Shaping the future of education: a cluster analysis of generative AI’s transformative impact Open Access

The recent developments in Generative Artificial Intelligence (Gen-AI) signify a substantial shift with far-reaching educational consequences. As we navigate this technological revolution, it is crucial to map the most recent research, understand the current state of knowledge, identify gaps and pave the way for future research. Activities previously considered beyond automation, especially those involving creative processes, are now being redefined by AI models like ChatGPT (Burger et al., 2023), leading to discussions about their impact on creativity and innovation management (Marr, 2023). The emergence of ChatGPT, known for its ability to create new content independently, has sparked increasing interest across various fields (Larsen and Narayan, 2023), including education (e.g. Baidoo-Anu and Ansah, 2023; Grassini, 2023; Lo, 2023).

However, the unprecedented rise of Gen-AI tools, such as ChatGPT, has transformed the educational landscape; yet, existing research remains fragmented, often lacking a cohesive synthesis of conceptual trends, ethical considerations and practical applications. Many reviews offer narrow analyses that fail to integrate structural and thematic bibliometric insights. This fragmentation hinders educators, policymakers and researchers from fully understanding how Gen-AI reshapes pedagogy, assessment and creativity in education. Addressing this gap, the present study systematically analyzes 817 peer-reviewed articles using bibliometric and content analysis to identify dominant research clusters, map intellectual trends and propose a framework and future research agenda for the responsible integration of AI in education.

A comprehensive literature review is essential to grasp the evolving research landscape of Gen-AI, particularly ChatGPT, in education. Examining existing knowledge can uncover trends, methodologies and new insights in this rapidly changing field. Moreover, systematically reviewing the literature allows us to understand the increased interest among researchers since the launch of ChatGPT and identify areas for future exploration. This paper aims to present the findings of a systematic literature review (SLR) on Gen-AI, explicitly focusing on the influence of ChatGPT in the education sector. In this review, our purpose is to add to the ongoing discussion about how ChatGPT-powered Gen-AI is changing the landscape of the education sector. By looking at existing literature, we hope to provide valuable insights for educators, researchers and policymakers who want to use ChatGPT in the education sector. We aim to answer below four research questions (RQ) in this study:

This paper systematically reviews the use of Gen-AI in the education sector, focusing on ChatGPT, examining publication trends, significant themes and its impact on educational practices. ChatGPT is a generative AI model based on the Generative Pretrained Transformer (GPT) architecture (Yenduri et al., 2024). GPT is a deep learning architecture that uses self-attention mechanisms to understand textual data relationships, enabling the model to generate new content based on input. It is designed to create human-like text by predicting the next word sequentially. This allows the generation of new content, such as responses or stories. Unlike other AI models that primarily classify or retrieve information, ChatGPT uses a transformer-based approach to process long-range dependencies in text, producing coherent and contextually relevant outputs (Brown, 2020; Radford et al., 2019). This study addresses the calls for diverse data utilization and theoretical development by Ansari et al. (2023) and Preiksaitis and Rose (2023) by using bibliometric review and exploring literature through comprehensive search strategies. It maps out educational AI’s structural and conceptual frameworks, identifying research gaps and suggesting directions for future studies to enhance creativity and innovation.

In addition, it assesses the effects of AI tools on student engagement and learning outcomes, underscoring the potential of AI to personalize and improve the educational experience. Therefore, through this study, we aim to offer scholars, educators and practitioners a detailed understanding of the impact of ChatGPT and Gen-AI on the education sector. By spotlighting topics of interest and identifying areas requiring further research, we aim to empower stakeholders to make informed decisions as we navigate the ChatGPT era.

The exploration of artificial intelligence (AI) in education can be traced back to seminal works like Pinker’s (1979) study, which combined theoretical linguistics with computational models. Pinker demonstrated how early AI techniques were applied to decipher complex cognitive tasks such as language learning, illustrating the merger of human cognition and machine processing from an early stage. Building on this, Buchanan’s (2005) work provided a historical analysis of AI’s evolution from philosophical and fictional origins, influenced by thinkers like Descartes and Leibniz and science fiction authors such as Verne and Asimov. Buchanan detailed how the post-Second World War boom in computer technology enabled the practical application of AI theories, marking an era where robotics evolved from simple mechanical devices to complex systems capable of sophisticated tasks. Continuing this exploration, Burkhardt and Rieder’s (2024) discussion focuses on deep learning’s impact and how models like OpenAI’s GPT use extensive data sets and computational power for various applications. They emphasize the role of users as codevelopers in refining AI applications and discuss the centralization of AI capabilities among large tech companies, raising issues of bias, opacity and energy demands. This narrative reflects AI’s intellectual and technological evolution, with significant socioeconomic implications. Thus, the historical narrative of AI, from its theoretical origins to its status as a powerful practical tool in education, reflects a continuum of intellectual and technological evolution with profound socioeconomic implications.

Based on training data, Generative AI (Gen-AI) techniques create new data or content, like images, text, music or videos. These models can be used for creative purposes like art or deepfake videos but raise ethical concerns about potential misuse for misinformation or media manipulation (Ali et al., 2021). Gen-AI has existed for several decades, as mentioned by Buchanan (2005), who wrote that early milestones in AI include knowledge representation, inference and problem-solving, with roots in natural language processing. Generative AI started being used in education around the early 2000s. The specific term “Gen-AI” may not have a precise starting point. However, generative AI as a field of AI research has been around since at least the early 2010s and has gained prominence with advancements in deep learning and neural network techniques. Its applications include image generation, text generation, music composition and video synthesis (Kaswan et al., 2023).

While many reviews have explored generative AI and ChatGPT in education, comprehensive analyses of 817 articles (2021–2024) using bibliometric and content methods to examine AI’s transformative potential, identify clusters and frameworks and address ethical, methodological and practical challenges remain scarce, as emphasized by Bond et al. (2024), who stress gaps in ethical considerations, methodological diversity, stakeholder inclusivity, geographical representation, interdisciplinary focus and longitudinal studies on AI’s long-term impacts in education. The literature indicates that AI has the potential to improve learning outcomes, engage students and assist educators, but there is a need for empirical research to address challenges effectively. Discussions suggest a promising future where AI, integrated thoughtfully into educational methods, could significantly enhance personalized and engaging learning experiences. Table 1 summarizes the various approaches to generative AI in each review and the connection between AI and education. Given the existence of over 31 SLRs on generative AI in education at the time of this study, this research prioritized selecting the most cited reviews and those published in the highest impact journals to ensure quality and relevance. This approach focuses on influential works representing key insights and trends in the field while maintaining a manageable scope. The criteria for selection ensured methodological rigor and the inclusion of reviews that have significantly shaped academic discourse.

