Generative AI in Education: Mapping the Research Landscape Through Bibliometric Analysis

Abstract

The rapid emergence of generative AI technologies has sparked significant transformation across educational landscapes worldwide. This study presents a comprehensive bibliometric analysis of GAI in education, mapping scholarly trends from 2022 to 2025. Drawing on 3808 peer-reviewed journal articles indexed in Scopus, the analysis reveals exponential growth in publications, with dominant contributions from the United States, China, and Hong Kong. Using VOSviewer, the study identifies six major thematic clusters, including GAI in higher education, ethics, technological foundations, writing support, and assessment. Prominent tools, especially ChatGPT, are shown to influence pedagogical design, academic integrity, and learner engagement. The study highlights interdisciplinary integration, regional research ecosystems, and evolving keyword patterns reflecting the field’s transition from tool-based inquiry to learner-centered concerns. This review offers strategic insights for educators, researchers, and policymakers navigating AI’s transformative role in education.

1. Introduction

Artificial intelligence (AI) refers to the field of computer science concerned with creating systems that simulate aspects of human intelligence, such as problem-solving, learning, reasoning, and perception [1]. Generative AI (GAI) is a subset of AI that focuses on the creation of new content, such as text, images, code, or other media, using learned patterns from existing data [2]. It includes models, like Generative Adversarial Networks (GANs), Generative Pre-trained Transformer (GPT) models, and Generative Diffusion Models (GDMs), which generate human-like outputs in response to prompts [3].

In recent years, the advent of GAI technologies has marked a profound paradigm shift across a broad spectrum of sectors, ranging from healthcare and finance to entertainment and education [4]. GAI, encompassing technologies that can autonomously produce text, images, code, music, and even synthetic voices, has rapidly evolved from a novel research area to a transformative force with wide-reaching implications [3,4,5]. These technologies rely on complex neural network architectures, particularly transformer-based models, to generate responses based on patterns and relationships identified in large-scale datasets used during training. These outputs do not stem from original cognition or understanding but rather from a synthesis of information embedded in the training data. As such, GAI responses reflect the scope, limitations, and biases of the data on which they were trained.

Among the most influential innovations in this domain are Large-Scale Language Models (LLMs), such as OpenAI’s ChatGPT, Google’s Gemini (formerly Bard), Meta’s LLaMA, and Anthropic’s Claude, which have significantly expanded the capabilities of AI in natural language understanding and generation [6]. The exponential growth of these models in scale and sophistication has not only elevated AI research but has also catalyzed a new wave of technological adoption and experimentation in pedagogical environments [7].

Education, by its very nature, is a dynamic and multifaceted domain that continuously adapts to technological advancements [2]. In recent years, GAI has emerged as both a catalyst and a challenge within this evolving landscape [8]. Educational institutions globally are beginning to explore and integrate these tools into instructional design, assessment strategies, learner support services, and administrative operations [9,10]. From generating personalized learning pathways and real-time tutoring support to enabling multimodal content creation and automating feedback provision, the potential of GAI to enhance educational equity, scalability, and efficacy is widely acknowledged [11]. Tools, such as ChatGPT, have been utilized to summarize complex readings, craft lesson plans, simulate interactive discussions, and provide writing assistance, thereby extending pedagogical resources in both formal and informal settings [12]. Additionally, generative image models, like DALL·E and Midjourney, are being incorporated into creative arts education, while coding assistants, like GitHub Copilot, support programming instruction [13].

However, the integration of GAI in education also presents a host of challenges and tensions. Ethical concerns about data privacy, surveillance, and algorithmic biases are increasingly pressing as these tools become embedded in educational infrastructures [14]. Moreover, issues of academic integrity have surfaced as students gain access to AI-powered writing and problem-solving tools that can easily circumvent traditional evaluation systems [15,16]. There are also pedagogical concerns about overreliance on AI, the risk of diminishing human agency in learning, and the need for both educators and learners to acquire new literacies to navigate AI-enhanced environments effectively [17]. These concerns have led some institutions to take formal action. For instance, in 2023, the New York City Department of Education initially restricted access to ChatGPT across its public schools due to concerns about misuse and a lack of clarity around educational value [18]. Similarly, SciencesPo in France issued a temporary ban on the use of ChatGPT for coursework, citing worries about academic dishonesty and undermining pedagogical aims [19]. Although some of these policies have since evolved, such examples underscore the institutional hesitation and governance dilemmas posed by rapidly advancing GAI tools. The rapid pace of innovation often outstrips institutional capacity to adapt policies, pedagogies, and curricula, creating a sense of urgency for scholarly inquiry and informed guidance [16].

Given these multifaceted implications, there is a growing necessity to systematically map and understand the research landscape surrounding GAI in education. Bibliometric analysis offers a robust and data-driven approach to achieving this objective. By employing quantitative techniques to analyze publication metadata, such as authorship, affiliations, keywords, citations, etc., bibliometric methods enable researchers to uncover the intellectual foundations, emerging trends, and collaboration networks within a field [20]. This macroscopic view complements more traditional narrative and systematic reviews by providing empirical evidence of research activity, impact, and evolution.

This study aims to conduct a comprehensive bibliometric review of the intersection between GAI and education. Specifically, we seek to:

• Analyze publication trends and productivity patterns across journals, authors, institutions, and countries.
• Identify prominent research themes and their evolution over time through keyword co-occurrence analysis.
• Map co-authorship and citation networks to understand collaboration and influence structures.
• Identify key gaps and propose future research directions.

Through this systematic exploration, we intend to provide an evidence-based roadmap for educators, policymakers, and researchers navigating the integration of GAI into educational systems. In doing so, we contribute to a more nuanced understanding of how emerging technologies can reshape teaching and learning in the 21st century.

2. Literature Review

The connection between AI and education is not new. Since the 1980s, AI has been employed in Intelligent Tutoring Systems (ITS), adaptive learning platforms, and automated grading tools [21]. Early AI applications focused on rule-based expert systems designed to simulate the decision-making of human educators [22]. These systems paved the way for adaptive learning environments capable of responding to individual learners’ needs in real-time [23].

Over the decades, advances in machine learning and data analytics have led to the proliferation of GAI, which can create new content by learning patterns from existing data [2]. Notably, GANs and GPTs has dramatically improved the contextual understanding and generation capabilities of AI systems, making them highly suitable for language-intensive domains, like education [24].

GAI offers numerous educational affordances. Instructors are exploring its use to develop syllabi, generate feedback, and create test questions [25]. Learners use it to seek explanations, draft essays, and brainstorm ideas. Its ability to simulate human-like conversation has sparked interest in roles, such as virtual tutors or peer learners [11]. Additionally, GAI supports differentiated instruction and inclusive education. Multilingual students and learners with disabilities can benefit from content tailored to their linguistic or cognitive needs [26]. Visual learners may engage more effectively with image-generating tools, like DALL·E, while students learning to code can receive personalized assistance from platforms, such as GitHub Copilot [13]. These applications hold promise for narrowing educational inequities when implemented with equity-focused strategies.

