Artificial Intelligence in Education (AIEd): Publication Patterns, Keywords, and Research Focuses

Abstract

Since the advent of generative AI, research on AI in Education (AIEd) has experienced explosive growth. This study systematically explores publication dynamics, keyword evolution, and research focuses in AIEd by analyzing 2952 papers from the Web of Science (1990–2024). Using bibliometric methods, 2800 English publications were screened, with analyses conducted via VOSviewer v1.6.20 and Python v3.11.5. Findings show a surge in publications post-2020, reaching 612 in 2023 and 1216 by November 2024. The US and China are leading contributors, with the University of London and the University of California system as core institutions. Keywords evolved from “AI” and “machine learning” (2018–2020) to “ChatGPT” and “ethics” (post-2022), reflecting dual focuses on technological applications and ethical considerations. Notably, 68% of highly cited papers address ethical controversies, while higher education and medical education emerge as primary application domains, involving personalized learning and intelligent tutoring systems. Cross-disciplinary research is evident, with education studies comprising the largest category. The study reveals AIEd’s shift toward socio-technical integration, highlighting generative AI’s transformative role yet identifying gaps in ethical governance and K-12 research. These insights inform policymakers, journals, and institutions, advocating for enhanced interdisciplinary collaboration and long-term impact research to balance innovation with educational ethics.

1. Introduction

Over the previous decades, the prospects of AIEd have drawn extensive focus, with scholars and researchers carrying out in-depth investigations. However, it was not until the emergence of generative AI tools such as ChatGPT that research on the application of AIEd witnessed exponential growth, signifying the shift from the era of informatization to the era of intelligence in education.

Despite this growing interest, existing studies demonstrate AIEd’s application across diverse educational tiers—from K-12 [1] and higher education [2,3] to professional fields like medicine [4] and computing [5]—yet reveal several critical limitations that hinder a comprehensive understanding of the field’s evolution. Specifically, prior bibliometric analyses [6,7,8] share three notable constraints: (1) Temporal Narrowness: Studies cover limited timespans (e.g., 2007–2017 [7], 2011–2020 [9]), failing to capture accelerating shifts post-2018 (e.g., ChatGPT emergence); (2) Thematic Fragmentation: Most existing reviews narrowly focus on specific domains (e.g., higher education) prevents detection of cross-disciplinary convergence trends; (3) Governance Blind Spots: Previous research identifies ethical concerns [2,3,10] and technology adoption barriers [3,5] but addresses them anecdotally rather than through systematic measurement. No prior study quantitatively assesses the citation-ethics paradox, in which ethical debates attract scholarly attention yet seldom translate into actionable governance frameworks [8,11]. Crucially, while geographic diversity is observed, existing studies have not explored structural inequalities within global collaboration networks, leaving knowledge-production imbalances unaddressed. Generative AI (e.g., ChatGPT) has triggered explosive growth in AIEd research since 2018, yet existing studies remain constrained by the three factors mentioned above.

To overcome these limitations and present a comprehensive perspective,

Table 1. Comparison of this study with prior bibliometric analyses in AIEd.

Study	Temporal Scope	Domains Covered	Generative AI	Ethical Governance	Global Collaboration
Chatterjee and Bhattacharjee, 2020 [8]	2010–2018	Higher Education	No	Limited	No
Holmes et al., 2022 [6]	2010–2020	General AIEd	No	Minimal	No
Wartman and Combs, 2018 [7]	2007–2017	Medical Education	No	None	No
Chan and Hu, 2023 [12]	2011–2022	Higher Education	Yes	Moderate	Limited
This Study	1990–2024	All Levels	Yes	Extensive	Yes

compares this study with prior bibliometric analyses, highlighting its broader temporal scope, cross-domain focus, and ethical emphasis. Building on this comparison, this study advances AIEd by quantifying publication trends, ethical governance deficits, and global collaboration imbalances, informing equitable and ethically grounded policy frameworks.

To address the multifaceted gaps outlined above, this study systematically analyzes global publication patterns, keyword trends, and research priorities in AI education from 1990 to 2024 to quantify disciplinary evolution, ethical governance deficits, and global collaboration imbalances. These objectives respond directly to the identified research need to move beyond isolated cases and measure interdisciplinary convergence and ethical gaps. To achieve this aim, we integrate multidimensional network analysis, temporal trend mapping, and computational linguistics techniques to systematically achieve the following:

(1)

Quantify cross-disciplinary evolution patterns across K-12, higher, and professional education—transcending isolated case studies and overcoming the temporal limitations of prior reviews;

(2)

Measure structural imbalances in global knowledge co-production networks, specifically assessing Global South exclusion highlighted by the under-representation in cited collaborations;

(3)

Operationalize the ethics-governance gap by correlating citation impact of ethical discussions with the development of actionable frameworks—resolving the anecdotal treatment of governance failures.

This integrated methodological approach moves beyond descriptive bibliometrics to provide explanatory diagnostics for the field’s critical imbalances—including the observed 89% higher-ed bias, 78% framework deficit, and the computer science-education citation disconnect—while capturing accelerating disruptions triggered by generative AI post-2018. Ultimately, our findings will equip policymakers, institutions, and educators with data-driven strategies to steer AIEd toward equitable, ethically grounded intelligence-era education.

2. Methods

AIEd papers available in the Web of Science database are screened and analyzed to understand the publication patterns, keywords, and research focuses of AIEd. To adhere to established bibliometric guidelines, this study follows a standardized four-stage framework: data collection, cleaning, analysis, and validation. The Web of Science (WoS) core collection was selected [13], prioritizing peer-reviewed publications for quality assurance. The search query combined education (TS = Education* OR TS = pedagog*) and AI (TS = AI OR TS = Artificial intelligence) terms, aligning with recommendations for thematic keyword clustering [14]. This database was chosen because most of the papers indexed within it undergo a rigorous peer-review process, ensuring a relatively higher quality compared to other sources. Using multiple databases introduces significant heterogeneity in coverage criteria, metadata quality, and indexing standards, complicating longitudinal trend analysis.

We used the following search query: (TS = Education* OR TS = pedagog*) AND (TS = AI OR TS = Artificial intelligence) AND (TI = (AI OR Artificial intelligence OR neural network* OR machine learning OR deep learning OR natural language processing)) AND (TI = educat*). Here, TS refers to “Topic” and TI to “Title”. This approach ensures that the topic must be relevant to education or pedagogy and also include artificial intelligence as a central theme. Additionally, for the second part of the query, we required that the titles of selected papers contain phrases such as “artificial intelligence,” “neural networks,” “machine learning,” “deep learning,” or “natural language processing,” along with words related to education (e.g., “educat*”). The wildcard * is used to cover variations of the root words.

The search yielded a total of 2952 papers. Remarkably, despite 2024 being incomplete (as of November), 1216 papers were already published in 2024. We retained these 1216 papers to maintain the dataset’s comprehensiveness. To ensure comparability across all papers, we included only English-language papers and excluded non-English ones. Furthermore, we implemented a programmatic filter to eliminate papers misclassified as 2024 publications but actually scheduled for 2025. We then manually review the title of the paper to ensure all selected papers are on the topic of AIEd.

2.1. Data Curation Protocol

Data cleaning followed Passas’s protocol [15]: (1) duplicate removal via WoS unique identifiers; (2) institutional name standardization (e.g., merging “University of London” and “UCL”) using a custom thesaurus; (3) exclusion of 2025-scheduled papers via publication date field validation. This process aligns with the PRISMA framework for systematic reviews and ensures data integrity [14]. After these adjustments, the final dataset consisted of 2800 papers. All relevant metadata (full records) were downloaded from Web of Science for further analysis. The Flowchart for Literature Screening is shown in Figure 1.