Three reviews before ChatGPT correctly foresaw AI’s role in facilitating personalized learning, tailored feedback and predictive analytics, as well as assisting educators by automating tasks and providing insights into student learning patterns. These reviews – by Zawacki-Richter et al. (2019), Sapci and Sapci (2020) and Song and Wang (2020) – also predicted AI’s continued ability to manage large data sets and improve learning outcomes, especially in virtual learning environments used in medical and health education for student profiling and performance predictions.

The reviews adopt various methodological approaches, from meta-analyses and SLRs to bibliometric analyses (Imran and Almusharraf, 2023; Ansari et al., 2023). Research also underlined the positive effects of AI on students’ learning outcomes and engagement. For instance, Wu and Yu’s meta-analysis (2024) strongly suggests that AI chatbots significantly enhance learning outcomes, mainly through short-term interventions. Similarly, Chiu’s systematic review (2024) emphasizes the potential of generative AI tools to improve student motivation, engagement and academic performance, urging further research from students’ perspectives to better understand AI’s implications in education. Also, Park and Doo (2024) focused on AI applications in blended learning, identifying trends in AI applications that predict learners’ status and achievement, indicating a promising role for AI in personalizing learning experiences. Systematic reviews revealed the diverse applications of AI in educational settings, from classroom discourse to writing assistance. For example, Wang et al.’s (2024) review sheds light on AI’s role in analyzing classroom discourse, enhancing both teaching and learning experiences. Alternatively, Baber et al.’s study (2023) provided an early-stage analysis of ChatGPT research, highlighting its use in natural language processing, dialogue systems and response generation.

While AI’s benefits are widely acknowledged, reviews also highlight the need for cautious optimism. For example, Preiksaitis and Rose (2023) discussed the speculative nature of generative AI’s benefits in medical education, stressing the importance of empirical studies to evaluate its real-world impacts. This sentiment echoes across the reviews, emphasizing the need for balanced approaches to leveraging AI technologies in education. Yet, a review by Bahroun et al. (2023) showcased the interdisciplinary nature of generative AI research in education. Their analysis found that education research spanned computer science, engineering and medical fields, but this emphasized the need for collaboration to tackle complex challenges and applications. Their study found generative AI is revolutionizing education by offering personalized learning experiences. Generative AI was also found to adapt educational content to individual needs, transforming education by creating practical, inclusive and engaging learning environments in the digital age.

Recent SLRs have highlighted several essential research gaps that need addressing to advance our understanding of generative AI’s role in education. First, there is a critical need to evaluate the role and impact of generative AI in blended learning and higher education settings. Researchers like Park and Doo (2024) emphasized the importance of identifying emerging trends and gaps to integrate these technologies better. Future reviews are encouraged to address ethical and privacy concerns related to AI applications in educational contexts. These concerns are not only academic but also aligned with international standards. For instance, UNESCO’s Recommendation on the Ethics of UNESCO (2021) emphasizes the importance of transparency, accountability, inclusiveness and the protection of human rights in the development and deployment of AI. This global ethical framework provides a critical lens through which the integration of AI tools in education – such as ChatGPT – should be evaluated, ensuring that innovation does not come at the cost of fairness, privacy or academic integrity. According to Baber et al. (2023) and Imran and Almusharraf (2023), gathering user feedback and ensuring responsible use are essential to addressing these critical ethical challenges.

Furthermore, Ansari et al. (2023) and Preiksaitis and Rose (2023) advocated using diverse data sources and comprehensive search strategies. They suggested developing theoretical frameworks for domains such as classroom discourse analysis and medical education, which could significantly optimize AI integration and enhance educational outcomes. Finally, the long-term effects of AI tools and their sustainability within academic environments, including K-12 settings, require thorough investigation. This would involve assessing the effectiveness of these tools in enhancing teaching, learning and research processes to ensure that their integration delivers sustained benefits. Fahd et al. (2022) provided a systematic analysis of research applying machine learning in higher education, focusing on student performance, at-risk student identification and attrition prediction, drawing from a combination of primary studies, systematic reviews and meta-analyses totaling 89 relevant publications.

In addition, Chen et al. (2020) assessed the impact of AI on education administration, instruction and learning using a qualitative research design and analyzing 30 sources, including journal articles and professional publications. Vargas-Murillo et al. (2023) investigated the impact of ChatGPT in higher education, exploring its applications, risks, challenges and overall effects on teaching and learning outcomes through a structured, SLR, consulting 16 studies across platforms like Scopus, ScienceDirect and others. This would involve assessing the effectiveness of these tools in enhancing teaching, learning and research processes to ensure that their integration delivers sustained benefits.

Finally, in their meta-SLR, Bond et al. (2024) analyzed secondary studies, including systematic reviews, meta-analyses and scoping reviews, to provide insights and recommendations for advancing AI in education. The review highlights the importance of adhering to rigorous frameworks like preferred reporting items for systematic reviews and meta-analyses, diversifying data sources and using advanced synthesis tools such as EPPI Reviewer to enhance SLRs. It emphasizes the need for diverse data collection methods, including qualitative, mixed-methods and design-based approaches, to ensure broad applicability across different populations. The authors advocate for expanding AI applications beyond STEM fields and into areas such as admissions, course scheduling and student support services, while addressing ethical concerns like data privacy and bias mitigation. In addition, they call for interdisciplinary collaboration, innovative methodologies like bibliometric analysis and focused exploration of emerging AI technologies to advance knowledge and practical applications in education. Finally, Baig and Yadegaridehkordi (2024) conducted a systematic review of 57 studies, offering a comprehensive analysis of ChatGPT’s role in higher education. Their research is notable for encompassing diverse stakeholder perspectives, including students, academic staff, researchers and nonacademic personnel, thereby providing a holistic understanding of ChatGPT’s applications, adoption measures and limitations within the educational context.

While the reviews in Table 1 focus on broader AI applications or specific educational tools, the current bibliometric review provides a deeper analysis of Gen-AI and ChatGPT’s integration into educational practices. It offers a detailed overview of the current state and future directions, advocating for frameworks for the responsible use of Gen-AI in education and setting a clear direction for future studies.

This study uses bibliometric analysis to investigate the landscape of AI in education, leveraging systematic methodologies to explore the existing literature comprehensively. Drawing on Donthu et al. (2021) and Linnenluecke et al. (2020) framework for conducting bibliometric analyses, our research integrates co-citation analysis and text mining to clarify AI’s structural, conceptual and theoretical foundations within academia. Paul and Criado (2020) stated that a bibliometric review is systematic. Still, it focuses on the quantitative analysis of publication data rather than the qualitative synthesis of research findings. It uses statistical tools to analyze trends, citations and co-citations in a large body of published research. Still, it focuses more on metrics like authors and affiliations than the theories, methods and constructs typically emphasized in other systematic reviews.

