinical diagnostic process, ChatGPT's performance, and neurosurgical education. These results emphasize the expertise of particular institutions in specific directions within the overall GAI-related medical education space, while also underscoring the importance of collaboration among various directions and institutions to propel the field forward.

The journals with the highest numbers of GAI-related medical education research publications are analyzed. The results reveal that several prominent journals are responsible for publishing a considerable proportion of the studies in this field over the past 2 years.

JMIR Medical Education is the most prominent journal in accordance to GAI-related medical education research output with 148 publications, followed by the Cureus Journal of Medical Science, PLoS Digital Health, the Journal of Medical Internet Research, Nature, Healthcare-Basel, Medical Teacher, Radiology, Academic Medicine, and BMC Medical Education (101, 91, 83, 82, 82, 75, 67, 67, and 66 publications, respectively). Based on the Impact Factor (IF) of the 2025 Journal Citation Reports, the top ten journals are CA-A Cancer Journal for Clinicians (IF = 232.4), LANCET (IF = 88.5), New England Journal of Medicine (IF = 78.5), JAMA-Journal of The American Medical Association (IF = 55), Nature Medicinen (IF = 50), Nature (IF = 48.5), Science (IF = 45.8), Circulation (IF = 38.6), LANCET Infectious Diseases (IF = 31), ACM Computing Surveys (IF = 28). The widely distributed nature of these publications across journals emphasizes the importance of collaborations among researchers from various disciplines as a means of advancing efforts in the GAI-related medical education field.

Cluster analysis of journals is performed based on title words (Figure 4).

This cluster analysis reveals that the predominant subject categories for these journals include medical undergraduate, artificial intelligence chatbot, health care professional, generating multiple-choice questions, ChatGPT-driven quantitative analysis, neurosurgical education, medical licensing examination, professional identity formation, and patient education tool, highlighting the importance of multidimensional direction for the publication of studies in the GAI-related medical education.

To provide further insights into particular clusters of interest, CiteSpace can extract noun phrases from titles, keyword lists, or abstracts. To better understand the structure and homogeneity of the generated network, the resultant clusters can then be assessed based on their modularity (Q = 0.6169) and weighted mean silhouette (S = 0.8549) scores.

In this study, clusters are numbered based on descending cluster size order, such that the largest cluster containing (size = 30) is numbered #0 and is related to prompt engineering. Other top clusters in this analysis include #1 graduate medical education, #2 equity, #3 patient education, #4 medical service, #5 innovations, #6 assessment, #7 multiple-choice questions, #8 academic writing, and #9 knowledge. Cluster analysis is performed based on keywords (Figure 5).

These clusters allow for the effective categorization of research topics in the GAI-related medical education field, providing key insights into the major areas of research interest.

Timeline analyses can offer a visual representation of progress pertaining to particular research keywords and themes over time (Chen, 2017; Lu, 2024). Here, the timeline visualization reveals the profound evolution of GAI-related medical education research over the past 2 years (Figure 6).

Based on Figure 6, the development of GAI-related medical education can be categorized as follows:

The GAI has been applied to different stages of medical higher education, including undergraduate (Luke et al., 2024), master's (Li et al., 2024a), and doctoral education (Turchoie et al., 2024). Besides the students, the educational objects also contain medical and healthcare professionals (Patel et al., 2024), interns and residents (Lower et al., 2023), medical clinicians (Patil et al., 2024), and patients (Srinivasan et al., 2024). At the same time, the GAI has been integrated with different disciplines of medicine, including preclinical medicine (Luke et al., 2024), clinical medicine (Miao et al., 2024b), anesthesiology (Khan et al., 2024), biomedicine (Khosravi et al., 2024), medical imageology (Monroe et al., 2024), ophthalmology and optometry medicine (Ciekalski et al., 2024), psychiatry (Li et al., 2024b), radiation medicine (Pandey et al., 2024), pediatrics (Ramgopal et al., 2024), stomatology (Balel, 2023), preventive medicine (Kassab et al., 2024), traditional Chinese medicine (Li et al., 2024a), laboratory medicine (Meyer et al., 2024), nursing science (Gosak et al., 2024), rehabilitation therapy (Sivarajkumar et al., 2024), pharmacy (Pradhan et al., 2024), obstetrics (Riedel et al., 2023), emergency medicine (Liu et al., 2024). In clinical medicine education, the disciplines include nephrology (Miao et al., 2024b), neurosurgery (Arfaie et al., 2024), orthopedics (Vaishya et al., 2024), otolaryngology (Grimm et al., 2024), rheumatology (Madrid-García et al., 2023), cardiology (Madaudo et al., 2024), urology (Park et al., 2024), gastroenterology (Gravina et al., 2024), andrology (Ergin and Sanci, 2024). According to the above statistical results, the GAI has integrated with medical education.