The dataset encompassed various analytical parameters, including the length of paper titles, types of publications, publication years, research institutions, keywords, and the journals in which these papers were published. Additionally, we examined citation counts for each paper, title lengths, total page counts, and how these parameters evolved over time. Unlike traditional bibliometric studies, this research did not prioritize author interconnections or individual contributions. Instead, it focused on contributions from different countries and institutions. This study employed performance analysis (e.g., publication trends, citation counts, keyword frequencies) and science mapping (e.g., co-occurrence and co-citation networks) to address the research questions. Performance analysis quantified trends to reveal disciplinary evolution, while science mapping visualized structures to identify research hotspots, aligning with established bibliometric frameworks. Furthermore, an examination of the top 50 most-referenced papers in the domain was carried out to determine their main focuses and spotlight the cutting-edge topics in AIEd. These techniques were selected to map publication patterns, keyword evolution, and research focuses, directly aligning with the research questions that seek to quantify disciplinary evolution patterns, ethical governance gaps, and global collaboration structures. Performance analysis quantified trends and impacts, while science mapping visualized thematic and intellectual structures, enhancing insights into AIEd’s development.

2.2. Analytical Tool Rationale

A custom Python script (with Pandas v2.0.3, Matplotlib v3.7.2, Seaborn v0.11.2) was developed to address three limitations of off-the-shelf tools [18]:

(1)

Flexibility for hybrid analysis: Existing tools (e.g., BibExcel) lack support for integrating title length trends with citation impact, a unique focus of this study.

(2)

Domain-specific customization: The script enabled nuanced cleaning of AIEd terminology (e.g., distinguishing “deep learning” from “machine learning” in title annotations).

(3)

Reproducibility via code transparency: The script (available upon request) ensures full reproducibility, surpassing proprietary software’s black-box limitations.

Validation followed two procedures: (1) Cross-tool consistency: Python (v3.11.5) and VOSviewer (v1.6.20) were used, yielding 98% agreement on the top 50 keywords. (2) Temporal stability test: Re-running the script on 2018–2022 subsets showed <3% variance in publication trend slopes, confirming reliability. Unless otherwise noted, these versions apply throughout the study.

Limitations include the following: (1) absence of automated author affiliation disambiguation (addressed via manual curation); (2) reduced efficiency for real-time data updates compared to commercial tools like Scopus Analyze. These trade-offs were justified by the study’s focus on longitudinal, cross-disciplinary analysis.

3. Results

To investigate the publication dynamics, keyword trends, institutional contributions, country-level distributions, journal preferences, and research areas within the field of Artificial Intelligence in Education (AIEd), we extracted relevant data from the Web of Science database. Figure 2, Figure 3 and Figure 4 were generated using Python packages Matplotlib, Seaborn, Pandas, and NumPy. These figures provide a comprehensive overview of the temporal trends, thematic focus, and global landscape of AIEd research, facilitating a deeper understanding of its evolution and current state.

3.1. Publication Dynamics of AIEd

3.1.1. Proportion of Literature Types

Among all types of literature, journal articles represent the largest share, accounting for nearly 55% of the total, far surpassing other types. A possible reason for this is that journal articles undergo peer review, which makes their results more authoritative. Consequently, researchers often prefer publishing their findings in the most authoritative format available. The second-largest category is proceeding papers, representing 19.93% of the total, indicating significant discussion around the application of AIEd, which is also frequently presented at conferences (Figure 2a). The third to fourth types, in order of quantity, are editorial papers and reviews, with document counts ranging between 170 and 200. In contrast, categories such as retracted articles, letters, meeting abstracts, book reviews, and book chapters each have fewer than 100 documents. It is noted that 69 papers have been retracted, and, since most papers in the dataset were published recently, the retracted papers are predominantly recent as well. This high number of retracted publications suggests that the field may be experiencing issues such as insufficient rigor in peer review, challenges in maintaining research integrity, or the fast-paced nature of publication, possibly leading to errors or ethical concerns. Further investigation into the reasons for these retractions could provide valuable insights into improving publication standards and research practices in this domain.

3.1.2. Publication Trends

This paper studies the publication trends from 1990 to 2024. This 34-year span captures AIEd’s complete evolutionary arc. While significant acceleration occurred after 2018, early publications provide crucial historical context. They establish the field’s foundational themes, enabling quantitative tracking of key shifts (e.g., keyword evolution from “expert systems” to “ChatGPT”). Excluding this period would obscure the field’s development trajectory.

From 1990 to 2010, the number of publications remained relatively stable, with some years not seeing a single paper published (Figure 2b). In most years, only two to three papers were published sporadically, and, even in the most productive years, the maximum number of publications reached was seven. Around the year 2020, with significant advancements in large language models, machine learning, and deep learning, the quantity of publications regarding AI applications in education began to grow, reaching 125 papers in 2020. This number saw a sharp increase to 364 papers by 2022. After the appearance of generative AI models such as ChatGPT at the close of 2022, the number of publications skyrocketed to 612 in 2023. As of 1 November 2024, with the year not yet complete, there have already been 1216 papers published on this topic. It is likely that the emergence of generative AI has brought revolutionary changes to the application of AIEd. Pre-2010 publications quantitatively represented less than 1% of the corpus but served as conceptual precursors. These studies explored foundational AI applications (e.g., early intelligent tutoring systems) that shaped later research themes. Analyzing them provided baseline metrics demonstrating the dramatic acceleration in publication velocity and interdisciplinary convergence characterizing the post-2018 era. From this tendency, it could be predicted that, in the coming years, the publication of papers on AIEd will continue to grow exponentially.

3.1.3. Trends in the Countries and Regions Published AIEd Papers

Similar to the tendency in the quantity of published papers, the quantity of nations with authors contributing to research on the application of AIEd has also followed a similar pattern. From 1990 to 2015, only authors from two or three countries were involved in this field sporadically (Figure 2c). The peak during this period was in 2015, with authors from seven different countries contributing. Around 2018, there was a noticeable increase in publications on this topic. By 2020, authors from 36 different countries had participated in this research. This number grew to 53 countries in 2022, 75 countries in 2023, and 93 countries in 2024. These figures indicate that researchers from an increasing number of countries are engaging with this topic, suggesting that the study of AIEd is continuously expanding. Considering that there are over 200 countries worldwide, it is unlikely that the number of participating countries will grow exponentially. However, it is reasonable to predict that more countries will engage in research on the application of AIEd in the coming years, further broadening the scope of this field.

3.1.4. Trends in Institutions and Authors Involved in AIEd

The trend of authors and institutions involved in research on the application of AIEd aligns closely with “the” trends in the quantity of published papers and the countries participating in this field. Before 2015, only a small number of authors and institutions were engaged in this research (Figure 2d). However, after 2015, the number of authors and institutions participating in this field began to grow significantly. From 2018 to 2020, this growth accelerated, showing an exponential trend. By 2024, approximately 4000 authors and 1500 institutions were involved in research on this topic. It is worth noting that the total number of authors exceeds the number of institutions, as an author may be affiliated with multiple institutions or list different affiliations throughout their career. Although this study did not manually track or programmatically identify cases where an author changed their institutional affiliation during their career or where an institution changed its name, it is evident that more new authors and institutions will continue to join this field in the coming years. This indicates a sustained expansion in the study of AI applications in education.

3.1.5. Trends in AIEd’s Paper Title Length over Time

A Python script is used to calculate the length of titles in papers on the utilization of AIEd. The findings showed that, before 2010, the title lengths fluctuated greatly, primarily due to the small number of papers published during this period, which caused statistical biases. In other words, the length of titles in the years with fewer papers was not statistically representative. From 2010 to 2015, the title lengths decreased, but the limited number of papers in this period may still have led to a lack of statistical representativeness (Figure 2e). Notably, from 2018 to 2024, the number of papers increased significantly, making the statistics more representative. During this period, we observed that the title lengths became progressively longer, and the variance in the number of papers remained almost unchanged, further confirming the statistical representativeness of this period. The increase in title length from 2018 to 2024 may be attributed to the expanding scope of research on AI applications in education, possibly involving interdisciplinary or cross-field studies. This expansion likely requires authors to use more words in their titles to better describe their work. If this trend continues, future paper titles will likely become even longer, as the implementation of AIEd will become more dispersed and involve more interdisciplinary fields.

