Educational Transformation and Governance Turn of Generative Artificial Intelligence in Higher Education: Knowledge Structure, Thematic Evolution, and Research Frontiers

Generative artificial intelligence (GenAI) has moved rapidly from a novel conversational technology to an infrastructure-level challenge for higher education. This study maps the development of research on GenAI in higher education and examines whether the field is shifting from an early emphasis on technological adoption and academic disruption toward educational transformation and institutional governance. A bibliometric dataset of 874 Web of Science records published between 2023 and 20 July 2026 was analysed using CiteSpace. Performance indicators, author and institutional collaboration, country participation, document co-citation, keyword co-occurrence, cluster structure, and temporal evolution were examined. The literature expanded from 34 publications in 2023 to 147 in 2024 and 344 in 2025; 349 records had already been indexed by 20 July 2026. The United States, China, England, Australia, and Spain were the most productive contributors, although brokerage centrality was not proportional to output. Co-citation analysis identified a compact intellectual core organised around AI policy, academic integrity, educational opportunities and risks, assessment, and technology acceptance. The keyword network was highly connected (N = 334, E = 1,194; largest component = 95%), with higher education, artificial intelligence, generative AI, academic integrity, and AI literacy as the most frequent terms. Cluster quality was acceptable (Q = 0.5022; weighted mean silhouette = 0.7856), revealing ten themes that were analytically consolidated into adoption and acceptance, teaching and assessment transformation, academic integrity, AI literacy and epistemic judgement, human-AI interaction, measurement development, and governance in diverse institutional settings. The results show a clear but incomplete governance turn: academic integrity and AI literacy have become central, yet technology acceptance models and behavioural intention remain dominant. The field therefore risks treating GenAI primarily as a user-adoption problem while underexamining institutional responsibility, assessment validity, disciplinary variation, equity, and long-term learning. A future research agenda is proposed around observable learning outcomes, task-specific human-AI collaboration, discipline-based AI literacy, policy effectiveness, assessment validity, and substantive educational equity.