However, the integration of GAI into education raises important pedagogical concerns. Its use in assessment tasks prompts questions about how to maintain the development of critical thinking and originality [2]. Students may become overly dependent on AI tools, limiting their practice of essential academic skills, such as critical reading, synthesis, and argumentation [27]. To address these challenges, educators must scaffold AI-supported learning in ways that foster metacognition and originality. Some institutions are beginning to pilot frameworks that treat AI as a collaborative partner rather than a substitute for student effort, emphasizing transparency and reflective use [28]. Finally, the emergence of GAI necessitates a rethinking of learning outcomes and assessment formats [2]. Traditional summative assessments may not be adequate in contexts where students can leverage AI to generate high-quality responses. Alternatives, such as open-book exams, project-based learning, peer assessment, and reflective journals, are gaining traction as more authentic indicators of student understanding in AI-enhanced learning environments [29].

Furthermore, the integration of GAI into education is fraught with ethical dilemmas. Key issues include data privacy, misinformation, and the potential for plagiarism. Students can use AI to generate essays or assignments that evade traditional plagiarism detectors, challenging existing academic integrity frameworks [30]. Moreover, generative models can reproduce biases present in their training data, potentially perpetuating stereotypes or inaccuracies [31].

Transparency and explainability are critical ethical dimensions [32]. GAI often functions as a ‘black box’ with outputs that appear intelligent but lack interpretability [33]. When students and educators cannot fully understand how AI reaches its conclusions, the risk of misuse and over-trust increases. There is an urgent need for AI tools in education to include mechanisms for clarifying how content is generated and what data sources underpin the outputs [32].

Another concern involves surveillance and data usage. Educational applications of AI often require tracking user behavior and personal data to tailor outputs. This raises questions about consent, data ownership, and the long-term consequences of educational datafication [34]. Institutions must establish clear data governance policies that safeguard learner privacy [14].

In response, scholars and practitioners are advocating for the inclusion of AI ethics and digital responsibility within curricula [17]. AI literacy should not only encompass technical proficiency but also critical awareness of AI’s social and ethical implications [35]. This includes recognizing algorithmic bias, questioning the neutrality of AI-generated content, and understanding the ethical use of generative tools [35].

Educators also need to reconsider their roles in upholding academic integrity [30]. Rather than relying solely on detection tools, institutions are exploring alternative strategies, such as honor codes, AI usage disclosures, and formative assessments, that emphasize learning processes [29]. These measures aim to cultivate ethical reasoning and a sense of accountability among students navigating AI-rich academic environments [36].

Building on this foundation, a growing body of systematic reviews and meta-analyses has explored the application of artificial intelligence in education, reflecting the increasing maturity of this research domain. These include broad reviews of AI applications in educational settings [37,38,39,40], analyses of theoretical and methodological gaps [41], and syntheses of ethical implications and instructional potential [42,43]. These studies provide critical overviews of AI’s integration in pedagogy, assessment, adaptive learning, and administrative functions.

However, these reviews predominantly focus on general or historical forms of AI, such as rule-based systems, machine learning classifiers, or adaptive learning technologies. The more recent wave of GAI, such as LLMs, like GPT-4, introduces new capabilities and challenges that are only beginning to receive systematic academic attention. For instance, Mao et al. [44] offers an important focused review on GAI’s implications for assessment, signaling a growing interest in this subfield. However, there remains a need to comprehensively map the research terrain surrounding GAI as a distinct category, particularly in terms of scholarly trends, intellectual structure, and global collaboration.

This bibliometric study addresses that gap by providing a focused and data-driven synthesis of the scholarship at the intersection of GAI and education. It highlights not only research productivity and thematic structure but also the characteristics of scholarly contributions and intellectual influences in this emerging field.

3. Methodology

This bibliometric review adopts a systematic and quantitative approach to analyze the research landscape at the intersection of GAI and education. The methodology comprises four main components, namely data query, inclusion/exclusion criteria, and bibliometric analyses.

3.1. Data Querying

Scopus and Web of Science (WoS) are widely regarded as the two leading academic databases for bibliometric studies due to their extensive coverage of peer-reviewed literature and consistent metadata quality [45]. Since Scopus and WoS have been shown to exhibit a high degree of overlap, using both databases is generally not expected to significantly alter overall bibliometric patterns, unless literature from highly specific niches is targeted. Therefore, we opted for Scopus based on its wider journal inclusion in the social sciences and its compatibility with VOSviewer, the bibliometric tool used in this study.

The data retrieval process was conducted on 4 May 2025, ensuring that the dataset reflects the most current research at the time of analysis. To capture the breadth of literature related to GAI and its applications in education, a targeted search query was constructed using Boolean logic and truncated keyword variations. The following search string was used: [TITLE-ABS-KEY (“generative artificial intelligence” OR “GAI” OR “large language model” OR “LLM” OR “ChatGPT” OR “artificial intelligence generated content” OR “AI-generated content”)] AND [TITLE-ABS-KEY (“teach” OR “educat*” OR “learn*” OR “universit*” OR “college” OR “school”].

This query was designed to capture scholarly articles addressing the educational implications, uses, and consequences of GAI technologies, including widely known tools, such as ChatGPT and other LLMs. The wildcard terms “teach*,” “educat*,” and “learn*” ensured coverage of institutional, personal, and instructional perspectives. The initial search yielded 8229 documents.

3.2. Inclusion and Exclusion Criteria

To ensure the relevance and academic rigor of the dataset, the following criteria were applied during the article selection process:

• Timeframe: Only articles published between 2022 and 2025 were included. This timeframe corresponds to the emergence and proliferation of GAI technologies in mainstream academic and educational discourse, particularly following the release of GPT-3 and GPT-4.
• Source Type: The dataset was limited to peer-reviewed journals to ensure the quality and credibility of the findings. Conference papers, editorials, reviews, book chapters, and grey literature were excluded.
• Document Type: Only original research articles were selected for analysis. The decision to exclude review articles was made to focus the analysis on empirical and conceptual studies that actively contribute new findings or frameworks to the field.

The dataset was cleaned and de-duplicated before proceeding to bibliometric analysis. Finally, a total of 3808 articles were retained for bibliometric analysis.

3.3. Bibliometric Analysis

Bibliometric analysis involves the quantitative examination of scholarly publications using mathematical and statistical techniques [46]. In this study, two complementary statistical approaches were employed to investigate the research landscape.