3.1.6. Trends in Number of Pages of AIEd Papers

Similar to the fluctuations in the length of paper titles, the number of pages in papers also experienced significant variation before 2015. This was again due to the small number of papers during this period, which caused the statistical representation of page numbers to be unreliable (Figure 2f). However, after 2018, the average number of pages in papers gradually increased each year. This may be because research on this topic became more in-depth, requiring longer papers to fully explain the research. It is evident that papers published before 2015 generally had shorter lengths compared to those published after 2015, although there were occasional exceptions. This trend further confirms that as research into AI applications in education becomes more detailed, papers are generally lengthier. Another difference post-2018 is the larger variance in the standard deviation of the number of pages, suggesting that while some papers in this field have grown significantly longer, others, such as newer or more concise discoveries, may still be relatively shorter. Regardless, the average page length of papers indicates that the trend will likely continue, with future papers in this field becoming longer, aligning with the earlier observation that paper titles are also lengthening.

3.1.7. Trends in the Number of References Cited over Time

The references in a paper play an essential part in supporting the paper’s topic and the arguments it presents. Statistical analysis reveals that, similar to the trends in paper length and title, the quantity of references in papers regarding the application of AIEd was relatively low before 2015, as indicated by the average value (Figure 2g). Between 1990 and 2015, the number of references varied significantly, likely due to the small number of papers published in certain years, which led to statistical biases in both the average and variance. Around 1998, one or two review papers were published, which had a larger number of references, causing a significant rise in both the quantity of references and the statistical variance. This aligns with the trends observed in paper length from the previous section. From 2018 onward, there has been a steady growth in the quantity of references cited in papers regarding the application of AIEd. This increase may be due to the expanding volume of research in this domain, expanding the pool of references available to researchers, giving them more options, and leading to a higher citation count. Additionally, as research on this topic has become more in-depth and interdisciplinary, authors need to cite more references to support their arguments. It is foreseeable that the number of references cited in papers on this topic will continue to rise.

3.2. Keywords in AI-Related Articles in Education

From the number of publications (Figure 2b), the development of AIEd began to make a qualitative leap only since 2018. Therefore, this paper analyzes the evolution of AIEd keywords from a timeline perspective (Figure 3) and finds that the evolution of keywords has experienced several changes, with the specific content shown in

Table 2. Evolution of keywords in AIEd by time period.

Time Period	Characteristics of Keywords	Main Keywords
2018–2020	Initial exploration stage, focusing on the preliminary integration of basic AI technologies and education	ai, education, artificial intelligence, machine learning, deep learning
2020–2022	Development and expansion stage. With technological advancements and changes in educational demands, keywords related to online education and intelligent technology applications emerged	online education, natural language processing, big data, e-learning, chatbot
2022–2024	Hot-spot outbreak and deep-diving stage. Generative AI became the focus, and considerations related to ethics emerged	ChatGPT, generative AI, GPT-3, GPT-4, ethics

.

3.2.1. Multi-Dimensional Analysis of the Top Ten Keywords

Some papers did not use keywords, while most included from four to six keywords. Our findings reveal that 1177 papers used “artificial intelligence” as a keyword, accounting for 66.5% of all papers. These appeared in 565 journals and were cited 15,523 times, with 57 papers being cited more than 57 times (H-index = 55). The second most common keyword was “education,” appearing in 343 papers (61.2% of the total) across 520 journals. Although fewer papers used this keyword, the total citation count was significant, reaching 15,107, comparable to that of “artificial intelligence.” Of these, 58 papers were cited more than 58 times (Figure 4a).

Despite ChatGPT only appearing as a keyword after the end of 2022, there have already been 229 papers using it, with 29 of them cited more than 29 times. The fourth and fifth most common keywords were “higher education” and “machine learning,” indicating that AIEd primarily focuses on higher education compared to other levels of education. While various AI methods are used, machine learning remains undoubtedly the most frequently adopted method in the realm of AIEd. The sixth most common keyword was “AI,” a shorthand for “artificial intelligence.” Although its frequency was relatively low, it still had a significant citation count and H-index. If we combine “AI” with “artificial intelligence,” the frequency of this keyword becomes even more prominent. Ranked seventh was “generative artificial intelligence,” with 140 papers, though its citation count was not very high. The eighth-ranked keyword was “ethics,” emphasizing the significance of ethical aspects in AIEd. The ninth and tenth most common keywords were “deep learning” and “generative artificial intelligence,” with 79 and 72 papers, respectively, and citation counts of 1052 and 1022 (Figure 4a). The keywords “artificial intelligence” and “generative AI” overlap, which means the actual ranking for “artificial intelligence” in the top ten positions would be even higher.

Among the top ten keywords, two are related to education, while the rest are associated with AI. This suggests that while research in AIEd is heavily focused on artificial intelligence itself, there is less attention given to education-related aspects. This may be because research is dispersed across various educational fields, applying AI methods to study different applications. Consequently, the related educational keywords did not rank in the top ten, as they are more fragmented across different areas.

3.2.2. Ten Keywords of AIEd Papers Affiliated by Countries and Regions

We further analyzed the application of these top ten keywords in different countries. In the United States, there are 197 papers that use “artificial intelligence” as a keyword in the context of its application in education, while 58 papers focus on “education” as a keyword (Figure 4b). In China, which ranks second, there are 204 papers using “artificial intelligence” as a keyword and 27 papers using “education.” This is easily understandable, given the large number of researchers and research teams in both the United States and China. Among the other countries, Spain, India, and the UK rank third, fourth, and fifth in terms of using “artificial intelligence” as the primary keyword. In these countries, the top two keywords are consistently “artificial intelligence” and “education.” The countries ranked sixth through tenth are Germany, Australia, South Korea, Taiwan, and Canada. These countries have relatively few papers regarding the application of AIEd, and while “artificial intelligence” ranks first, “education” does not always appear as the second most used keyword.

In the chart, we highlight the top two keywords for each country in a brighter color, while dimming the other keywords. One noticeable pattern is that most countries focus on “artificial intelligence” as a keyword. However, when it comes to studies on the application of AIEd, different countries emphasize different aspects. Of course, there are some countries with too few relevant studies, which may lead to significant statistical errors. Nevertheless, it is clear that both China and the United States have comparatively high frequencies of using these top ten keywords, with “artificial intelligence” and “education” consistently ranking as the top two keywords. This is because our search criteria required that both “artificial intelligence” and “education” appear in the title of the papers. Other keywords were not specified in the search parameters. Therefore, keywords like “higher education,” “machine learning,” “generative AI,” “ethics,” “deep learning,” and “generative artificial intelligence” remain some of the most frequently employed in the context of AIEd.

3.3. Institutes, Countries, Journal, and Research Area of AIEd

3.3.1. Institute Research on AIEd

Of the AIEd research, the University of London has the highest number of publications, totaling 82 articles and accounting for 1.12% of the total (Figure 5a). This is followed by the University of California system, with 71 articles, representing 0.97% of the total. The third, fourth, fifth, and sixth rankings are held by the State University System of Florida, the University of Toronto, and Harvard University, respectively. The quantity of papers they have released regarding this topic ranges from 61 to 69, and the gap between them and the top-ranked University of London is not very large. While there are few universities in mainland China, the number of papers published by Chinese universities is relatively high. This indicates that research on this topic in mainland China is more dispersed across different universities, while, in some European and American countries, studies on the utilization of AIEd tend to be concentrated in a few key universities.