In these applications, the main fields include medical exams (Zong et al., 2024; Danehy et al., 2024; Moglia et al., 2024; Lubitz and Latario, 2024), the performance assessment of GAI models (McGrath et al., 2024; Yamaguchi et al., 2024; Chen et al., 2024), the improvement of teaching methods and modes (Wojcik et al., 2024; Naamati-Schneider, 2024), and curriculum reform (Houssaini et al., 2024).

Medical exams play a significant role in improving the ability of medical professionals and students and contribute to the development of medical education in which the GAI models act as virtual teaching assistants and tutors (Zong et al., 2024). To pass these exams, a deep understanding of medical knowledge, clinical skills, and medical scenarios is needed. The medical exams cover various fields, including licensing examinations (Danehy et al., 2024), qualifying examinations (Vaishya et al., 2024), entrance exams (Khosravi et al., 2024), didactic tests (Moglia et al., 2024), and training examinations (Lubitz and Latario, 2024). These medical exams generally take the form of single-choice questions (Liu et al., 2024), multiple-choice questions (Khan et al., 2024; Vaishya et al., 2024), or a structured questionnaire (McGrath et al., 2024). Data sources of the medical exams are from medical textbooks and documents, medical contents of the web (Khan et al., 2024), question banks (Yang et al., 2024), question archive database (Ciekalski et al., 2024), and online platforms (Yamaguchi et al., 2024).

In these studies, the helpfulness for learning is one aspect of concern; on the other hand, the goal is to evaluate the performance of the GAI models. The GAI models can effectively understand medical knowledge and provide contextually relevant and appropriate responses (Oh et al., 2023; Li et al., 2024c; Sengar et al., 2025; Sikarwar et al., 2025; Kumar et al., 2023; Balakrishnan and Sengar, 2024; Sengar et al., 2023). However, it also has some limits, such as limited knowledge, inaccuracies, and the necessity for verification (Mu and He, 2024; Boscardin et al., 2024; Li et al., 2024c). Researchers have employed various technologies, such as prompt engineering, fine-tuning, and low-rank adaptation, to increase the performance of GAI models (Maitin et al., 2024). In these techniques, prompt engineering has gained increased attention (Mesko, 2023). A prompt (Liu et al., 2023) is a set of instructions provided to the GAI that programs the GAI by customizing it and/or enhancing or refining its capabilities. Prompt engineering is the means by which GAI is programmed via prompts and is employed to optimize the performance of GAI models (Maharjan et al., 2024).

The GAI models are helpful to enrich the teaching methods in medical education. According to research achievements, the GAI model is considered an assistance tool for academic writing, homework assignments, exam preparation (Wojcik et al., 2024), understanding of course knowledge and clinical information (Luke et al., 2024; Oh et al., 2023), case study (Bakkum et al., 2024), assessment of medical literature (Parente, 2024), and development of clinical skills (Ba et al., 2024). It can also simulate medical settings (Parente, 2024; Gosak et al., 2024), make narrative assessments of clinical learners, generate learning assessments, and create lesson plans (Meşe et al., 2024). Especially, some researchers investigated customized innovative GAI models to enhance medical education (Collins et al., 2024; Kiyak and Kononowicz, 2024; Huang et al., 2024).

The GAI models can promote innovation in teaching modes and improve educational outcomes. Naamati-Schneider (Naamati-Schneider, 2024) proposed a novel pedagogical framework that integrated problem-based learning (PBL) with the use of ChatGPT for undergraduate students. Divito et al. (Divito et al., 2024) addressed the factors of implementation and described how ChatGPT can be responsibly utilized to support key elements of PBL.