First, descriptive statistics were compiled to assess the productivity and impact of key bibliometric entities, including documents, journals, authors, institutions, and countries. Performance metrics, such as the number of publications, citation counts, publication years, source distribution, etc., were organized and summarized using Excel spreadsheets.

Second, relational analysis was conducted using VOSviewer 1.6.20 to generate network visualizations that illustrate the scientific structure of the field [47]. Specifically, keyword co-occurrence analysis was used to map thematic areas and identify emerging research trends based on author-supplied keywords. Meanwhile, document co-citation analysis was employed to uncover the intellectual foundations of the field by identifying seminal and highly influential works.

4. Results

4.1. Overview

Figure 1 shows the number of publications on GAI and education over the past four years. In 2022, the publication count was relatively modest because GAI had initially emerged. However, starting in 2023, the field experienced a marked acceleration in output, with the number of publications suddenly jumping to 465. The upward trend continued into 2024; the number of publications further increased to 1951, more than four times that in the previous year. The 2025 data, while appearing lower, only include records retrieved up to 4 May and should therefore be interpreted as partial-year data. This caveat is important when observing temporal trends.

The distribution of publications across subject areas reveals the multidisciplinary nature of the field and highlights the leading academic domains (

Table 1. Subject areas of articles on GAI and education research.

Rank (nth)	Subject Area 1	Number of Papers 2	%
1	Social Sciences	2512	66.0
2	Computer Science	1259	33.1
3	Medicine	540	14.2
4	Engineering	374	9.8
5	Arts and Humanities	343	9.0
6	Psychology	248	6.5
7	Business, Management and Accounting	208	5.5
8	Health Professions	139	3.7
9	Mathematics	126	3.3
10	Multidisciplinary	102	2.7

¹ Only the subject areas that have at least 100 articles are shown; ² The total count is larger than the total number of articles because some articles belong to more than one subject area.

). The productive journals on GAI and education paints a coherent picture of a broadly interdisciplinary yet socially grounded field (

Table 2. Journals published articles on GAI and education research.

Rank (nth)	Journal 1	Articles	Citescore	SJR 2024
1	Education and Information Technologies	152	11.8	1.65 Q1
2	Computers and Education: Artificial Intelligence	99	28.7	5.22 Q1
3	Education Sciences	69	5.5	0.73 Q1
4	JMIR Medical Education	50	11.0	1.97 Q1
5	IEEE Access	46	9.0	0.85 Q1
6 and 7	Frontiers in Education	39	2.9	0.65 Q2
	Interactive Learning Environments	39	13.8	1.98 Q1
8	Journal of Applied Learning and Teaching	37	10.4	1.76 Q1
9 and 10	Applied Sciences (Switzerland)	36	5.5	0.521 Q2
	IEEE Transactions on Learning Technologies	36	7.2	1.01 Q1
11	BMC Medical Education	33	4.4	0.95 Q1
12 and 13	Journal of Chemical Education	32	4.7	0.6 Q2
	Sustainability (Switzerland)	32	7.7	0.69 Q1
14	British Journal of Educational Technology	30	17.6	2.69 Q1
15	TechTrends	28	6.8	1.09 Q1
16 and 17	International Journal of Educational Technology in Higher Education	27	27.7	3.91 Q1
	Scientific Reports	27	6.7	0.87 Q1
18 to 20	European Journal of Education	24	3.6	0.98 Q1
	Medical Teacher	24	6.5	1.25 Q1
	PLoS One	24	5.4	0.8 Q1
21	International Journal of Artificial Intelligence in Education	23	15.4	1.96 Q1
22	Journal of University Teaching and Learning Practice	22	6.2	0.91 Q1
23 and 24	European Public and Social Innovation Review	21	0.1	0.17 Q3
	International Journal of Learning Teaching and Educational Research	21	2.3	0.32 Q3
25	Cogent Education	20	2.9	0.6 Q2

¹ Only the journals that have at least 20 articles are shown.

).

Social Sciences (2512 articles, 66.0%) dominate the research landscape, with Education and Information Technologies (152 articles), Education Sciences (69 articles), British Journal of Educational Technology (30 articles), Journal of Applied Learning and Teaching (37 articles), TechTrends (28 articles), International Journal of Educational Technology in Higher Education (27 articles), European Journal of Education (24 articles), International Journal of Artificial Intelligence in Education (23 articles), Journal of University Teaching and Learning Practice (22 articles), European Public and Social Innovation Review (21 articles), International Journal of Learning Teaching and Educational Research (21 articles), and Cogent Education (20 articles) as the primary outlets in this subject area. These journals are interested in the human-centered dimensions of AI integration, such as ethical challenges (e.g., plagiarism, cheating), teacher preparedness, and student engagement, indicating that GAI facilitates a fundamental transformation of education norms, teaching practices, assessment models, student engagement, and institutional policy.

Computer Science (1259 articles, 33.1%) and Engineering (374 articles, 9.8%) represent the technical basis that supplies the architecture and computational methods on which educational applications are built, underscoring the essential role of interdisciplinary collaboration between AI developers and education researchers. Productive journals, such as Computers and Education: Artificial Intelligence (99 articles), Interactive Learning Environments (39 articles), IEEE Access (46 articles), IEEE Transactions on Learning Technologies (36 articles), and IEEE Access (36 articles), represent the engineering contributions.

Medicine accounts for 8.1% (540 articles), suggesting the growing adoption of AI tools in medical and health professional training contexts. This is likely due to the structured, assessment-heavy nature of medical education, which makes it a fertile ground for early adoption and rigorous evaluation of GAI tools, especially in case-based learning, exam preparation, and patient education [48]. Productive journals in this subject area are JMIR Medical Education (50 articles), BMC Medical Education (33 articles), and Medical Teacher (24 articles).

Less represented but still meaningful contributions come from Arts and Humanities (343 articles, 9.0%), Psychology (248 articles, 6.5%), Business, Management and Accounting (208 articles, 5.5%), Health Professions (139 articles, 3.7%), and Mathematics (126 articles, 3.3%), reflecting specific niches and applications of GAI in education. The category Multidisciplinary (102 articles, 2.7%) encompass a broad set of published works with overlapping or emerging disciplinary identities, represented by a cluster of multidisciplinary and open-access journals, such as Frontiers in Education (39 articles), Applied Sciences (Switzerland) (36 articles) Sustainability (Switzerland) (32 articles), Scientific Reports (27 articles), and PLOS One (24 articles). These journals accommodate cross-sectoral studies that explore the role of AI in broader educational reform, digital equity, and sustainable learning.

4.2. Three Levels of Contributions

Figure 2 and

Table 3. Productive institutes on GAI and education research.