Bibliometric reviews help summarize extensive bibliographic data to reveal the intellectual structure and emerging research topic or field trends. It aids in identifying prolific contributors and shows the field’s bibliometric and philosophical structure, enabling high-impact research (Donthu et al., 2021). Typically, bibliometric analyses use statistical techniques to scrutinize vast quantities of published studies on a particular theme, shedding light on patterns, citations, co-citations and variables like authorship, publication dates or journals. In contrast to other review types, bibliometric reviews emphasize the roles of authors, their affiliations, countries and citation networks rather than focusing primarily on the theories, methodologies and concepts involved. Our approach, depicted in Figure 1, consists of selecting samples, gathering data, conducting analyses and presenting findings. We undertake an SLR that adopts a science mapping strategy using Scopus and Web of Science for analytical purposes. This method facilitates exploring connections among the contributors to research and investigates themes via coword analysis in the chosen articles (Donthu et al., 2021; Linnenluecke et al., 2020).

This study adopts the approach of Aguinis et al. (2023) to enhance the rigor and utility of its review. It uses a broad spectrum of relevant studies sourced from Scopus and Web of Science databases through a Boolean search string (e.g. “generative ai” AND “educ*”). A transparent screening process was applied, resulting in 817 final documents analyzed using bibliometric tools like VOSviewer and RStudio. These tools facilitated the visualization of the intellectual structure, identification of study relationships and clustering through co-occurrence mapping. The clusters were further examined through content analysis to define themes and research streams, providing a comprehensive overview of the field (Tables 2 and 3 in the manuscript).

This study conducted a comprehensive literature search using Scopus and Web of Science to identify peer-reviewed journal articles on generative AI in education, covering the period from 2021 to 2024. A Boolean search strategy was used, combining terms such as “generative AI” and “ChatGPT” with education-related keywords (e.g. “educ,” “teach,” “learn,” “student”) to ensure comprehensive coverage of relevant studies.

The initial search retrieved 1,364 records. After applying inclusion criteria – focusing on English-language, peer-reviewed journal articles discussing AI applications in educational contexts – and removing irrelevant or duplicate entries, a final data set of 817 articles was obtained. These articles were sourced from journals across multiple disciplines, including social sciences, decision sciences, arts and humanities, business, psychology, economics and multidisciplinary research.

Metadata (e.g. titles, abstracts, authors, keywords, publication years and journal names) were exported in BibTeX format and analyzed using Bibliometrix (R package) and VOSviewer. Manual checks ensured accurate keyword normalization and cluster identification. Two researchers independently conducted content analysis following established guidelines (Gaur and Kumar, 2018; Salipante et al., 1982), focusing on key article features such as theoretical frameworks, research questions, data sources, variables and main findings.

Leveraging content analysis, mainly through unstructured ontological discovery, enabled extracting and identifying concepts, themes and patterns within the textual data of AI in education literature. This approach verified the conceptual structure and thematic progression and complemented the bibliometric review by offering a nuanced interpretation of the content within articles. Unlike statistical and network analysis, content analysis focuses on qualitative aspects of research, including the development of concepts, theoretical debates and methodological approaches. It enhances the understanding of thematic clusters identified by bibliometric techniques like coword analysis, providing a comprehensive perspective (Donthu et al., 2021).

In this section, we have conducted a bibliographic compilation using R-software and a biblioshiny package. Subsequently, we identified the trends observed in publications (RQ1) and identified noteworthy journals, articles and authors (RQ2) per the performance analysis. Furthermore, we have characterized the research domain by identifying themes and clusters (RQ3).

Table 3 presents a synopsis of the data encompassing document contents, authors, author collaborations and document types. Over the years 2021–2024, 817 articles were generated (note: A negative annual growth is because it has just been a quarter of 2024, and the articles for 2023 are for a complete year, which might be higher than those in the first quarter of 2024). The articles demonstrated citation rates, averaging around six citations per article and accumulating over 18,650 citations collectively in a short span. Collaboration among authors was prevalent, with most papers coauthored by three to four individuals, while sole authors authored 105 articles (22.2%). However, most research activities were concentrated within specific countries, with only around 23% of articles originating from international scholars.

Clustering by bibliographic coupling was used to identify and categorize research streams within the Gen-AI_ES domain. This technique groups articles based on shared references if publications citing similar sources are intellectually related. The data set, comprising 817 articles from Scopus and Web of Science, was analyzed using VOSviewer, a bibliometric mapping tool that applies modularity-based clustering algorithms to visualize the relationships between publications. Each publication’s reference list was extracted and processed to compute bibliographic coupling strengths, with the algorithm grouping documents into distinct clusters based on the density of their shared references to ensure clusters were both internally homogeneous and distinct. Following this, the clusters were refined using content analysis, where the titles, abstracts and keywords of the constituent articles were reviewed to define overarching themes. This combined approach ensured the cluster titles accurately represented the intellectual and thematic priorities of the research domain, providing a robust foundation for mapping intellectual structures and uncovering key thematic areas in the literature, as adapted from methodologies such as Rafols et al. (2012) and Waltman et al. (2010).

The trend in scientific production over time is a key indicator of the field’s development (Bornmann et al., 2021). Figure 2 depicts the annual growth of publications in the Gen-AI_ES domain. A significant surge in output occurred in 2023 and the first quarter of 2024, coinciding with the introduction of ChatGPT. This growth highlights the increasing engagement of researchers with generative AI technologies in education. Such patterns underscore the rising importance of Gen-AI_ES as a research area.

Using bibliometrics is helpful to get insights into journal prominence (Donthu et al., 2021). This section highlights journals that have contributed to the Gen-AI_ES domain. The Scimago Journal Rank (SJR) journal ranking systems were consulted, offering distinct methodologies in journal assessment. SJR uses a citation-based quartile ranking system, with Quartile 1 (Q1) representing the highest impact factor (Mason and Singh, 2022). Expert-based rankings involve rigorous peer review by scientific community members to evaluate journal quality, whereas citation-based rankings reflect community preference based on citation patterns (Walters and Zheng, 2023).

Table 4 outlines the most prolific journals in the Gen-AI_ES domain, detailing the number of articles published and their respective rankings according to SJR. The tabulation features the top ten productive journals, with the Computers and Education: Artificial and Education and Information Technologies journals leading with 38 articles each, trailed by the Medical Education with 31 articles and Computers and Education: Artificial Intelligence with 24 articles. Several other journals in the list have published eight or more articles. Notably, most of these journals predominantly fall within the Q1 or Q2 categories in SJR rankings. Furthermore, most journals in the list are indexed in fields such as AI and education, aligning with the focus of this study’s research domain.