The curriculum is one of the most important parts of the medical education system. Additionally, the development of the curriculum is integrated with the GAI technique, merged with constructive alignment principles, the design thinking method, and GAI. Houssaini et al. (Houssaini et al., 2024) designed a new medical curriculum to guide educators in generating student-centered learning experiences. Huang and Lin (2024) proposed a GAI-based model to design a professional identity formation course. Turchoie et al. (2024) generated teaching activities to introduce GAI to students enrolled in a nursing data science and visualization course. Benboujja et al. (2024) employed GAI language models to produce a multilingual curriculum for online medical education.

Overall, this period reflects important ongoing efforts to explore and expand the GAI to meet medical education needs.

Subsequently, a keyword co-occurrence analysis is carried out, and a keyword co-occurrence network is obtained (Figure 7). The network consists of 161 nodes and 257 co-citation links for studies published from 2023 to 2024. The density for this keyword co-occurrence network is 0.02. The links among these keywords are labeling of their co-occurrence relationships, with thicker lines indicating closer relations (Lu, 2024).

In the generated keyword co-occurrence map, denser, thicker interactions are observed among certain keywords, including “artificial intelligence”, “large language model”, “ChatGPT”, “medical education”, “generative artificial intelligence”, “patient education”, and “natural language processing”, suggesting that these keywords are areas of notable interest within medical education. Furthermore, “machine learning”, “surgical education”, “medical exam”, “medical students”, “medical licensing examination”, “performance”, “health information”, and “public health” are identified as keywords that are closely linked to core keywords.

Keyword co-occurrence maps can provide detailed insight into important themes and relationships in the context of GAI-related medical education research, emphasizing important areas of study and opportunities for additional collaboration and development. In the keyword co-occurrence analysis, the two top fields in GAI-related medical education are identified as AI-related technologies and the research in medical education, accounting for 46% and 35%, respectively.

In the field of AI-related technologies, the five top keywords are identified as follows:

These keywords are all related to the AI discipline field, in which the GAI is one of the most advanced fields. In addition, the LLM and GAI are the outcomes of AI technologies, which encompass natural language processing, deep learning, and machine learning. The most frequently mentioned GAI models are ChatGPT and Bard from Google (Vaishya et al., 2024).

In the GAI-related medical education fields, the five top keywords are identified as follows:

These keywords are all related to medical education, in which the frequently mentioned fields include surgical education, patient education, nursing education, and continuing medical education. In the field of health education, researchers pay more attention to public health, digital health, health literacy, and health care. For the medical students, clinical skills training and medical knowledge are equally important. The topics that are frequently discussed include clinical decision-making, clinical reasoning, clinical practice, and clinical management. Moreover, the medical examination is one of the most effective teaching methods, and it is integrated with the GAI technology deeply. In addition, it includes multi-choice questions, the medical licensing examination, and the clinical informatics board examination. The other typical keywords also include curriculum development, learning outcomes, problem-based learning, educational technology, and interactive learning.

Overall, these results suggest that a large proportion of GAI-related medical education studies have focused on AI-related technologies and medical education. Their deep integration improves the development of these disciplines.

Although some research achievements are gained based on a detailed bibliometric analysis, there are some limitations in the article. First, only the Web of Science dataset is utilized to analyze in the investigation. Certainly, the dataset can be expanded to other datasets, including DOAJ, EBSCO, and Scopus. With the expansion of the database, the number of relevant articles will continue to increase, and the analysis results will be enriched. Second, the search strategy is designed using typical terms. However, many generative artificial intelligence models will be generated with the development of advanced technologies. With the emergence of different generative artificial intelligence models, the keywords and the analysis results will be enriched.

In summary, these results highlight the importance of future comprehensive, systematic research areas focused on multidisciplinary collaboration and the implementation of various methodologies and perspectives to study the GAI-related medical education. The research fields can include understanding the abilities of GAI models, integrating GAI technologies into medical education further, solving resistance from educators or students, and ensuring ethically responsible use of GAI. Further studies will also be needed to explore mutual relationships between different areas of the GAI-related medical education field. The additional elucidation of these relationships has the potential to offer insight into how best to improve the GAI-related medical education.