Rank (nth)	Institute 1	Country	Articles	QS World Ranking 2025
1	The Education University of Hong Kong	Hong Kong	51	12 *
2	The University of Hong Kong	Hong Kong	43	17
3	Monash University	Australia	40	37
4	Chinese University of Hong Kong	Hong Kong	38	36
5	Beijing Normal University	China	34	271
6	University of Toronto	Canada	30	25
7 and 8	Nanyang Technological University	Singapore	29	15
	East China Normal University	China	29	501
9 and 10	Harvard University	USA	25	4
	King Saud University	Saudi Arabia	25	201
11	Purdue University	USA	24	91
12	The Hong Kong Polytechnic University	Hong Kong	23	57
13 to 17	Universiti Sains Malaysia	Malaysia	22	146
	Arizona State University	USA	22	200
	King Abdulaziz University	Saudi Arabia	22	149
	Stanford University	USA	22	6
	University of Florida	USA	22	216
18 and 19	Tecnológico de Monterrey	Mexico	21	185
	National University of Singapore	Singapore	21	8

¹ Only the institutes that have at least 20 articles are shown. * Subject ranking (education).

and

Table 4. Productive authors on GAI and education research.

Rank (nth)	Author 1	Institute	Country	Articles	h-Index 2024
1	Chiu, T.K.F.	Chinese University of Hong Kong	Hong Kong	13	38
2 to 4	Gašević, D.	Monash University	Australia	11	35
	Hwang, G.J.	National Taichung University of Education	Taiwan	11	90
	Zou, D.	The Hong Kong Polytechnic University	Hong Kong	11	36
5	Mishra, P.	Arizona State University	USA	10	29
6 to 8	Annamalai, N.	Universiti Sains Malaysia	Malaysia	9	15
	Henriksen, D.	Arizona State University	USA	9	22
	Kohnke, L.	The Education University of Hong Kong	Hong Kong	9	21
9 to 13	Al-Emran, M.	The British University in Dubai	United Arab Emirates	8	50
	Gu, X.	East China Normal University	China	8	23
	Lee, H.-Y.	National Yunlin University of Science and Technology	Taiwan	8	12
	Moorhouse, B.L.	Hong Kong Baptist University	Hong Kong	8	24
	Rudolph, J.	Research and Learning Innovation	Singapore	8	13
14 to 20	Bozkurt, A.	Anadolu University	Turkey	7	29
	Khlaif, Z.N.	An-Najah National University	Palestine	7	16
	Strzelecki, A.	University of Economics in Katowice	Poland	7	17
	Tan, S.	Kaplan Higher Education Academy	Singapore	7	13
	Tlili, A.	Beijing Normal University	China	7	25
	Tu, Y.F.	National Taiwan University of Science and Technology	Taiwan	7	19
	Wu, T.T.	National Taiwan University of Science and Technology	Taiwan	7	28

¹ Only the authors that have at least 7 articles are shown.

show the national, institutional, and individual contributions to the field, respectively. Taken together, these three levels of analysis reveal how the field is shaped by both geopolitical research ecosystems and individual academic leadership, with visible patterns of regional specialization and scholarly concentration, thus providing a rich and nuanced understanding of the research landscape in GAI and education.

The United States leads by a substantial margin with 935 articles, affirming its dominant position in research on both AI and education. It is also home to prolific authors, like P. Mishra (10 articles) and D. Henriksen (9 articles), and key institutions, such as Harvard University (25 articles), Purdue University (23 articles), Arizona State University (22 articles), University of Floride (22 articles), and Stanford University (22 articles), each of which appears in the top tier of institutional contributors.

China ranks second with 508 articles. Its contributions are reinforced by institutions, like Beijing Normal University and East China Normal University, and notable scholars, such as A. Tilili and X. Gu.

The United Kingdom (237 articles) and Australia (225 articles) follow as key Commonwealth contributors. Australia is home to leading author D. Gašević (11 articles) of Monash University (40 articles), a researcher known for bridging learning analytics and AI pedagogy.

Saudi Arabia (168 articles), Turkey (119 articles), and United Arab Emirates (107 articles) are leading countries on GAI and education in the Gulf region and the Middle East. Notably, King Saud University (25 articles) and King Abdulaziz University (22 articles) are institutional leaders, while M. Al-Emran (8 articles) of The British University in Dubai and Z.N. Khlaif (7 articles) of An-Najah National University represent some of the region’s most prolific authors in this field.

An interesting phenomenon is the significant concentration of scholarly output in Hong Kong (147 articles), comparable to Spain (163 articles) and Germany (153 articles) but higher than India (142 articles). Three institutions from Hong Kong, namely The Education University of Hong Kong (51 articles), The University of Hong Kong (43 articles), and The Chinese University of Hong Kong (38 articles), rank among the top five institutes in this field. This density reflects Hong Kong’s strategic investment in AI-enhanced education and is further underscored by prolific scholars, such as T.K.F. Chiu (13 articles), D. Zou (11 articles), L. Kohnke (9 articles), and B.L. Moorhouse (8 articles).

Other high-performing countries are mainly from Asia, such as South Korea (129 articles), Taiwan (104 articles), Malaysia (95 articles), Indonesia (74 articles), Singapore (174 articles), and Japan (70 articles). Significant institutes include Nanyang Technological University (29 articles), Universiti Sains Malaysia (22 articles), and National University of Singapore (21 articles), while prolific authors are G.J. Hwang (11 articles), N. Annamalai (95 articles), H.-Y. Lee (8 articles), J. Rudolph (8 articles), S. Tan (7 articles), Y.F. Tu (7 articles), and T.T. Wu (7 articles).

4.3. Keywords

The analysis of frequently occurring keywords highlights the key topics driving the literature on GAI and education. Overall, these keywords reflect the breadth of the research field (

Table 5. Frequently occurring keywords on GAI and education research.

Rank (nth)	Keyword 1	Occurrence
1	Artificial Intelligence (AI)	1803
2	ChatGPT	1686
3	Generative Artificial Intelligence (GAI)	973
4	Large Language Model (LLM)	739
5	Higher Education	555
6	Education	478
7	Machine Learning	382
8	Students	368
9	Medical Education	298
10	Language Model	286
11	Chatbot	272
12	Teaching	251
13	Learning	219
14	Contrastive Learning	203
15	Procedures	203
16	Natural Language Processing (NLP)	172
17	Federated Learning	118
18	Academic Integrity	116
19	Educational Technology	110
20 and 21	Educational Measurement	101
	Medical Student	101
22	Engineering Education	100

¹ Only the keywords that have at least 100 occurrences are shown.

). The top four most frequently occurring keywords are “AI” (1803 occurrences), “ChatGPT” (1686 occurrences), “GAI” (973 occurrences), and “LLM” (739 occurrences), reflecting that GAI and ChatGPT are central themes in current scholarly discussions on the applications of GAI to education.