In terms of citation counts, the University of London leads with 1318 citations, followed by the University College London with 1085 citations. To assess citation impact, we calculated citation rates by dividing the overall citations by the quantity of publications (TC/TP). The University College London and the Education University of Hong Kong (EdUHK) exhibit the highest citation rates, at 19.38 and 18.03, respectively.

Regarding the nature of authorship, most publications are multi-institutional collaborations, with only a few independently authored papers from the University of Toronto (one article) and Harvard University (two articles). Such papers, rarely authored by a single author, are relatively uncommon in other fields, indicating that research on the application of AIEd likely requires collaboration among authors from different disciplines, as it is inherently an interdisciplinary topic. This preference for collaborative publications suggests that applying AIEd may require expertise across fields; institutions that excel in both AI and education are better positioned for independent publications.

We also examined the distribution of first-author publications (AU1) and found similar patterns to overall publication counts, where the University of London and the University of California System leading with 58 and 50 articles, respectively. A similar distribution was observed for corresponding authors (AUC). In the ranking of H-index, the University of London and the University of Toronto have the highest values, which are 21 and 20 respectively, reflecting high recognition in the academic community.

The prominence of the University of London and the University of California system can be attributed to their strong academic positions in both AI and education. The University of London, particularly UCL’s Computer Science Department, is a top institution in the UK with significant achievements in AI fields such as computer vision, machine learning, and natural language processing. Notable alumni include Demis Hassabis and David Silver, creators of the AlphaGo algorithm, further highlighting the institution’s influence in AI. Similarly, the University of California system hosts leading AI research programs at campuses like UC Berkeley, which is renowned for its robotics and intelligent systems labs, as well as UCLA and UC San Diego, both prominent in AI research and education.

3.3.2. Top 10 Countries and Regions Research on AIEd

Regarding the quantity of papers released on the AIEd, China and the United States are the two leading countries. The United States has published 529 papers, while China has published 518, together accounting for about one-fifth of the total number of papers (Figure 5b). The citation rates for both countries are also similar, with the U.S. papers being cited 4914 times and China’s papers 4146 times. Although the average citation rate is relatively low, these two countries have a significant amount of collaborative work. Additionally, Chinese scholars tend to be more limited to being independent authors of their papers.

Despite the similar number of papers and citations in both China and the U.S., the H-index of U.S. papers is relatively higher. The countries and regions ranked fifth to eighth in terms of contribution are England, Australia, and Germany, with each having published between 89 and 104 papers. Countries ranked from sixth to tenth include Canada, South Korea, India, Spain, and Taiwan. Although Taiwan has only published 48 papers, it has an average citation rate of 18.5 times per paper, which is the highest among all regions and countries, with an average citation rate of approximately 40 times per paper.

3.3.3. AIEd Papers in Different Journals

In terms of total publication volume, Sustainability leads, with 68 articles (2.4%), followed closely by Education and Information Technologies, with 64 articles (2.3%) (Figure 5c). Interestingly, the theme of Sustainability as a journal is not directly related to the application of AIEd (Figure 5c). However, it has published the majority of papers on this topic, with a citation count of 1017, ranking second among the top ten journals. The journals ranked from third to fifth are Education Science, IEEE Access, and International Journal of Artificial Intelligence in Education, with 40, 35, and 33 papers published, respectively. The citation rates for these journals are relatively low. The journals ranked fifth to tenth have a similar number of papers published, ranging from 20 to 24 papers. Considering that there are 2800 AIEd papers, the fact that no single journal in the top ten has more than 70 papers on this topic suggests that research in this field is widely distributed across different journals, rather than being concentrated in one specific journal as is often the case in other fields. Because these journals have relatively few papers published, their H-index values are also relatively low. Interestingly, the International Journal of Technologies in Higher Education has published only 20 papers on this topic but has been cited 1619 times, with an average of more than 81 citations per paper per day. This indicates that the journal maintains high quality standards for the papers it publishes, and the overall quality of the papers is also relatively high. The ranking of disciplines for publishing AI-related education papers

3.3.4. Research Area Category of AIEd Papers

Each paper is categorized under a specific theme upon publication. Among the papers concerning the application of AIEd, 936 papers (19.8%), or about one-fifth, are classified under the theme “Education and Educational Research” (Figure 5d). These papers have been cited a total of 9950 times, with a relatively high citation rate. Notably, 52 of these papers have been cited more than 52 times. The second most common theme is “Computer Science and Artificial Intelligence,” with 406 papers, accounting for approximately 6.8%. However, the total citation count for this theme is relatively low. This may indicate that papers on the application of AIEd classified under this theme are considered peripheral to the core areas of computer science and artificial intelligence, which could explain their lower citation rates. The third theme, “Computer Science in Disciplinary Applications,” is similar to the second theme and includes 360 papers, also with relatively few citations. The fourth and fifth themes are “Education, Scientific Discipline” and “Computer Science, Information Systems,” with 336 and 247 papers, respectively. The citation count for the fourth theme, “Education, Scientific Discipline,” is 3368, which is significantly higher than that for the third theme. This further suggests that papers concerning the application of AIEd tend to receive more attention when published in education-related fields. In contrast, when published in computer science-related journals, these papers tend to have lower citation rates. The themes ranked from sixth to tenth are all related to computer science and have not been categorized under education. Therefore, research on the AIEd still requires further in-depth exploration to bridge this gap.

4. Review of High-Cited AIEd Papers

Among the top 50 most highly-referenced articles in the domain of AIEd (as shown in

Table A1. Top 50 highly cited AI articles in education.