This section outlines the top ten authors distinguished by their productivity within the Gen-AI_ES domain and their publication output (Tables 5 and 6) by author and country. Pack A emerged as the most prolific author, contributing seven publications. Pack A is closely followed by Xu X, Chan Cky and Cowling M, with six, five and five publications, respectively. Notably, all the top 1 authors have contributed at least three papers within the Gen-AI_ES domain. Figure 6 illustrates the most prolific authors based on publication volume, with authors such as Tan S., Wang C. and Rudolph J. leading in the number of published documents. In contrast, Table 6 highlights the most influential authors based on total link strength in the bibliographic coupling network, which reflects scholarly impact rather than productivity. This distinction accounts for the differences in author names across the two outputs, as some authors may have a greater influence through citation networks despite publishing fewer articles.

Global citations (GCs), derived from databases such as Scopus and Web of Science, measure a publication’s worldwide impact and visibility. In contrast, Local citations (LCs) represent the number of citations a document receives within the analyzed collection (Apriliyanti and Alon, 2017). Combined, GCs and LCs help contextualize the influence of records within the reviewed data set. This section highlights the most locally cited documents in the Gen-AI_ES domain, emphasizing the significance of citations in advancing knowledge (Bornmann et al., 2021). Table 6 lists the top ten most locally cited documents and their corresponding LC, GC and LC/GC ratios.

The most locally cited documents are Chan (2023), with eight LCs; de Winter (2023), with five, and Chan (2023), with four, as presented in Table 5. While Chan (2023) leads in LC count, Rudolph et al. (2023a) appear more influential in terms of overall network connectivity and thematic relevance despite receiving fewer LCs. This broader influence is reflected in Table 6, which considers bibliographic coupling and link strength. Collectively, four articles authored by Chan exhibit significant local impact, positioning them as key contributions to the Gen-AI_ES discourse.

Figure 3 illustrates the prevailing trend topics identified through authors’ keywords within the Gen-AI_ES domain. To ensure clarity and coherence, synonymous terms were consolidated to prevent redundancy (e.g. artificial intelligence, ai, generative ai, etc.). The top five recent and most significant topics (with a frequency exceeding 100) encompass ChatGPT (528), artificial intelligence (349), education (137), generative AI (120) and higher education (105). ChatGPT and AI have gained substantial attention in studies from 2023 and 2024. In addition, topics related to education (not exclusively higher education), academic integrity and large language models were identified as key themes in this domain. These terms are recurrently depicted in the subsequent thematic maps and evolution diagram, illustrating their growing significance within the Gen-AI_ES domain.

Understanding the interrelationships between terms and identifying underlying clusters is essential in bibliometrics analysis. Clustering by coupling, introduced by Kessler (1960), involves grouping related items based on their co-occurrence patterns, revealing underlying structures and themes within data sets. Coupling frequently appearing words helps identify cohesive clusters, providing valuable insights into complex information landscapes (Gao et al., 2020). In our study, we explore the utility of clustering by coupling authors’ keywords to shed light on clusters and facilitate knowledge discovery within scholarly literature (Figure 4).

We identified four interconnected research streams: (Cluster 1 red) – Generative AI in transforming education (Generative AI in Advancing Education), (Cluster 2 green) – AI in learning and teaching: transforming curriculum (OpenAI and its impact on pedagogy), (Cluster 3 blue) – Integrity and AI in higher education (AI in Higher Education: Integrity and Assessment), (Cluster 4 yellow) – Adopting ChatGPT: enhancing creativity and knowledge. Cluster 1 represents “The transformative impact of AI on education, highlighting its role in enhancing teaching methodologies, reshaping learning experiences and offering insights into cutting-edge technology integration.” The first cluster points to AI’s educational potential by introducing innovative teaching and assessment methods and reshaping educational settings and outcomes. Cluster 2 represents “The influence of AI technology on language processing, focusing on large language models, and its implications in the field of AI.” The second cluster stresses AI’s potential to improve information literacy, transform learning experiences and enhance educational outcomes, highlighting its potential for future teachers and students. Cluster 3 represents “The ethical use of ChatGPT in education assessment, focusing on its impact on academic integrity and assessment practices in the age of advanced AI.” This cluster also features the role of AI in enhancing creativity in education, promoting innovative teaching methods, personalizing learning experiences and reshaping educational practices to foster innovation. Furthermore, cluster 4 explores the integration of ChatGPT in education, focusing on creativity, knowledge adoption and technology acceptance models (TAM). It focuses on how AI technologies enhance learning experiences, provide personalized content and support knowledge adoption through instant access to information and customized learning paths.

The first cluster comprises keywords such as “artificial intelligence,” “chatgpt,” “higher education,” “academic integrity,” “large language models,” “plagiarism,” “academia” and “academic advising.” Cluster 1 highlights how AI tools, such as ChatGPT, are reshaping education by personalizing learning, enhancing teaching efficiency and transforming instructional practices.

While ChatGPT introduces powerful new tools for learning and interaction, its use raises concerns about academic integrity. Its ability to generate text and simulate human responses has prompted institutions to reevaluate existing policies. For example, Alkhaaldi et al. (2023) assessed medical students’ perceptions of ChatGPT, noting its potential to enhance education efficiency. Another instance involves Šedlbauer et al. (2024), who examine the interactions of undergraduate students with ChatGPT, highlighting its advantages for learning and critical thinking support.

Papers on this theme primarily focus on how AI technologies like ChatGPT influence creative processes, idea generation and the enhancement of creative skills among students and educators. For instance, Alshahrani (2023) suggested modifying learning objectives to enhance students’ creativity and critical thinking skills. In another example, Hallal et al. (2023) examined the potential of AI chatbots in organic chemistry education, suggesting the development of innovative teaching and learning methods by improving chatbots’ performance to foster engagement.

Most studies in Cluster 1 do not apply formal theoretical frameworks. Instead, they focus on applied concepts from AI, machine learning and educational technology to explore ChatGPT’s influence on teaching and learning. Karakose et al. (2023) discussed ChatGPT in education through the lens of Kranzberg’s (1986) view on technological neutrality, Henriksen’s (2016) ideas on transformative teaching and Luo et al.’s (2024) concept of intelligence augmentation (IA). Gilson et al. (2023) emphasized ChatGPT’s role as an interactive tool in medical education, underscoring the importance of AI literacy. They highlighted its ability to support assessment and deliver personalized, contextually relevant feedback. Similarly, Dwivedi et al. (2023) examined ChatGPT’s potential to enhance learning and teaching efficiency while also raising concerns about plagiarism and distinguishing between AI-generated and human-authored content. They also proposed future research into AI-related skills, ethical risks, human–AI collaboration and algorithmic bias. The key studies from Cluster 1 are summarized in Table 7. This cluster highlights how Gen-AI, particularly ChatGPT, is transforming pedagogy, personalized learning and academic processes. It helps resolve the fragmented understanding in the field by offering a clear synthesis of Gen-AI’s educational impact, supporting educators with insights into its innovative potential.