The frequent mention of “Higher Education” (555 occurrences), education (478 occurrences), students (368 occurrences), teaching (251 occurrences), learning (219 occurrences), contrastive learning (203 occurrences), federated learning (118 occurrences), educational technology (110 occurrences), and educational measurement (101 occurrences), further underscores the educational focus of the corpus.

Notably, domain-specific applications, such as medical education (298 occurrences), medical student (101 occurrences), and engineering education (100 occurrences) appear prominently, indicating interdisciplinary exploration across academic and professional training fields.

To explore the associations between the keywords on GAI and education, a co-occurrence analysis was conducted on 65 keywords that appeared at least 65 times across the dataset. These keywords were grouped into five distinct clusters (Figure 3).

The red cluster includes 26 keywords, encompassing a broad array of keywords, such as GAI, ChatGPT, AI in education, AI literacy, digital literacy, technology acceptance, technology adoption, educational technology, higher education, university students, academic performance, and academic writing. This cluster can be aptly titled “GAI in Higher Education”, as it represents a substantial body of research centered on the integration of GAI tools within university contexts. The focus lies in examining how these technologies influence learning outcomes, academic productivity, and students’ digital competencies. The inclusion of terms, like engineering education, nursing education, and programming education, points to emerging, discipline-specific applications of AI in technical and professional training contexts. Additionally, keywords, such as attitude, motivation, self-efficacy, technology acceptance model, and student engagement, reveal a parallel interest in the psychological dimensions of AI use, particularly how students perceive, adopt, and are influenced by these tools in their learning processes.

In contrast, the green cluster, best described as “Technological Foundations and Linguistic Interfaces”, revolves around keywords, such as deep learning, machine learning, GPT, LLM, NLP, AI, OpenAI, chatbot, and readability. This cluster positions the AI tools themselves as the primary object of inquiry, examining what they are capable of, how they generate and communicate language, and where their limitations emerge. Emphasis is placed on the technological infrastructure behind GAI, including platforms, like OpenAI and Bard, and the functioning of LLMs that enable these systems to process, produce, and engage with human language in chatbot form.

The blue cluster, centered around keywords, such as AI tools, language learning, EFL, writing, and feedback, can be aptly labeled “AI and Writing.” These keywords highlight the role of AI as a powerful language assistant, capable of enhancing writing fluency, providing targeted feedback, and supporting both collaborative and personalized learning experiences, particularly in second-language education. The inclusion of terms, like medical education and medical students, reflects the strong emphasis on language and communication skills in medical training, where students must produce clinical documentation, communicate effectively with patients, and articulate diagnostic reasoning with precision.

The yellow cluster, composed of keywords, such as education, creativity, ethics, technology, and students, reflects a broader, cross-cutting theme “Critical Perspectives on Technology and Society”. This cluster interrogates the societal, psychological, and philosophical dimensions of AI in education. The inclusion of creativity and learning also hints at deeper questions about what it means to learn or create in an age of algorithmic assistance.

The purple cluster, organized around terms, like academic integrity, plagiarism, prompt engineering, and AI ethics, may be defined as “Integrity and Ethics”. Closely related to the yellow cluster, but narrower in scope, this cluster centers on questions of reliability, transparency, and responsible use of AI tools. It addresses how prompts shape outcomes (i.e., prompt engineering), how AI-generated content can be misused (e.g., plagiarism), and what constitutes ethical boundaries in AI-supported environments. The emphasis is less on technological innovation and more on regulation, trust, and accountability.

Finally, the cyan cluster, with terms, such as assessment, science education, teacher education, and teaching and learning, forms a smaller thematic group best labeled “Assessment”. This cluster points to the discussions on how GAI affects the assessment of teaching and learning.

The temporal overlay visualization of keyword co-occurrence provides a dynamic view on the changes in research priorities with time (Figure 4). By assigning colors to keywords based on their average year of occurrence, with purple and blue indicating earlier usage and green and yellow signifying more recent attention. This color gradient enables us to track shifts in research priorities over time.

Early interests, as indicated by the prevalence of blue and purple keywords, were dominated by technologically foundational and instructional topics. Keywords, such as machine learning, deep learning, NLP, GPT, and Bard, indicate an initial emphasis on the core technologies underpinning GAI systems. Alongside these, educational terms, like teaching, learning, and students, highlight that early research was largely focused on traditional instructional settings, where researchers began to explore how AI could enhance or challenge conventional learning processes. Notably, terms, such as academic integrity and plagiarism, also emerged during this period, underscoring early concerns about the ethical implications of AI-generated content and its potential to undermine assessment credibility.

The intermediate phase, transitioning into green keywords, shows a transition toward application-oriented and pedagogically contextualized research. Keywords, such as AI in education, higher education, LLM, ChatGPT, and GAI, signal growing attention to how specific AI tools and platforms are being integrated into teaching and learning environments. The inclusion of terms, like educational technology, pedagogy, language learning, EFL, and AI literacy reflects a broader interest in discipline-specific adoption and teacher–learner preparedness. This stage also introduced digital literacy, academic performance, and prompt engineering, indicating a diversification of research into both skills required to use AI effectively and evaluation of educational outcomes tied to AI usage.

The latest topics are captured by yellow keywords, demonstrating a marked shift toward student-centered experiences, behavioral responses, and participatory engagement. Emerging keywords include university students, student perceptions, perceptions, technology adoption, and feedback, reflecting a surge of interest in how learners interpret and adapt to AI-enhanced environments. The appearance of academic writing suggests renewed focus on how GAI intersects with student expression, originality, and scholarly communication. Other terms, such as science education, collaborative learning, and Gemini (i.e., the advanced successor to Bard), highlight a push toward specific content domains and novel platforms, indicating that researchers are expanding the field’s scope to explore diverse learning contexts and tools.

4.4. Seminal and Foundation Works

The most cited articles highlight the scholarly impact of seminal works in the field (

Table 6. Highly cited articles on GAI and education research.