Ref.	Cited	Type	Contents Summary	Education Level
[1]	751	Review	Analyzes AI applications in higher education, focusing on academic support, personalization, and ethical challenges.	Higher educations
[2]	484	Review	Analyzes artificial intelligence applications in education, emphasizing pedagogical impact, ethical challenges, and research trends.	-
[3]	258	Editorial	Open AI platforms in nursing education offer both opportunities for academic advancement and pose risks of ethical misuse.	Nursing students
[4]	253	Article	Explores ChatGPT’s potential in science education, addressing ethical issues, pedagogical applications, and implications for teaching and research practices.	-
[46]	253	Article	Analyzes generative AI’s paradoxes and transformative potential for reformation in management education.	-
[5]	212	Editorial	Examines AI and chatbots’ impact on plagiarism, academic integrity, and assignment design in education.	college students
[10]	195	Conference Proceeding	Identifies AI’s transformative potential in personalized learning, assessment, teaching methods, and communication.	-
[6]	156	Article	Proposes a framework addressing ethics in AI education, emphasizing fairness, transparency, and accountability.	-
[7]	156	Article	Advocates reforming medical education by integrating AI, data analytics, and compassionate care principles.	Medical students and healthcare professionals.
[8]	156	Article	Explores factors influencing AI adoption in education using structural equation modeling.	-
[11]	155	Review	Reviews AI applications in education from 2010-2020, identifying trends, challenges, and research directions.	-
[9]	144	Article	Proposes an AI policy framework addressing ethical, pedagogical, and operational dimensions in higher education.	University learners and employees in higher education.
[28]	140	Review	Explores ChatGPT’s opportunities and challenges in education, focusing on university settings and teaching strategies.	University students and educators in higher education.
[29]	138	Review	Presents a comprehensive assessment of AI applications in advanced education, spotlighting tendencies and voids.	Undergraduate students
[12]	138	Articles	Explores university students’ perceptions of generative AI in higher education, emphasizing advantages and obstacles.	Undergraduate and postgraduate in Hong Kong.
[31]	134	Review	Examines challenges “and” future directions for integrating AI and big data in education.	-
[33]	133	Book	Advocates transforming higher education to adapt to AI-driven societal and workforce changes.	Higher education students and institutions.
[30]	131	Review	Synthesizes empirical research on AI in online higher education, highlighting functions, algorithms, and outcomes.	Learners and instructors in online higher education
[34]	127	Articles	Proposes integrating machine learning education into medical curricula to prepare future healthcare professionals.	Medical students “and” healthcare professionals
[35]	121	Review	Reviews future trends in medical education emphasizing technology, humanism, and adaptive curricula.	Undergraduate medical students
[26]	120	Review	Examines ChatGPT’s transformative potential in education, focusing on personalized learning and ethical considerations.	-
[36]	119	Guide	Discusses integrating AI into medical education, emphasizing role adaptation and ethical considerations.	Medical students and healthcare professionals
[21]	118	Review	Analyzes two decades of AIEd, centering on trends, collaborations, obstacles, and future directions.	-
[38]	116	Article	Proposes “AI”-enabled assessment ecologies emphasizing formative feedback, collaborative learning, and multimodal knowledge representation.	-
[39]	112	Article	Develops a framework to assess surgical expertise using machine learning and virtual reality simulations.	Medical students and surgical trainees
[40]	108	Article	Develops ethical principles for AIEd, centering on transparency, inclusiveness, and human-centeredness.	-
[41]	102	Article	Develops a comprehensive AI literacy framework for education, spanning from kindergarten to university.	From Kindergarten children to university students
[23]	102	Article	Analyzes bibliometric trends of AI in higher education, emphasizing research impact and global interest.	Students and educators in higher education institutions
[42]	100	Article	Introduces an Intelligent-TPACK framework integrating ethical considerations for AI-based education tools.	K-12 teachers
[20]	98	Review	Examines AI’s roles and trends in math education using bibliometric mapping and comprehensive review.	Elementary, junior high, and higher education students.
[45]	98	Review	Evaluates AI’s integration in medical education, focusing on curriculum gaps and competency frameworks.	Medical and health informatics students
[43]	98	Article	Analyzes the political economy of AI in Chinese education, emphasizing policy and private sector dynamics.	University students and broader education stakeholders in China
[44]	97	Article	Advocates interdisciplinary partnerships for designing AI-driven educational technologies informed by learning sciences.	-
[27]	96	Article	Explores sustainable curriculum planning for K-12 AI education using self-determination and curriculum design theories.	K-12 school students and teachers.
[47]	95	Conference Proceeding	Discusses prospects and hurdles of AI-driven code formation in introductory coding education.	University-level novice programming students and educators.
[32]	94	Review	Explores AI integration into dental education, emphasizing curriculum updates and ethical considerations.	Undergraduate and postgraduate dental students.
[48]	91	Article	Proposes a smart classroom integrating real-time AI-driven feedback for improving presentation skills.	University students and educators
[49]	88	Editorial	Explores historical, ethical, and governance challenges of AIEd, proposing critical research directions.	-
[50]	88	Review	Introduces AI-based precision education framework in radiology, enhancing personalized training and decision-making.	Radiology trainees and medical students in diagnostic imaging.
[37]	87	Article	Develops a self-determination theory framework promoting inclusion and diversity in primary and secondary AI education.	K-12 students, including boys, girls, high achievers, and low achievers.
[51]	86	Article	Examines ChatGPT’s prospects and hurdles for higher education, emphasizing academic integrity and instructional innovation.	Higher education students and educators
[52]	86	Article	Investigates students’ viewpoints of AI teaching aides, emphasizing usability and communication ease in education.	Undergraduate students in higher education.
[53]	86	Article	Explores AI-enabled technologies integration in foreign language education, focusing on teacher preparation.	foreign language teachers and teacher trainees.
[54]	85	Articles	Reviewing machine learning’s applications in precision education, focusing on predictions, interventions, and individual learner differences	University students
[55]	85	Article	Investigates AI and machine learning’s prospects and hurdles in higher education institutions.	University students and educators in higher education institutions.
[56]	85	Article	Examining the controversies surrounding deep learning’s application in educational performance prediction, focusing on data, methods, and socio-cultural tensions.	high school students
[57]	84	Article	Explores generative AI’s transformative impact on K-12 education in teaching, learning, and assessment.	K-12 students and teachers.
[58]	83	Conference proceeding	Designs an AI curriculum with social robots for early childhood education, promoting hands-on learning.	aged 4 to 7 years in early childhood education
[59]	83	Article	Proposes a framework for applying ChatGPT in education, addressing advantages, obstacles, and ethics.	-
[60]	81	Review	Reviews AI’s state-of-the-art applications in education, analyzing their potential, limitations, and ethical challenges.	-

in Appendix A), 19 concentrated on higher education, while 5 explored the K-12 group, which refers to students from kindergarten through the 12th grade. This review complements the bibliometric findings by identifying dominant themes (e.g., higher education ethics, ChatGPT applications) and unaddressed gaps (e.g., K-12 research scarcity), providing qualitative depth to the quantitative analysis of publication trends and keyword evolution. The medical-focused studies examined various sub-disciplines such as pharmacy, dentistry, nursing, and surgical training. Our classification analysis revealed that journal articles made up the largest proportion of these highly-cited works (24 publications), followed by reviews (15 articles), with editorials, conference proceedings, and general articles each contributing only 3 publications. Despite the growing interest, AIEd papers have relatively few citations overall, with only 6 papers cited over 200 times and 29 cited over 100 times, while the most highly-cited paper garnered 751 citations.

The examination of the top 50 most-referenced articles uncovered substantial focus on ethical matters related to the application of AIEd, with 13 articles addressing potential moral and ethical dilemmas. Researchers examined these concerns across various populations and disciplines, highlighting the ethical implications of AI integration. Beyond ethics, many scholars contributed to the development of discipline-specific measurement scales to enhance AI’s application in educational settings. Notably, ChatGPT, a widely recognized generative AI tool, has garnered substantial academic interest, with 5 of the top 50 articles specifically focusing on its role and applications in education.

This study employs a dual-dimensional analytical framework to examine highly cited literature. First, co-citation network analysis (Figure 6) utilizes VOSviewer to construct a visual representation of objective citation associations based on co-occurrence frequencies—representing how frequently pairs of publications are jointly cited.

This data-driven approach identifies four dominant knowledge clusters (

Table 3. Four knowledge clusters identified through co-citation analysis.

Cluster Name	Core Content	Refs.
Generative AI and Educational Practice (in blue)	Focuses on application cases of tools like ChatGPT in teaching and their impact on the learning experience.	[1,20,21]
AI Ethics and Policy Framework (in green)	Discusses ethical principles (data privacy, academic integrity) and policy-governance systems.	[22,23,24]
AI Technology Trends and Methodology (in yellow)	Focuses on technological trends of machine learning and big data in education, and methodological research on precision education and intelligent tutoring systems.	[7,10,25]
Application in Higher Education and Professional Fields (in red)	Explores the implementation of AI in higher education management and professional fields such as medicine and dentistry.	[9,26,27]

). Second, thematic classification categorizes publications subjectively into four research domains—educational ethics, higher education applications, ChatGPT implementation, and medical education innovation—according to their primary research questions. The methodologies are complementary yet distinct: co-citation clustering reflects objective scholarly affinity through citation patterns (e.g., grouping ethics-focused literature based on mutual citation frequency), while thematic classification addresses subjective conceptual boundaries (e.g., assigning cross-domain works to core thematic domains). Illustrating this complementarity, Holmes et al. [6] appear in both the ethics co-citation cluster and the ethics thematic category due to their citation linkages and content focus. Collectively, this integrated framework—synthesizing objective network associations and subjective thematic induction—reveals the intellectual structure of the field while highlighting research priorities across domains.

4.1. AIEd in Higher Education

In the 50 highly cited articles, 19 focus on higher education. This relatively high proportion suggests that AI applications in education are predominantly oriented toward the higher education sector. Among these 19 articles, nine explore AI applications in higher education from various perspectives, including opportunities and ethical challenges, trends and gaps in AI adoption, scientometric studies, the utilization of AI and machine learning, and approaches to fostering AI literacy across different age groups. The diverse applications of AIEd in higher education across stakeholders are outlined in

Table A2. AI-enabled education in higher education context.