The second cluster includes keywords such as “AI,” “chatbots,” “artificial intelligence,” “generative AI,” “ChatGPT,” “academic libraries,” “GPT-3” and “academic assignments.” OpenAI technologies are reshaping education by influencing pedagogical strategies and altering student learning perceptions, prompting renewed interest in models like the TAM. Studies in this cluster focus on AI policy development, the integration of tools like ChatGPT to enhance information literacy and understanding the affordances, challenges and student perceptions of AI. For instance, Chan (2023) proposed AI policy frameworks for higher education, while Johnson et al. (2024a, 2024b) examined the role of ChatGPT in library instruction to strengthen student information literacy. Gao et al. (2024) emphasized ChatGPT’s potential to transform learning experiences and Crompton and Burke (2024) discussed its educational benefits and limitations.

The second theme examines how AI enhances creativity in education by improving teaching and learning, facilitating the generation of creative content, enabling human–AI co-creation and addressing challenges such as misinformation. Amaro et al. (2023) examined the use of ChatGPT in creative tasks, such as summarization and programming, while raising concerns about the dissemination of fake information. This cluster also highlights AI’s transformative role in reimagining educational practices, including its impact on mathematics instruction, the use of custom GPTs for evaluating design tasks and shifts in teaching and assessment strategies. For example, Bower et al. (2024) analyzed how AI is reshaping assessment and instructional methods to foster innovation across learning environments.

Most Cluster 2 studies do not explicitly apply theoretical frameworks to examine the educational impact of ChatGPT, although notable exceptions exist. Johnson et al. (2024a, 2024b) used Universal Design for Learning (UDL) and constructivist theories to design an activity for librarians based on the ACRL Framework, promoting adaptable and hands-on learning for information literacy. UDL supports inclusive learning environments, while constructivism emphasizes interactive learning. Similarly, Bower et al. (2024) adopted the Unified Theory of Acceptance and Use of Technology (UTAUT2) model (Venkatesh et al., 2012) to investigate educators’ motivations for adapting their teaching and assessment practices. The relevant studies are summarized in Table 8. This cluster examines ChatGPT’s role in shaping curriculum and assessment, highlighting its influence on evolving teaching practices. It addresses the need for practical guidance by illustrating how Gen-AI supports educators in adapting instruction to AI-enhanced learning environments.

Cluster 3 addresses the ethical challenges arising from the use of generative AI in higher education, including issues related to plagiarism, cheating, authorship and fairness in assessment. Several studies have examined the implications of students using tools like ChatGPT to generate assignments or write exam responses without proper acknowledgment, raising complex questions about originality and intellectual ownership (Baidoo-Anu and Ansah, 2023; McGee, 2023).

Rather than outright bans, some researchers advocate for policy reform, emphasizing the importance of clear guidelines and AI literacy. For instance, Nguyen (2025) suggested repositioning generative AI as a pedagogical tool under ethical supervision. Similarly, Osman et al. (2024) proposed that educators develop transparent assessment criteria that clarify acceptable AI use.

This cluster directly addresses the research problem of the lack of ethical clarity in AI integration. It aligns with global frameworks, such as UNESCO’s Recommendation on the Ethics of Artificial Intelligence (UNESCO, 2021). These findings underscore the importance of institutions establishing comprehensive governance strategies and fostering ethical awareness among students and faculty (Table 9).

Cluster 4 highlights the role of generative AI in fostering creativity and supporting personalized knowledge construction. Several studies report how tools like ChatGPT enable learners to brainstorm, refine writing and receive iterative feedback – functions that can stimulate higher-order thinking and self-directed learning (Lo, 2023; Zhang et al., 2023).

Generative AI is seen not only as a writing assistant but also as a partner in cognitive exploration. For example, Huang and Lin (2024) documented how students use ChatGPT to generate novel ideas and enhance their understanding of complex concepts. Similarly, Zhao et al. (2023) demonstrated that learners value the tool’s ability to scaffold individual learning paths and adapt to their unique needs and proficiency levels.

This cluster addresses the gap in understanding how Gen-AI supports educational creativity and learner agency, reinforcing the potential of AI as a driver of personalized and inquiry-based learning. However, the authors also caution that uncritical reliance on these tools may inhibit original thought, underscoring the need for guided integration into instructional practice (Table 10).

Synthesizing insights across the four clusters reveals a dynamic interplay of pedagogical innovation, ethical considerations and student-centered learning. The focus on instructional redesign in Cluster 1 converges with curriculum and assessment strategies in Cluster 2. Ethical dilemmas identified in Cluster 3 are deeply tied to the practical use of AI tools in learning environments, as described in Cluster 4.

Several studies (e.g. Kohnke et al., 2023; Rodrigues and Rodrigues, 2023) point to the need for integrated frameworks that consider both technological affordances and human values. This cross-cluster narrative indicates that effective AI adoption in education must go beyond tool deployment, encompassing institutional culture, pedagogical intentionality and ethical foresight.

Overall, these interconnections highlight that Gen-AI’s influence on education is multifaceted. As AI tools continue to evolve, a holistic approach – combining ethics, creativity and pedagogy – is necessary to guide their responsible and impactful integration.

To explore the thematic evolution of the Gen-AI_ES domain, the data set was divided into four annual time slices (2021–2024). For each period, a thematic keyword co-occurrence map was generated using the Bibliometrix software. Figure 5 presents the thematic structure for the first time slice (2021), showing clusters of terms based on their co-occurrence frequency, density and centrality. A thematic map categorizes research themes into four quadrants – motor, basic, niche and emerging or declining – based on their centrality (x-axis) and density (y-axis) metrics (Cobo et al., 2011). Centrality reflects a theme’s importance and its connections to other themes within the research field, while density indicates the theme’s internal development and cohesion. Motor themes, located in the upper-right quadrant, are both well-developed and central, signifying influential research areas. In contrast, themes in the lower-left quadrant are less developed and peripheral, representing either emerging topics or areas of declining interest. This mapping approach provides a comprehensive overview of the research domain’s structure and evolution, facilitating the identification of key areas of growth and emerging significance.

The thematic map categorizes research themes into four quadrants based on centrality (x-axis) and density (y-axis) metrics (Cobo et al., 2011). Motor themes (upper-right quadrant) are both highly developed and central, indicating robust and influential areas of research. Basic themes (lower-right quadrant) are essential but less developed, representing foundational topics with potential for further exploration. Emerging or declining themes (lower-left quadrant) are underdeveloped and peripheral, possibly indicating nascent or waning areas of interest. Niche themes (upper-left quadrant) are well-developed but isolated, reflecting specialized topics with limited connections to other themes.