Rank (nth)	Articles 1	Title	Year	Journal	Citations
1	Dwivedi et al. [49]	“So what if ChatGPT wrote it?” Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy	2023	International Journal of Information Management	2111
2	Gilson et al. [48]	How does ChatGPT perform on the United States medical licensing examination? The Implications of large language models for medical education and knowledge assessment	2023	JMIR Medical Education	1140
3	Rudolph et al. [50]	ChatGPT: Bullshit spewer or the end of traditional assessments in higher education?	2023	Journal of Applied Learning and Teaching	921
4	Cotton et al. [30]	Chatting and cheating: Ensuring academic integrity in the era of ChatGPT	2024	Innovations in Education and Teaching International	916
5	Chan & Hu [51]	Students’ voices on generative AI: perceptions, benefits, and challenges in higher education	2023	International Journal of Educational Technology in Higher Education	561
6	Pavlik [52]	Collaborating with ChatGPT: Considering the implications of generative artificial intelligence for journalism and media education	2023	Journalism and Mass Communication Educator	557
7	Farrokhnia et al. [53]	A SWOT analysis of ChatGPT: Implications for educational practice and research	2024	Innovations in Education and Teaching International	506
8	Rahman & Watanobe [54]	ChatGPT for education and research: Opportunities, threats, and strategies	2023	Applied Sciences (Switzerland)	506
9	Chan [55]	A comprehensive AI policy education framework for university teaching and learning	2023	International Journal of Educational Technology in Higher Education	487
10	Rudolph et al. [56]	War of the chatbots: Bard, Bing Chat, ChatGPT, Ernie and beyond. The new AI gold rush and its impact on higher education	2023	Journal of Applied Learning and Teaching	421
11	Khan et al. [57]	ChatGPT-Reshaping medical education and clinical management	2023	Pakistan Journal of Medical Sciences	395
12	Sullivan et al. [58]	ChatGPT in higher education: Considerations for academic integrity and student learning	2023	Journal of Applied Learning and Teaching	382
13	Perkins [59]	Academic Integrity considerations of AI large language models in the post-pandemic era: ChatGPT and beyond	2023	Journal of University Teaching and Learning Practice	375
14	Dergaa et al. [60]	From human writing to artificial intelligence generated text: examining the prospects and potential threats of ChatGPT in academic writing	2023	Biology of Sport	353
15	Yeo et al. [61]	Assessing the performance of ChatGPT in answer- ing questions regarding cirrhosis and hepatocellu- lar carcinoma	2023	Clinical and Molecular Hepatology	351
16	Noy & Zhang [62]	Experimental evidence on the productivity effects of generative artificial intelligence	2023	Science	348
17	Crawford et al. [63]	Leadership is needed for ethical ChatGPT: Character, assessment, and learning using artificial intelligence (AI)	2023	Journal of University Teaching and Learning Practice	324
18	Michel-Villarreal et al. [64]	Challenges and opportunities of generative AI for higher education as explained by ChatGPT	2023	Education Sciences	317
19	Yan [65]	Impact of ChatGPT on learners in a L2 writing practicum: An exploratory investigation	2023	Education and Information Technologies	313
20	Kamalov et al. [66]	New era of artificial intelligence in education: towards a sustainable multifaceted revolution	2023	Sustainability (Switzerland)	305

¹ Only the articles that have at least 300 citations are shown.

). Leading the list is the article by Dwivedi et al. [49] (2111 citations), which offers a sweeping overview of the challenges and opportunities posed by GAI across domains. Its rapid uptake suggests that the academic community well accepts the discourse that frames the integration of GAI in knowledge production and education. Similarly, Gilson et al. [48] achieves remarkable impact with 1140 citations for their analysis of ChatGPT’s performance on the U.S. Medical Licensing Examination, highlighting the disruptive potential of LLMs in medical training and evaluation.

A key theme among the highly cited works is the concern for academic integrity and assessment. For instance, Rudolph et al. [50] (921 citations) critically explores how GAI challenges traditional assessment models, while Cotton et al. [30] (916 citations) delves into ethical dilemmas around cheating and plagiarism facilitated by AI tools. Notably, the authors disclosed that the article itself was co-written with the assistance of ChatGPT based on the GPT-3 model, making it both a subject and an example of AI’s impact on academic authorship. Studies, like Crawford et al. [63] (324 citations) and Michel-Villareal et al. [64] (317 citations), touch on the need for ethical leadership and institutional readiness, reinforcing the importance of strategic and moral frameworks in navigating this technological transition. These contributions urge educational institutions to rethink assessment design and uphold academic standards amid AI advancements.

The role of student perceptions and experiences is also well-represented. Chan and Hu [51] (561 citations) explores student voices on the use of GAI in higher education, shedding light on perceived benefits and concerns. Likewise, Farrokhnia et al. [54] (506 citations) captures the strengths, weaknesses, opportunities, and threats of ChatGPT in teaching and learning contexts, pointing to the complexity and multifaceted impact of such tools in educational ecosystems.

The growing focus on policy and governance is illustrated by Chan [55] (487 citations) comprehensive AI policy framework, which has become influential in discussions on institutional responses to AI integration. Meanwhile, Rudolph et al. [56] (421 citations) draws attention to the competitive landscape and varied capabilities of AI tools, like Bing Chat, Ernie, and ChatGPT.

Medical and clinical education emerges as a specialized yet impactful area of interest. For example, Khan et al. [57] (395 citations) addresses how ChatGPT is reshaping medical education and clinical decision-making, and Yeo et al. [61] (351 citations) assesses ChatGPT’s performance in answering clinical questions. On the other hand, Noy and Zhang [62] (348 citations) examine AI’s productivity effects in a broader workplace context, demonstrating the cross-sectoral influence of GAI research.

To explore the intellectual structure underpinning the field, co-citation analysis was conducted on 49 articles using 40 citations as the classification threshold, and five clusters were identified (Figure 5). The red cluster encompasses the strategic inquiries on the transformative impact of GAI (especially ChatGPT) on traditional education paradigms, and reflections on the potentials and constraints. For example, Kasneci et al. [67] offers a balanced discussion on the opportunities and challenges that large language models present to educators and learners. Grassini [68] articulates the promise of AI in reshaping pedagogy, advocating for a reimagined educational future grounded in technological advancement. While Rudolph et al. [50] reviews how ChatGPT is changing learning, teaching, and assessment in higher education, Lo [69] examines ChatGPT’s capabilities across subject domains in the context of education. These articles portray ChatGPT not merely as a tool but as a catalyst for educational transformation.

In contrast, the green cluster encompasses the concrete applications of GAI, with a focus on language education and writing support. This cluster highlights how ChatGPT can be used to enhance learners’ language proficiency, engagement, and writing capabilities. For example, Kohnke et al. [70] discusses the use of ChatGPT in language teaching and learning. Su et al. [71] explores the possibilities of ChatGPT in assisting students to improve their writing skills. Barrot [72] examines the technology’s capacity to scaffold writing instruction for English as a Second Language (ESL) students, while acknowledging the need for pedagogical safeguards. Meanwhile, Minutomo and Eguchi [73] explore the use of ChatGPT for automated essay scoring, pointing toward a future where AI assists in both instruction and assessment.

The blue cluster is characterized by its meta-analytical and conceptual orientation, aiming to map, critique, and guide the integration of GAI in education. Common among these documents is a focus on systematic reflection, such as Bahrourn et al. [74], who uses bibliometric and content analysis to chart the research landscape. While Chen et al. [38] provides an overview of AI applications in education, Celik et al. [75] reviews AI’s impact on teacher roles. Several works propose conceptual frameworks to reinterpret educational practices, such as Mishra et al. [76] updating TPACK to include GAI and Lim et al. [77] framing the debate around transformation versus disruption.