Stakeholder	AI Applications	Benefits	Challenges	Solutions
Students	Personalized learning, intelligent tutoring, language support [30,33]	Improved learning outcomes, equity for diverse backgrounds, critical thinking [30,33,34]	Privacy risks, over-reliance on AI, fairness issues [11,28]	Data protection, promote independent learning, transparent feedback [2,28,34]
Educators	Automated grading, emotion recognition, data analytics [34,35]	Greater efficiency, real-time engagement insights, tailored curriculum [34,35]	Balancing automation with teaching, bias in grading, need for training [26,35]	Ethical AI training, fair grading policies, maintain human oversight [26,30,35]
Administrators	Predictive analytics, resource allocation [33,34,35]	Data-informed decisions, timely interventions [33,34,35]	Privacy/fairness concerns, over-dependence on AI [11,28,36]	Transparency in analytics, ethical guidelines, balance with human judgment [2,34,35]
Institutions	AI curriculum integration, online learning support, research tools [29,30]	Career readiness, accessible online learning, improved research quality [29,30]	Rapid AI changes, academic integrity, equitable access [26,28]	Regular curriculum updates, anti-misuse policies, promote tech access [29,30]

. Some articles also examine AI’s role in online higher education, which is discussed in detail below.

Review studies such as those by Zawacki-Richter et al. [1], Grassini [28], and Crompton and Burke [29] provide broad overviews across disciplines, whereas Ouyang et al. [30], Han et al. [12], and Hwang and Tu [20] focus on specific fields within higher education. These comprehensive analyses indicate that the majority of AIEd publications originate from computer science and STEM fields. The main areas of application include academic studies and administrative tasks. However, research on AI’s implications for educational philosophy and ethics remains limited [8]. Applications in teaching mainly encompass appraisal and evaluation, student learning management, and intelligent tutoring mechanisms, particularly in language acquisition and student performance prediction. Despite the increasing diversification of AI applications in education, issues such as fairness, technological dependency, and privacy concerns persist [28]. As one of the most well-known AI tools, ChatGPT demonstrates significant potential in personalized learning, automated grading, and resource development. However, its rapid adoption has also raised concerns regarding academic integrity, bias, fairness, and the possible displacement of educators. Addressing these challenges will require updated educational policies, continuous professional development for teachers, and enhanced transparency in AI operations. Importantly, AI should be regarded as an assistive tool rather than a replacement for human educators [2].

Four of the articles mentioned above explore AI applications in online higher education, medicine, mathematics, and dentistry. In online higher education, Ouyang et al. [30] propose integrating learning theories into AI-based online learning and leveraging AI techniques to gather and analyze real-time data. In medicine, research emphasizes preparing students for the digital era by improving AI literacy while maintaining humanistic and societal values [31]. Both Han et al. [12] and Thurzo et al. [32] highlight the revolutionary impact of AI on medical education. Han et al. [12] focus on combining technology with humanistic values to meet societal needs, while Thurzo et al. [32] emphasize rapidly updating curricula to improve students’ AI skills in dentistry. Both underscore that adapting medical education to emerging technologies requires maintaining ethical and humanistic integrity.

Beyond reviews, six journal articles investigate specific utilizations of AI in higher education. These range from supporting individual learners to optimizing educational management and assessment processes, ultimately forming a comprehensive loop from teaching to administration. For students, AI enables personalized learning and intelligent guidance, providing tailored learning paths and real-time feedback based on individual learning histories and behavioral data [33]. AI can also help students from diverse linguistic backgrounds overcome language barriers, promoting equity and inclusion in learning [30]. In classrooms, AI-powered intelligent systems equipped with emotion recognition technologies can assess students’ emotions and engagement in real time, offering actionable insights for teachers to adjust their instructional strategies. Multimodal sensing and real-time feedback systems create interactive learning environments that enhance teaching effectiveness [34]. Additionally, AI supports research activities, from generating ideas and synthesizing literature to summarizing data, drafting manuscripts, and offering translation and language editing for non-native English speakers [30].

For educators, AI offers both opportunities and challenges. Positively, it enhances teaching efficiency and quality by automating duties like course planning, marking, and student assessments, enabling educators to concentrate on instructional innovation and student engagement [35]. AI also supports educators by analyzing data to refine teaching methods and tailor curricula to student needs [34]. For administrators, AI promotes data-informed decision-making and education management by analyzing student data to predict academic outcomes, such as dropout risks, enabling timely interventions [33]. Additionally, AI optimizes resource allocation, improves institutional operations, and ensures teaching quality [34]. However, AI also presents significant challenges in areas such as assessment and academic integrity. While AI-driven automated grading tools enhance efficiency and consistency, detecting and addressing issues related to generative AI content pose significant challenges [26]. Some institutions are developing policies to prevent AI misuse and ensure effective and fair academic assessments [30].

Recognizing AI’s growing importance, universities are increasingly incorporating AI-related courses into their curricula, covering its principles, applications, and ethical considerations to prepare students for future career demands [35]. These courses aim to develop students’ abilities to use AI tools to solve complex problems in interdisciplinary contexts [36]. In the future, AI applications will become increasingly widespread across various academic disciplines.

Among the 50 highly cited articles, five focus on K-12 education, highlighting significant interest in this demographic. Compared with higher education, AI applications in K–12 differ in three primary aspects: target audience, curriculum design, and teaching roles. Higher education emphasizes in-depth academic research and advanced coursework for undergraduate and graduate students [2], whereas K-12 education focuses on fostering interest and inclusivity in early learning [1]. In curriculum design, higher education targets complex academic skills and interdisciplinary research capabilities [8], while K-12 education emphasizes foundational skills such as logical thinking, programming, and teamwork [1]. In terms of teaching roles, higher education instructors act more as supervisors and researchers of AI technologies, while K-12 teachers are seen as guides in implementing AI tools and addressing students’ psychological needs [37].

4.2. Ethic of AIEd

Many articles concerning the application of AIEd have highlighted that, while AI offers numerous advantages, ethical, moral, and accountability concerns remain significant challenges. The ethical issues associated with AIEd can be categorized into three levels: the limitations of the technology itself, the adaptability of the educational system, and the broader societal and policy-level impacts. These issues span both technical and societal dimensions. Among them, technical ethical concerns have been the focus of the most extensive discussions.

One of the most prominent concerns is privacy. The data collection required for AIEd poses risks to the privacy of students and teachers, including matters related to data ownership, access privileges, and possible misuse of data [21]. Privacy concerns are particularly acute in personalized learning and predictive analytics, where robust data protection mechanisms are needed. Furthermore, the reliance of AI systems on sensitive data during decision-making may lead to the unintentional exposure of personal information [11].

Another major issue is bias and fairness. AI systems trained with biased data can produce unfair or inaccurate decisions, disproportionately affecting certain groups of students. For example, language learning tools may fail to recognize diverse accents, and grading systems may disadvantage underrepresented groups [7]. Additionally, the “black box” nature of AI decision-making renders it hard for teachers and students to understand the reasoning behind certain decisions, exacerbating ethical risks [11].

Lastly, the complexity of education is another challenge. Education is inherently a nuanced process that involves emotions, social interactions, and individual differences, which cannot be fully addressed through standardized AI models. Many studies emphasize that education goes beyond mere knowledge transfer, encompassing socialization and personalized development, areas where AI has limited capacity [38]. AI is better suited as a complementary tool rather than a substitute for the pivotal role of human educators.