To complement our thematic analysis, we adopt the approach of Železnik et al. (2017), Kim and So (2022), which involves examining the top 10 most cited articles within each quadrant for each time scale (TS1 and TS2). Figures 6 and 7 outline the four primary themes of research that encompass most keywords within the collection. Based on the figures for TS1, there is one motor theme, two primary themes, one at the intersection of basic and emerging/declining themes, and one niche theme centered on language education.

Figures 5–7 illustrate the thematic evolution and scholarly contributions within the domain of Gen-AI_ES. Figure 5 categorizes themes into motor, basic, niche and emerging/declining categories, highlighting areas such as “large language models” and “prompt engineering” as central, while foundational themes, including “ChatGPT” and “artificial intelligence,” dominate the discourse. Figure 6 identifies prolific contributors, such as Wardat et al. (2023) and Wang et al. (2023), emphasizing the interdisciplinary nature of the field. Figure 7 visualizes the progression of keywords from 2021 to 2024, showing how terms like “human,” “internet” and “ChatGPT” evolved into themes such as “education.” This shift reflects the increasing application of generative AI in pedagogy, focusing on integrating technologies to enhance teaching, learning and student engagement while addressing ethical and practical implications.

A key concern in integrating generative AI into education is “hallucinations,” where models confidently generate false or fabricated information (Sakib et al., 2024). AI hallucinations pose significant risks in education, where students and educators may unknowingly rely on inaccurate content. Ahmad et al. (2023) emphasized the need for safeguards to ensure the accuracy of AI-generated materials, while Weise et al. (2023) pointed to the limitations of these models, which generate content based on patterns rather than factual understanding. These challenges impact the overall reliability and trust in AI systems (MIT n.d.). Strategies such as user warnings and detection frameworks show promise, but ongoing research and consideration of ethical implications are crucial to strengthen AI robustness further and limit misinformation (Ji et al., 2023).

The motor theme explores the articles around internet Web-based learning and curriculum. Two primary themes have been identified in this research domain. The first basic theme, with the most significant number of articles, explores issues related to AI and academic integrity in education. The second basic theme includes articles exploring ChatGPT and assessment issues. Papers on generative AI emerge as an intersection between basic and emerging/declining themes, while language education emerges as a niche theme.

The motor theme explores the articles around large language models, creative thinking and undergraduate students. There are three basic themes in the second-time slice. The first basic theme, with the most significant number of articles, explores issues related to AI and integration in education. The second basic theme includes articles that explore the issues around Chatbots and critical thinking. The final basic theme explores the article on academic dishonesty. One emerging theme is prompt engineering, possibly due to the emerging ChatGPT. One niche theme focuses on self-regulated learning, academic writing and digital multimodal composing.

Hence, discernible shifts in focus and thematic priorities are evident throughout the thematic evolution within the Gen-AI_ES domain from 2021 to 2024. Initially, between 2021 and 2023, the predominant themes revolved around internet Web-based learning, curriculum development and the intersection of AI with academic integrity in educational contexts. This period saw a concentration of research articles exploring the implications of AI technologies, particularly in maintaining academic honesty and assessing students through platforms like ChatGPT. However, by 2024, the thematic landscape had undergone a notable transformation. The discourse shifted toward a broader exploration of the applications of large language models, emphasizing creative thinking and their integration into educational settings.

The emergence of themes surrounding chatbots and critical thinking signaled a deeper inquiry into the pedagogical potential of AI, expanding beyond mere assessment concerns. Furthermore, identifying prompt engineering as an emerging theme underscored a growing recognition of the importance of tailored interactions in optimizing AI-driven educational experiences. Concurrently, niche themes such as self-regulated learning, academic writing and digital multimodal composing came to the fore, highlighting specific areas of interest within the evolving landscape of AI-enhanced education. These thematic evolutions reflect the maturation of research within the Gen-AI_ES domain and a dynamic response to the advancing capabilities and emerging challenges posed by AI in educational contexts. Table 11 aligns the thematic clusters with the derived propositions, offering a consolidated view that enhances the coherence and accessibility of our analysis.

This paper systematically reviews and synthesizes the growing body of literature on generative AI in education, with a focus on ChatGPT. Analyzing 817 peer-reviewed articles from 2021 to 2024, it combines bibliometric analysis, thematic mapping and content analysis to uncover key publication trends, thematic developments and the integration of Gen-AI into educational practice. The findings highlight a surge in research activity, growing interest in personalized learning and improvements in student engagement and outcomes. By identifying four evolving research clusters and aligning them with global ethical standards such as UNESCO’s AI Ethics Recommendation, the study offers a conceptual framework, ten theoretical propositions and a forward-looking research agenda – contributing both scholarly insight and practical guidance for educators, policymakers and AI developers. Table 12 summarizes how each identified research cluster aligns with the study’s problem statement and contributes to addressing key gaps in the literature.

The findings emphasize the importance of strategically embedding Gen-AI tools, such as ChatGPT and large language models, into educational frameworks to enhance personalized learning, automate assessments and foster innovation in pedagogy. This integration can significantly improve student engagement and learning outcomes (Dwivedi et al., 2023; Gao et al., 2024). For instance, ChatGPT’s ability to provide tailored feedback and generate creative prompts supports active learning and critical thinking (Crompton and Burke, 2024). Institutions should also prioritize the development of AI-driven ethical frameworks to address challenges such as plagiarism, data privacy, algorithmic bias and misinformation, ensuring responsible AI use by establishing robust guidelines and protocols that balance innovation with ethical integrity.

Addressing ethical concerns such as plagiarism, misinformation and biases is paramount. This study highlights the need for robust ethical guidelines and frameworks to ensure responsible AI use in education. For example, Rudolph et al. (2023a) discussed the potential misuse of AI for academic dishonesty and emphasize the importance of implementing AI-driven plagiarism detection and monitoring systems. Similarly, Bin-Nashwan et al. (2023) called for comprehensive frameworks to mitigate ethical challenges.

Educators and students must be equipped with AI literacy to maximize the benefits of Gen-AI tools while understanding their limitations and ethical implications. Training programs can empower stakeholders to use tools like ChatGPT effectively and responsibly (Fuchs and Aguilos, 2023). Johnson et al. (2024a, 2024b) also advocated for hands-on activities to improve educators’ and students’ understanding of AI’s role in information literacy.

The integration of Gen-AI requires collaboration between educators, AI developers, psychologists and policymakers. Studies such as Jeon and Lee (2023) underscored the importance of designing AI tools that address diverse pedagogical and cognitive needs. Collaboration can ensure that tools like ChatGPT are inclusive and adaptive to various educational contexts.