The yellow cluster is grounded in behavioral models or theories that provide the conceptual scaffolding for empirical studies on AI in education. Core models, such as the Theory of Planned Behavior [78] and the Technology Acceptance Model [79,80] underpin investigations into user intent and trust in ChatGPT. Measurement frameworks like structural equation modeling support robust data analysis [81,82].

Lastly, the purple cluster, although smaller and more peripheral, appears to represent domain-specific implementations and technical evaluations. These works interrogate the tension between the tool’s perceived utility and its potential to generate false or unverifiable content, especially in scientific and medical contexts [83,84].

5. Discussion

5.1. Research Landscape of GAI and Education

The dramatic increase in scholarly publications on GAI in education from 2022 to 2025 marks a paradigm shift in educational research and practice. The exponential increase from a modest number in 2022 to over 2000 articles in 2025 not only reflects the technological disruption brought by GAI but also signals growing institutional and academic commitment to understanding its educational implications. This rapid trajectory highlights not just intensified scholarly interest but also a recognition of the field’s relevance to pressing scientific, technological, and societal challenges [48].

Several factors may account for this growth. First, the maturation of LLMs, like ChatGPT, Bard, and Claude, have moved them from experimental tools to widely adopted educational technologies [57]. Second, expanded funding opportunities and supportive policy initiatives have likely stimulated research in this area. Third, interdisciplinary collaborations and advances in research methodologies have enabled scholars to explore novel questions and produce impactful findings. Finally, the proliferation of academic journals and open-access platforms has accelerated the dissemination of research outcomes.

The trajectory in publications not only reflects the rising scholarly momentum but also underscores the multidisciplinary nature of the topic. Subject area analysis reveals a strong representation from the Social Sciences, followed by Computer Science and Medicine. This distribution suggests a dual focus: one on how GAI is transforming pedagogy, assessment, and institutional governance; another on understanding the underlying technical architectures enabling these changes [9]. The notable presence of Medicine also indicates domain-specific enthusiasm, especially in high-stakes and structured learning contexts where AI supports simulations, diagnostics, and formative assessments [49].

Moreover, the diversity of contributing journals underscores both the broad academic reach and societal significance of this emerging field. While many leading journals are grounded in either social science or technical disciplines, numerous education-specific and interdisciplinary outlets increasingly call for reflective engagement with the pedagogical and epistemological implications of GAI, the bibliometric trends identified in this study suggest that much of the published output focuses on mapping trends, challenges, and adoption patterns rather than advancing interpretive or critical perspectives. More interpretive, theory-driven, or critical work is still needed to advance the field beyond documentation toward conceptual depth.

5.2. Global and Regional Leadership

The three levels of contributions, namely country, institution, and author, reveal a research ecosystem that is globally distributed yet regionally concentrated. These patterns underscore both the growing internationalization of the field and the emergence of regional centers of excellence, suggesting a research landscape that is globally interconnected but contextually grounded.

The dominance of the United States in terms of publication output reflects, in part, its early and extensive investment in digital infrastructure, artificial intelligence research, and higher education [85]. However, this quantitative presence should not be interpreted solely as academic leadership. Rather, it underscores existing global inequalities in access to resources, institutional capacity, and technological ecosystems that enable scholarly production. The lower representation from regions, such as the Global South may not indicate a lack of engagement or interest, but rather structural barriers to participation in the international research landscape.

China’s strong performance suggests a parallel trajectory driven by centralized policy initiatives and an emphasis on scaled AI deployment in education [86]. Notably, despite the relatively small area of Hong Kong, its outsized contribution indicates a highly focused investment in bilingual education, ethics, and critical AI literacy. Authors, such as Thomas K.F. Chiu, and institutions, like The Education University of Hong Kong, have played pivotal roles in shaping scholarly discourses on teacher adaptation and student engagement in AI-mediated classrooms [43].

Meanwhile, contributions from emerging regions, such as the Middle East and Asia, reflect a diversity of motivations [87]. In the Gulf region, for instance, countries, like Saudi Arabia and the United Arab Emirates, are pursuing policy-driven digital transformation [88]. In Asia, localized initiatives, such as AI-assisted education reforms in Malaysia [89] and the promotion of self-regulated learning in Taiwan [90], illustrate how these countries are tailoring AI integration to their specific educational needs and cultural contexts.

5.3. Emerging Focuses and Research Gaps

GAI in education is a rapidly evolving field, propelled by intersecting forces of pedagogical demands, technological innovation, and shifting societal expectations. The analyses of keywords and documents underscore the transformative potential of GAI in reconfiguring educational practices, particularly within higher education. These tools are influencing an increasingly wide array of teaching and learning processes, from instructional design and curriculum development to classroom engagement and educational assessment. Their integration has been linked to improvements in student learning outcomes, more dynamic instructional strategies, and streamlined academic workflows. Notably, GAI is reshaping educators’ roles, prompting them to shift from traditional knowledge transmitters toward facilitators of critical thinking, digital literacy, and AI literacy.

The rapid advancement of LLMs, particularly those based on transformer architectures, has been a key driver of GAI’s integration into educational contexts. Tools, such as GPT-3.5, GPT-4, and their multimodal successors, have significantly expanded the capabilities of AI to understand, generate, and translate complex content across diverse academic domains. Technological developments have lowered the barrier to entry through user-friendly interfaces and API integration, enabling institutions to embed GAI into learning management systems and academic support services.

Although GAI systems are capable of creating content in multiple formats, the generation of text remains their most dominant and impactful educational application. The popularity of ChatGPT, in particular, reflects its widespread adoption across academic institutions and its positioning as both a writing tool and an interactive cognitive assistant. Students use it for brainstorming, outlining, summarizing, editing, and even for acquiring foundational explanations of complex concepts. The impact is especially pronounced in disciplines that heavily emphasize language processing and technical communication, such as engineering, computer science, nursing, and medicine, where writing and coding tasks are central. At the same time, this increased reliance on GAI has sparked pedagogical reflection about how to balance AI-assisted outputs with the cultivation of independent analytical and creative skills. Educators are exploring how to adapt course design and learning objectives to harness AI’s benefits while maintaining academic rigor.

However, the growing adoption of GAI has intensified ethical debates surrounding its role in education. Chief among these concerns are issues of academic integrity, such as plagiarism, unauthorized assistance, and the blurred line between original student work and AI-generated content [91]. As generative tools become more sophisticated, traditional assessment methods are increasingly challenged, prompting educators to rethink what constitutes authentic learning and valid demonstration of knowledge. Additionally, concerns about algorithmic bias, data privacy, and the over-reliance on AI for critical thinking tasks have raised questions about equity and responsible use. These ethical considerations demand not only institutional guidelines and regulatory frameworks but also a cultural shift in how students and educators engage with emerging technologies.

