AI Summary & Key Takeaways
AI InsightBeyond describing the field, the study contributes a fully reproducible bibliometric protocol and a critical reading of the field’s ChatGPT-centrism, its collaboration deficit and its early signs of thematic saturation.
📌 3 Key Takeaways
Abstract
Generative artificial intelligence (GenAI) has reshaped teaching, learning and assessment in higher education faster than the literature describing it has been consolidated. Existing bibliometric reviews of this area are typically built on a single database, organised around ChatGPT alone, or extended to education at all levels and they rarely report the analytical parameters required for replication. This study addresses that gap by mapping the scientific production structure, intellectual and conceptual organisation and thematic evolution of GenAI research in higher education on a dual-database (Web of Science Core Collection and Scopus), PRISMA 2020-screened and fully parameter-transparent corpus of 2,992 articles and reviews published between 2022 and 2025, drawn from 728 sources and 9,948 authors. Descriptive, keyword co-occurrence, correspondence-analysis-based conceptual structure, strategic (thematic) map and thematic evolution analyses were triangulated across Bibliometrix/Biblioshiny and VOSviewer. Output grew from 5 documents in 2022 to 2,001 in 2025 a 400-fold expansion, with 66.9% of the entire corpus published in the final year alone at a mean of 18.46 citations per document. Impact is steeply concentrated: the ten most-cited documents (0.33% of the corpus) account for 8.75% of all citations. Production is led by China (n = 890), the United States (n = 617) and Australia (n = 401), yet international co-authorship stands at only 10.11%, indicating a field that is collaborative but overwhelmingly domestic. Türkiye (n = 96) ranks in the same output band as the United Arab Emirates, Peru and Hong Kong, but below regional peers such as Saudi Arabia, Indonesia and Jordan. Two independent clustering procedures each return a four-cluster architecture, but not the same four: the co-occurrence network built on indexed vocabulary separates the field by disciplinary setting instructional technology, health-professions education, technology acceptance and a general artificial intelligence core whereas correspondence analysis on author keywords separates it by conceptual concern into technological, educational-application, ethical-assessment and pedagogical dimensions. Generative AI is the single motor theme and the field shifts clearly from technological exploration towards pedagogical integration and academic integrity. Beyond describing the field, the study contributes a fully reproducible bibliometric protocol and a critical reading of the field’s ChatGPT-centrism, its collaboration deficit and its early signs of thematic saturation.
Özet
Generative artificial intelligence (GenAI) has reshaped teaching, learning and assessment in higher education faster than the literature describing it has been consolidated. Existing bibliometric reviews of this area are typically built on a single database, organised around ChatGPT alone, or extended to education at all levels and they rarely report the analytical parameters required for replication. This study addresses that gap by mapping the scientific production structure, intellectual and conceptual organisation and thematic evolution of GenAI research in higher education on a dual-database (Web of Science Core Collection and Scopus), PRISMA 2020-screened and fully parameter-transparent corpus of 2,992 articles and reviews published between 2022 and 2025, drawn from 728 sources and 9,948 authors. Descriptive, keyword co-occurrence, correspondence-analysis-based conceptual structure, strategic (thematic) map and thematic evolution analyses were triangulated across Bibliometrix/Biblioshiny and VOSviewer. Output grew from 5 documents in 2022 to 2,001 in 2025 a 400-fold expansion, with 66.9% of the entire corpus published in the final year alone at a mean of 18.46 citations per document. Impact is steeply concentrated: the ten most-cited documents (0.33% of the corpus) account for 8.75% of all citations. Production is led by China (n = 890), the United States (n = 617) and Australia (n = 401), yet international co-authorship stands at only 10.11%, indicating a field that is collaborative but overwhelmingly domestic. Türkiye (n = 96) ranks in the same output band as the United Arab Emirates, Peru and Hong Kong, but below regional peers such as Saudi Arabia, Indonesia and Jordan. Two independent clustering procedures each return a four-cluster architecture, but not the same four: the co-occurrence network built on indexed vocabulary separates the field by disciplinary setting instructional technology, health-professions education, technology acceptance and a general artificial intelligence core whereas correspondence analysis on author keywords separates it by conceptual concern into technological, educational-application, ethical-assessment and pedagogical dimensions. Generative AI is the single motor theme and the field shifts clearly from technological exploration towards pedagogical integration and academic integrity. Beyond describing the field, the study contributes a fully reproducible bibliometric protocol and a critical reading of the field’s ChatGPT-centrism, its collaboration deficit and its early signs of thematic saturation.