Policymakers should promote scalable AI adoption strategies for underprivileged schools and underserved regions. Subsidized programs and infrastructure development can bridge the digital divide, ensuring equitable access to AI-driven education. Alshahrani (2023) emphasized the potential of AI to democratize access to quality education, especially in blended learning environments.

Incorporating Gen-AI into curricula is essential to prepare students for the demands of an AI-driven workforce. AI tools can foster critical thinking, creativity and problem-solving skills, as discussed by Amaro et al. (2023) and Bower et al. (2024). Tailoring teaching strategies to integrate AI technologies will help educators optimize learning outcomes.

This research contributes to advancing theoretical frameworks in AI-integrated education by building on established models such as the TAM and Social Cognitive Theory (SCT). It provides insights into how factors like perceived ease of use, usefulness and cognitive aspects influence the adoption and efficacy of Gen-AI tools in educational contexts (Fuchs and Aguilos, 2023; Bin-Nashwan et al., 2023). In addition, the study maps the thematic evolution of AI education research, offering a deeper understanding of shifting research priorities and identifying emerging themes such as prompt engineering and niche areas like self-regulated learning.

By mapping thematic evolution, this study illustrates the shift in research priorities over time. Emerging themes like prompt engineering and niche areas like self-regulated learning reflect evolving focus areas in the Gen-AI_ES domain (Cobo et al., 2011). The thematic maps presented in Figures 5 and 7 highlight how foundational themes like “ChatGPT” and “artificial intelligence” have matured into applied contexts like education and critical thinking (Gao et al., 2024).

This research provides a foundation for developing normative frameworks addressing ethical challenges such as bias, academic dishonesty and data privacy. For example, Rudolph et al. (2023a) and Steele (2023) discussed the implications of AI use in assessment and propose solutions for maintaining academic integrity while leveraging AI.

The study explores how Gen-AI tools like ChatGPT influence creativity and innovation in education, providing a theoretical lens to examine their interplay with pedagogy. Studies such as Crompton and Burke (2024) and Johnston et al. (2024) demonstrated how AI tools facilitate creative thinking and collaborative learning, offering new opportunities for educational innovation.

The study identifies gaps and proposes areas for further exploration, such as longitudinal studies on the sustained impact of Gen-AI, tailored prompt optimization and AI literacy (Bringula, 2024). The propositions outlined provide a roadmap for advancing theoretical inquiries into AI’s integration in education.

This study underscores the dynamic nature of the Gen-AI_ES domain, with shifts in thematic priorities from foundational topics to applied and ethical considerations. Key themes such as academic dishonesty, critical thinking and self-regulated learning reflect the evolving discourse in AI-enhanced education. Insights from this research can inform innovative teaching strategies, promote stakeholder activation and inspire ethical frameworks for responsible AI adoption (Bin-Nashwan et al., 2023; Gao et al., 2024). By addressing research gaps and fostering interdisciplinary collaboration, this study enriches the knowledge base on Gen-AI in education, inspiring stakeholders to integrate these technologies thoughtfully and effectively into educational practices while maintaining ethical standards and enhancing learning experiences.

To ensure rigor and replicability, the study followed a four-step methodology to develop a future research agenda:

1. Identification of influential articles: A bibliographic coupling map was used to identify 40 highly cited articles that represent core contributions to the Gen-AI_ES literature.

2. Content analysis: These articles were qualitatively analyzed to extract key themes, research gaps and emerging trends shaping the intellectual landscape of the field.

3. Development of research propositions: Drawing on the analysis, a set of targeted research questions and propositions was formulated to address underexplored areas and guide future inquiry.

4. Validation of agenda: The proposed agenda was cross-checked against existing literature to ensure novelty, remove redundancies and prioritize innovative research directions.

This review addressed four key research questions. First (RQ1), the study examined the evolution of scientific production on generative AI in education. The bibliometric analysis revealed a significant surge in scholarly output starting in 2022, indicating a rapid growth in interest in the topic. Second (RQ2), the review identified the dominant thematic clusters within the literature, which include pedagogical transformation, curriculum and teaching strategies, ethics and academic integrity and creativity and knowledge personalization. Third (RQ3), the study explored how these themes have evolved, revealing a shift from initial concerns about AI adoption toward more nuanced discussions on pedagogical redesign, assessment practices and institutional preparedness. Finally, based on the analysis of influential articles, a forward-looking research agenda was proposed (RQ4). This agenda highlights several underexplored areas, such as the integration of ethical frameworks, cross-disciplinary theoretical perspectives and comparative evaluations of different generative AI tools in educational settings.

This structured methodology establishes a transparent and replicable approach for identifying future research directions. The proposed research questions and areas are distinct from the study’s results, as they address unexplored dimensions, leverage emerging technologies and prioritize practical and theoretical advancements in AI-integrated education.

The literature on generative AI in education is expanding rapidly across various research areas. However, certain aspects of generative AI in education warrant further exploration. To set recommendations for future research endeavors, we adopted a four-step methodology. Initially, we identified 40 influential articles through a bibliometric coupling map for all four clusters. Subsequently, we conducted content analysis on these influential articles to ascertain a prospective research agenda. Third, we transformed the potential research agenda into research questions and propositions. At last, we cross-checked and eliminated identified research questions that scholars have previously addressed. This process yielded 12 future research questions identified by various authors (refer to Table 13). To contribute further to the literature, using thematic analysis maps, we identified seven areas of future research in our study and suggested propositions to address them.

Our review thus revealed several gaps in our knowledge about generative AI in education related to the context (Alshahrani, 2023), methods (Bin-Nashwan et al., 2023; Rejeb et al., 2024), theoretical frameworks (Su and Yang, 2023; Tiwari et al., 2024) and measurements (Iwasawa et al., 2024). We discuss these research gaps and our proposals for dealing with them below. We propose various topics and propositions for future research from our thorough review of each cluster’s papers’ research questions (as in Table 5) and thematic progression in the Gen-AI_ES domain from 2021 to 2024. These future avenues offer scholars opportunities to enrich knowledge and comprehension in AI-integrated education, tackling emerging challenges and leveraging opportunities to enhance student learning experiences.