The incorporation of AI into education offers both opportunities and challenges, particularly in ensuring fairness, inclusivity, and respect for student rights. While AI has the potential to enrich learning experiences, it may also exacerbate existing educational inequalities. For instance, schools and regions with better technological resources can leverage AI more effectively, potentially widening the educational divide [39]. Resource disparities may further marginalize disadvantaged groups, highlighting the need for inclusive technology design and equitable policy implementation [7]. Additionally, the application of AIEd often overlooks the protection of student rights. Concerns have been voiced regarding the potential misuse of AI-based classroom monitoring tools, such as facial recognition and behavioral tracking, which could infringe upon students’ autonomy and privacy [39]. However, most articles fail to provide in-depth discussions on these ethical issues, suggesting that the implications of educational surveillance are not receiving sufficient attention.

The growing incorporation of AIEd gives rise to substantial worries at the societal level, particularly regarding its impact on employment. Although AI has the potential to improve teaching efficiency and ease educators’ workloads, it may also reduce the demand for traditional teaching roles, thereby threatening certain jobs [38]. However, most discussions tend to focus more on AI’s potential to assist teaching rather than addressing its employment-related risks. Another pressing issue is the lack of a comprehensive ethical governance framework, which could lead to disorder as AI adoption becomes more widespread in education. Existing ethical frameworks, such as those proposed by UNESCO and the OECD, are often too broad and lack specific guidelines for educational contexts. For example, while global ethical principles exist, their implementation and monitoring mechanisms remain unclear [21]. Some studies argue for more tailored policy frameworks to address the unique challenges of educational settings [40]. This gap in governance is partly due to insufficient ethical research in the field. The depth and scope of studies on AI ethics in education remain limited [21], with many articles calling for stronger interdisciplinary research to better understand the ethical implications and provide theoretical backing for the standardized utilization of AIEd [41].

To handle these hurdles, scholars have devised multiple frameworks aimed at mitigating the negative impacts of AIEd. Holmes et al. [6] propose a multidisciplinary approach emphasizing fairness, transparency, and inclusivity. Chan [9] introduces the AI Eco-Education Policy Framework for safe, responsible AI in teaching, governance, and operations. Nguyen et al. [40] build on global guidelines to prioritize privacy, fairness, and inclusivity, while Celik [42] expands the TPACK model to integrate ethical competencies for educators. These frameworks consistently stress the importance of transparency, fairness, and inclusivity, advocating for active involvement from diverse interested groups like teachers, students, and policymakers, to align AI applications with diverse needs. They also underscore the necessity for teacher training to guarantee effective and ethical use of AI within educational contexts.

Table A3. Summary of ethical challenges and proposed solutions in AIEd.

Ethical Issue	Summary	Key Implications	Proposed Solutions
Privacy [11,21]	Extensive data collection risks privacy, ownership, and misuse	Unintentional exposure of personal information in learning and analytics	Robust data protection; transparent, inclusive frameworks; privacy-focused policies
Bias and Fairness [7,11]	Biased training data can cause unfair outcomes	Exacerbates inequity; “black box” decisions reduce trust	Mitigate data bias; transparent decision-making; integrate ethical knowledge (e.g., expanded TPACK)
Complexity of Education [11,38]	AI struggles with emotional, social, and individual learning needs	May reduce human educator role in personalization	Use AI as a complement to educators; adopt multidisciplinary inclusive design
Educational Inequality [8]	Unequal access to AI widens the education gap	Marginalizes disadvantaged groups	Design inclusive technologies; equitable policy frameworks
Student Rights and Surveillance [8]	AI monitoring tools may infringe autonomy and privacy	Raises concerns about rights and surveillance impacts	Policies to prevent misuse; research on surveillance effects
Employment Impact [38]	AI may reduce traditional teaching roles	Potential job displacement without adequate safeguards	Position AI as assistive, not replacement
Lack of Ethical Governance [9,21,41]	No education-specific ethical frameworks; global guidelines lack specificity	Weak implementation in educational contexts	Develop tailored ethical frameworks; encourage multi-stakeholder input; strengthen interdisciplinary AI ethics research

summarizes the key ethical issues, their implications, and proposed solutions for AIEd, providing a comprehensive overview of the challenges and frameworks discussed.

4.3. Applications of ChatGPT in AIEd

As one of the highly recognized generative AI utilities, ChatGPT has transformative potential in education. Its applications encompass various aspects of the educational process, ranging from resource design and learning support to enhancing educational efficiency and addressing ethical concerns, forming a comprehensive ecosystem. The specific applications and corresponding educational stages are outlined in

Table A4. Applications of ChatGPT in AI-enabled education.

Stage	Applications	Benefits	Challenges	Solutions
Preparation (Curriculum and Resource Design)	Generates course outlines, quizzes, rubrics, multilingual materials; creates flipped classroom and interactive content [23,40]	Saves educator time, diversifies resources, boosts engagement [23,40]	Over-reliance may reduce originality [30]	Educators review and adapt AI content for relevance and quality [40]
Delivery (Teaching and Learning)	Customized learning plans, virtual tutors, real-time feedback; facilitates discussions and debates [28,42]	Supports personalized, multilingual, and reflective learning [23,28]	May weaken critical thinking and independence [30]	Use as supplement to active learning strategies [28]
Assessment (Grading and Evaluation)	Automates grading, provides targeted feedback, validates assessment design [28,42]	Improves efficiency, consistency, and early detection of challenges [28]	Risks of bias, fairness concerns, and academic integrity issues [28]	Maintain teacher oversight, validate AI feedback [42]
Support (Academic and Research Assistance)	Assists with writing, translation, summarization, questionnaire design, literature review drafting [40]	Improves performance for diverse learners, reduces repetitive educator tasks [40]	Risks of plagiarism, copyright issues, reduced reflection [30]	Follow ethical standards, verify data, address copyright [28]
Management (Administrative Efficiency)	Automates FAQs, generates basic materials, analyzes student needs for teaching management [40]	Reduces workload, enables focus on innovation and guidance [40]	Over-dependence may reduce teacher agency [30]	Train educators to use ChatGPT to enhance decision-making [40]

. In curriculum design and teaching resource development, ChatGPT can generate diverse educational content, such as course outlines, teaching cases, and multilingual materials, significantly saving educators’ time. For instance, ChatGPT is widely used to assist teachers in designing lesson units, creating quizzes, developing grading rubrics, and preparing instructional plans [40]. ChatGPT can produce personalized teaching materials, such as resources for flipped classrooms and interactive course content, thereby enhancing student engagement and improving teaching efficiency [23].

During the learning process, ChatGPT provides customized learning plans and real-time tutoring tailored to students’ abilities and progress. By creating virtual tutors and offering personalized recommendations, ChatGPT delivers individualized learning support and immediate feedback, helping students overcome learning obstacles [42]. Additionally, ChatGPT supports students from multilingual backgrounds by adapting to different learning paces and language needs, thus promoting educational equity [23]. It also facilitates critical thinking and reflective learning by engaging students in interactive discussions, fostering debates, and enhancing their problem-solving skills [28]. On the teaching side, ChatGPT assists in validating the effectiveness of assessment designs and fosters the development of reflective teaching practices [30].

In teaching management, ChatGPT offers significant potential to reduce teachers’ workload and enhance efficiency. First, it excels in automated grading and evaluation, capable of generating semi-automated, targeted remarks. This feature aids instructors in quickly identifying students’ educational obstacles and supplies personalized counseling [28]. This technology is particularly effective in standardized testing, although teacher intervention is essential to ensure fairness and transparency in the grading process [42]. Additionally, ChatGPT can assist students with academic writing, translation, and literature summarization, providing valuable language support for non-native English speakers. Teachers can also leverage ChatGPT to streamline research activities, such as designing questionnaires and drafting literature reviews [40]. Furthermore, ChatGPT can reduce the burden of repetitive tasks for teachers, such as answering frequently asked questions or generating basic instructional materials, enabling them to focus on innovative teaching strategies and personalized student guidance [40].