Despite these developments, several key research gaps remain. First, much of the literature remains descriptive and focuses on mapping trends, challenges, and adoption patterns, often without grounding in learning theories or insights into the cognitive effects of AI use. While functional capabilities are well-documented, little is known about how these tools impact learning strategies, metacognition, or knowledge construction [92]. To address this, future research should integrate perspectives from the learning sciences, such as cognitive load theory, constructivism, and self-regulated learning, to examine how GAI influences thinking, comprehension, and motivation in different educational settings [93].

Second, while student perceptions and attitudes have gained attention, most studies rely on short-term surveys or small samples. As a result, there is a fragmented understanding of how learners adapt to and interact with GAI tools across time, contexts, and disciplines. There is little clarity on how student agency, skill development, or engagement evolves in AI-mediated learning environments [94]. To fill this gap, future research should conduct longitudinal and mixed-methods studies that trace learner behavior and experience across diverse educational systems and cultural settings [95]. Such work would deepen understanding of sustained usage patterns, equity concerns, and pedagogical value.

Third, ethical and philosophical issues remain underdeveloped. While academic integrity and plagiarism are often cited, they are typically treated procedurally rather than critically. Broader questions about algorithmic bias, epistemic justice, and surveillance in AI-supported education are rarely explored [96]. These issues underscore that GAI, while powerful, lacks human capacities, such as contextual interpretation, moral judgment, and lived experience, all of which are central to educational inquiry. Future work should apply frameworks from critical pedagogy and digital ethics to examine how GAI reshapes relationships of power, knowledge, and responsibility in education [97]. This would encourage not only responsible use but also value-informed institutional responses.

5.4. Theoretical and Practical Implications

The findings of this study highlight both theoretical and practical implications for understanding the educational integration of GAI. Thematically, the clustering patterns and keyword trajectories suggest that existing educational theories, such as constructivism, the Technology Acceptance Model (TAM), and the Theory of Planned Behavior (TPB), continue to inform the field, but must now account for AI–human co-agency. As learners interact with AI not just as a tool but as a semi-autonomous partner in meaning-making, theoretical models, such as Theory of Planned Behavior [80] and the Technology Acceptance Model [81,82], must be expanded to address co-authorship, distributed cognition, and affective trust in machine-generated output.

Practically, the convergence of ethical, pedagogical, and technical dimensions necessitates a transdisciplinary approach to educational AI policy and implementation. Policymakers must establish clear, inclusive guidelines that address not only risks (e.g., plagiarism, bias, data privacy) but also enable innovation across different institutional and cultural settings [14,30]. Teacher training must become a central priority. As GAI transforms instructional roles, from knowledge transmission to facilitation, educators need support to build AI literacy, ethical reasoning, and pedagogical adaptability [68].

Moreover, GAI challenges traditional assessment systems. Institutions must consider new formats that reward reflection, process, and originality rather than rote output. Frameworks that promote transparency and co-creation (e.g., disclosing AI assistance in assignments) may foster critical awareness while supporting learning. Globally, the diversity of contributions, from Hong Kong to the Gulf countries, also suggests the need for context-sensitive strategies that align AI adoption with local pedagogical values and educational objectives.

5.5. Limitations and Recommendations for Future Research

Despite its breadth, this study faces several limitations. First, it relies solely on the Scopus database, which primarily indexes English-language peer-reviewed journal articles. This may omit relevant regional or non-English contributions, especially from countries actively deploying AI in education but publishing outside mainstream indexing systems. Future studies could consider a cross-database approach to enhance comprehensiveness.

Second, while bibliometric mapping identifies research patterns, it does not capture the quality, nuance, or classroom-level dynamics of AI integration. The clustering analysis reflects dominant themes but cannot explain how these themes play out in practice, particularly across diverse institutional or sociocultural contexts.

Future research should therefore pursue deeper, mixed-method approaches. Qualitative studies, ethnographies, and classroom case studies can illuminate how GAI is actually experienced by students and educators, especially in underrepresented contexts, such as K–12 education, vocational training, and non-Western systems. Longitudinal research is also needed to understand the sustained impact of GAI on learning outcomes, identity formation, and pedagogical change.

Finally, regional disparities in research output call for closer scrutiny. While countries, such as the United States, China, Hong Kong, and the Gulf countries, dominate the literature, European contributions appear underrepresented relative to their research capacity and tradition in educational technology. This imbalance may reflect differences in publication language, database indexing practices, or more cautious policy environments regarding GAI adoption in education. For instance, stricter data governance under GDPR or ongoing ethical debates within the EU may delay widespread implementation or publication. Alternatively, European scholars may publish in regional outlets not indexed in Scopus, or conduct more critical, theoretical work that takes longer to appear in bibliometric datasets.

Future research should explore these regional dynamics more systematically. Comparative bibliometric analyses, including Web of Science, ERIC, or national repositories, may reveal hidden contributions. Moreover, qualitative inquiry into institutional attitudes and funding strategies in European contexts can help explain whether this apparent lag reflects slower uptake, different research priorities, or structural barriers to visibility. Understanding such asymmetries is crucial if the field is to promote inclusive, global knowledge exchange on GAI in education.

6. Conclusions

This bibliometric review has mapped the rapidly expanding research landscape at the intersection of GAI and education from 2022 to 2025. Drawing on a dataset of peer-reviewed journal articles indexed in Scopus, the study analyzed publication trends, thematic foci, geographic and institutional contributions, and intellectual structures. The findings reveal that GAI is no longer a peripheral topic in educational discourse but has quickly become a central concern across disciplines and regions.

The analysis identified five major thematic clusters, namely pedagogical innovation, curriculum and teacher adaptation, technical infrastructure, ethical governance, and assessment and creativity. Each of them reflects distinct but interconnected lines of inquiry. Over time, the literature has shifted from foundational technical and ethical questions toward more applied, learner-centered, and context-sensitive concerns. This shift suggests that researchers and practitioners are moving beyond tool evaluation toward grappling with how GAI reshapes educational roles, practices, and values.

The study also highlights uneven global participation, with strong contributions from the United States, China, Hong Kong, and select Gulf and Asian countries. European representation appears relatively limited, pointing to structural or policy-related factors that warrant further exploration.

Overall, this review contributes an evidence-based synthesis that not only clarifies the state of research but also identifies critical gaps and future directions. As educational systems continue to confront the affordances and dilemmas of GAI, interdisciplinary, inclusive, and critically informed scholarship will be essential. The field must continue to interrogate how these technologies influence learning equity, academic integrity, and the human dimensions of teaching and knowledge creation.