1. The long-term effects of AI integration in education: While there is growing research on the immediate impact of AI in education, there is a notable absence of studies examining its sustained effects over time. Longitudinal studies are essential to track how AI influences pedagogical practices, student learning outcomes and the overall educational landscape (Bringula, 2024; Zekaj, 2023). Therefore, we propose that:

2. The impact of AI on specific subjects within blended learning environments: Given the shift toward exploring the integration of large language models and chatbots into educational settings, future research could investigate more deeply the pedagogical approaches, such as blended learning, that leverage AI technologies. This includes examining how these tools can be effectively incorporated into teaching practices to enhance student learning outcomes and foster critical thinking skills. Blended learning environments combine traditional classroom instruction with online learning, presenting unique opportunities and challenges. However, there is a lack of research on how AI tools like ChatGPT affect learning outcomes in specific subjects within such environments, hindering our understanding of their efficacy and applicability (Alshahrani, 2023). Therefore, we propose that:

3. Ethical considerations and academic integrity: As AI tools become increasingly prevalent in educational environments, addressing ethical concerns and maintaining academic integrity are paramount. Current research highlights these issues but lacks a thorough exploration of systematic management approaches. Therefore, it is proposed to conduct interdisciplinary research to develop comprehensive frameworks and guidelines for ethical AI usage in educational settings (Vargas-Murillo et al., 2023). This research would focus on establishing standards to ensure data privacy, reduce bias, enhance algorithmic transparency and prevent academic dishonesty, ultimately supporting AI’s responsible and ethically sound integration in education. Therefore, we propose that:

4. Tailored prompt engineering: The emergence of prompt engineering as an area of interest suggests a growing recognition of the importance of tailored interactions in educational AI systems (Denny et al., 2023). Future research could explore the design and optimization of prompts to enhance student engagement, motivation and learning outcomes across various educational contexts. Therefore, we propose that:

5. Niche themes exploration: Niche themes such as self-regulated learning, academic writing, language education and digital multimodal composing represent areas of specific interest within the Gen-AI_ES domain. Future research could explore these topics deeper, examining how AI technologies can support and enhance these aspects of the learning process and address any challenges or limitations. As AI becomes increasingly integrated into education, ensuring students possess the necessary literacy skills to navigate and use these technologies effectively is crucial. However, there is a dearth of research on students’ comprehension of AI literacy and their ability to use AI tools responsibly, highlighting the need for empirical studies in this area (Iwasawa et al., 2024; Rudolph et al., 2023a, 2023b). Therefore, we propose that:

6. Longitudinal studies: Despite emerging evidence on the short-term benefits of AI in education, there is a significant gap in understanding its long-term impacts. Research has focused on immediate outcomes, neglecting how these technologies affect teaching practices, student learning and ethical considerations over extended periods. Therefore, it is crucial to initiate longitudinal studies to systematically evaluate the sustained impacts of AI on educational effectiveness (Tiwari et al., 2024). These studies would track changes in pedagogical approaches, student engagement and academic performance over time, providing a comprehensive understanding of AI’s long-term implications in educational settings. Therefore, we propose that:

7. Cross-disciplinary collaboration: Given the interdisciplinary nature of AI in education, future research could benefit from cross-disciplinary collaboration between researchers in education, computer science, psychology and other relevant fields (Bahroun et al., 2023). de Neufville and Baum (2021) emphasized the importance of collective action in developing and regulating AI in education. This is also consistent with our findings, which state that deploying AI technologies could have significant implications for teaching methodologies, student engagement and educational outcomes. This collaborative approach can foster a deeper understanding of the complex interactions between AI technologies and educational practices.

The intellectual structure of generative AI in education reveals contributions from education, computer science and psychology, forming distinct research clusters (Shafique, 2013). For example, Yildiz (2023) showed that chatbots in vocabulary teaching integrate these disciplines by enhancing teaching methods, engaging learners and improving vocabulary retention through natural language processing and machine learning. Psychology contributes by understanding learner behavior and motivation. Similarly, Dwivedi et al. (2023) discussed the multidisciplinary nature of AI in education, focusing on personalized learning and diverse needs, with computer science providing the technical foundation and psychology examining motivation and engagement. Rudolph et al. (2023a, 2023b) emphasized the importance of pedagogical strategies, student engagement and assessment practices, with computer science and psychology exploring AI’s capabilities and learning impacts. Johnson et al. (2024a, 2024 b) highlight ChatGPT’s application in library instruction, where education focuses on pedagogy, computer science on AI tools and psychology on cognitive and emotional learning factors. Therefore, we propose that:

Despite the significance of our findings within the Gen-AI_ES domain, this study is subject to several limitations. First, our analysis is limited by relying exclusively on research papers indexed in the Scopus and Web of Science databases. Future investigations could expand by incorporating papers from additional databases such as Google Scholar, IEEE Xplore and EBSCOhost. Second, there is potential for deeper analysis by using alternative software tools like Gephi and CiteSpace. In addition to these limitations, it is essential to recognize that our study primarily focused on academic literature, potentially overlooking insights from gray literature, industry reports and other nontraditional sources. Incorporating a broader range of sources could provide a more holistic understanding of Gen-AI in the education landscape.

Furthermore, while our analysis provides a snapshot of the current state of research in Gen-AI_ES, it may not capture ongoing developments or emerging trends in real time. Continuous monitoring and periodic updates could ensure that our findings remain relevant and reflective of the evolving nature of the field.

Despite these constraints, our study contributes valuable insights that can inform future research directions, policy decisions and practical implementations of Gen-AI in education, with its unique value underscored by several key aspects that set it apart, particularly in addressing Bond et al.’s (2024) concern about the proliferation of reviews recycling the same studies and duplicating efforts. First, this study analyzed 817 articles from 2021 to 2024 using an extensive Scopus and Web of Science data set, incorporating innovative analytical techniques, namely, bibliometric and content analyses. Second, the study focused on the transformative potential of generative AI in education, specifically through tools like ChatGPT, focusing on its impact on creativity, curriculum development and academic integrity. Third, although this study used an inductive, literature-driven approach to uncover themes, we recognize that frameworks like UTAUT, technological pedagogical content knowledge and substitution, augmentation, modification, redefinition could enhance understanding of pedagogical impacts. Future research could apply these models to assess how Gen-AI tools like ChatGPT influence teaching practices, educator readiness and technology use in various educational settings. Fourth, this study focuses on ChatGPT due to its dominant presence in educational research. However, we recognize that other Gen-AI tools, such as Bard, Claude and Copilot, are also gaining use. The identified themes, creativity, ethics and instructional innovation are likely relevant across tools. Future research could expand by comparing the use and impact of different Gen-AI technologies. Fifth, this review focuses on Gen-AI trends in education. It does not cover recent developments in agile enterprise architecture, digital transformation, sociotechnical views, or related fields such as knowledge management, design science and organizational studies what are key to understanding systemic change. Future research could adopt cross-disciplinary frameworks to explore their intersection with Gen-AI in education. Finally, the paper uncovered seven future research areas, emphasizing ethical guidelines, innovative methodologies, integrating Gen-AI into personalized learning frameworks, advancing academic discourse and real-world application in education.