However, the widespread adoption of ChatGPT raises concerns about issues such as academic integrity, privacy protection, and student dependency, which require collaborative efforts from educators and society to address. Although ChatGPT is a powerful tool, excessive reliance on it may weaken students’ critical thinking and independent learning skills [30]. Similarly, it may diminish teachers’ capacity for reflective thinking. To ensure effective application, educators must understand and manage potential biases in ChatGPT, assess data reliability, and address copyright issues, while adhering to ethical usage standards [28].

4.4. AIEd in Medical Education

Among the numerous studies on the implementation of AIEd across various disciplines, exploration in the domain of medicine has shown a particularly high representation. Notably, within the top 50 highly cited articles, eight focus specifically on medical education. These studies investigate applications targeting diverse groups, including nurses, pharmacy students, dental practitioners, medical students, and surgical interns. The specific AI applications, educational impacts, and ethical considerations for different stakeholders in medical education are outlined in

Table A5. AI-enabled education applications, impacts, and considerations in medical education.

Target Audience	Key Applications	Educational Impact	Ethical and Practical Considerations
Medical Students (nursing, pharmacy, dental, surgical trainees)	Virtual patient systems, intelligent tutoring, VR simulations, data analytics for personalized learning [27,12]	Enhances clinical skills, diagnostic/treatment planning, and data-driven decision-making [27,31,45]	Privacy risks, AI tool reliability, fairness/transparency in feedback [43,44]
Educators	Automated grading, dashboards, course planning, assessment automation [7,20]	Increases efficiency, supports pedagogical design, enables real-time adjustments [7,20]	Balance automation with human teaching, address grading bias, provide AI ethics training [7]
Curriculum Designers	Integrating AI-related courses (ML, analytics, ethics), updating curricula for evolving AI [31,45]	Prepares students for data-informed, interdisciplinary work; keeps curricula relevant	Challenge of rapid updates, need to embed ethical and fairness principles [31,43]
Clinical Practitioners (via training)	AI-assisted diagnostics (imaging), big data–based treatment planning [44]	Improves diagnostic accuracy, bridges theory with clinical practice [12,31]	Algorithm transparency, bias prevention, critical evaluation of AI outputs [43,44]
Interdisciplinary Stakeholders (CS, biomedical research)	Cross-disciplinary collaboration to develop AI educational tools [20]	Expands skills through partnerships, fosters innovation [20]	Integration complexity, need for standardized collaboration frameworks [20]

. The disproportionately high representation of research in medical education can be attributed to several factors. Most significantly, medicine is intrinsically linked to human health, making it a prime area of interest for the technical and ethical implications of AI applications. Additionally, the COVID-19 pandemic in recent years has spurred an increase in medical research publications, contributing to the greater volume and proportion of studies in this domain.

In medical education, AI enhances learning and teaching by contributing to various domains, including personalized learning and intelligent feedback, improved teaching efficiency, curriculum design, and education on ethics and transparency. In terms of personalized learning and intelligent feedback, AI leverages virtual patient systems, intelligent tutoring platforms, and data analytics to offer students customized learning routes and instantaneous feedback, thereby optimizing learning outcomes [12]. Automated grading systems and teacher dashboards empower instructors to track student advancement and adjust teaching strategies swiftly [20].

AI also improves teaching efficiency by reducing educators’ workloads, such as in course planning and assessment assignments, permitting them to focus on pedagogical design and student support [7]. In the age of artificial intelligence, curriculum content and instructional design in medicine increasingly incorporate AI-related elements. Wartman and Combs [7] advocate integrating AI-related courses, like machine learning, deep learning, and data analytics, into medical education to equip students with data-informed decision-making and big data analysis skills. Thurzo et al. [32] emphasize the importance of curriculum updates, covering foundational AI knowledge, algorithm principles, and applications, as well as ethical considerations. While AI offers numerous conveniences, it also necessitates attention to ethics and transparency in education. Students must understand the operational principles and potential biases of AI systems, including transparency, fairness, and accountability, to address potential ethical challenges [43]. Additionally, they need to learn to evaluate the limitations of AI tools to ensure their reliability and safety in clinical applications [44].

AI serves as a bridge between theory and practice in clinical skills training. Combined with virtual reality technology, AI enables the simulation of realistic clinical scenarios, aiding students in enhancing diagnostic, treatment planning, and surgical skills [12]. AI-supported simulation training generates data to analyze student performance, providing a foundation for evaluating and improving skills [27]. In clinical practice, AI optimizes medical processes, particularly in diagnostic support, treatment optimization, and data-driven decision-making. AI algorithms, through big data analysis and deep learning, assist physicians in making faster and more accurate diagnoses and optimizing treatment plans. AI also processes medical images, such as X-rays and CT scans, providing students with practical opportunities for automated diagnostic support and treatment planning [44]. By learning to interpret and apply AI-generated diagnostic data, students enhance their practical skills for clinical applications [31]. Furthermore, AI’s capacity to dissect copious quantities of medical information helps cultivate data-driven decision-making skills in medical students and professionals. For instance, students are trained to use AI technologies for disease prediction, risk assessment, and treatment decisions, strengthening their data analysis abilities to meet the demands of the AI era [45]. Looking ahead, the interdisciplinary application of AI, spanning fields such as computer science, statistics, and biomedical sciences, is a growing trend. This integration facilitates collaboration between medical education and clinical research, providing students with comprehensive knowledge and skills for the future [20].

The top 50 highly cited AIEd papers focus on Higher Education (38%, 19 papers), Ethics (26%, 13 papers), ChatGPT Applications (10%, 5 papers), and Medical Education (16%, 8 papers). The specific article information is shown in Annex 1. These themes align with explorations of personalized learning and administrative efficiency in Higher Education, addressing opportunities and ethical challenges. Ethical discussions highlight privacy, bias, and governance gaps, emphasizing the need for transparent frameworks. ChatGPT Applications cover curriculum design, automated grading, and academic integrity concerns, showcasing transformative potential. Medical Education emphasizes virtual patient systems, curriculum updates, and ethical integration, enhancing clinical training. This synthesis underscores AIEd’s interdisciplinary nature, integrating technology, pedagogy, and ethics to advance educational innovation across diverse contexts.

5. Conclusions

This bibliometric study provides a comprehensive longitudinal overview of Artificial Intelligence in Education (AIEd) research from 1990 to 2024, highlighting explosive growth in publications—from a few papers annually in the 1990s to 612 in 2023 and over 1200 by 2024. This surge is particularly notable post-2018, coinciding with advancements in generative AI technologies, like ChatGPT, driving unprecedented research activity. The analysis further reveals a concentration of studies within higher education contexts, with comparatively fewer in K-12 and vocational education settings. Moreover, despite substantial enthusiasm for AI-driven educational innovation, the development of rigorous ethical frameworks and governance guidelines significantly lags behind technological advancements, creating a pronounced ethics–governance gap. Geographically, research output predominantly originates from the Global North, notably the United States and China, underscoring limited engagement from Global South countries.

The findings also indicate disciplinary disparities, as studies published in education journals receive greater citations compared to those in computer science journals, highlighting disciplinary silos and limited interdisciplinary communication. Additionally, research on generative AI applications exhibits a short-term orientation, focusing primarily on immediate outcomes rather than sustained long-term impacts, signaling a clear need for longitudinal evaluations. Collectively, our study maps critical trends, identifies significant imbalances, and provides an empirical foundation for future AIEd development.

To address these gaps, we recommend that researchers prioritize interdisciplinary collaboration, incorporate underrepresented regions and longer-term impact studies, and systematically integrate ethical frameworks into AIEd research designs. Educators should be trained to effectively and ethically utilize AI tools, ensuring these technologies enhance rather than replace human teaching. Policymakers are encouraged to establish targeted governance standards and support equitable AIEd access, particularly by bridging digital divides in resource-constrained regions. Addressing these recommendations will guide AIEd toward a more inclusive, ethically grounded, and impactful future.