Mapping the Intellectual Architecture of Education for Sustainable Development in Higher Education: A Bibliometric and AI-Assisted Thematic Synthesis

Background Despite the substantial growth of research on Education for Sustainable Development (ESD) in higher education, the field remains characterized by geographic concentration, methodological constraints, and conceptual fragmentation between established competence frameworks and emerging demands such as climate literacy and green skills. No prior review has simultaneously mapped the bibliometric architecture and conducted an artificial intelligence (AI)-assisted inductive thematic synthesis of this domain. Methods A dual-method design integrated bibliometric network analysis—co-citation, bibliographic coupling, and keyword co-occurrence—with AI-assisted inductive thematic synthesis. A total of 442 peer-reviewed documents were retrieved from Scopus, Web of Science Core Collection, and ERIC (2015–2026) through systematic multi-database search, deduplication, and PICO-adapted relevance screening evaluated by a specialized multi-rater panel (Fleiss’ κ=0.55 – 0.69 , 98.4% unanimous agreement on the retained corpus). Thematic extraction was performed using two complementary large language model architectures (Qwen3-235B and DeepSeek V3.2), governed by a stringent verification protocol achieving a 92.0% alignment score; 37 hallucination flags were audited and corrected via human expert review. Results The corpus exhibits substantial and accelerated scientific expansion since 2019 ( R2=0.91 for exponential fit; compound annual growth rate of 18.4%), with Spain, the United Kingdom, and Germany accounting for over one-third of output. Five thematic clusters emerged: curriculum integration, transformative pedagogies, institutional and faculty roles, sustainability literacy and behavioral outcomes, and digital innovation. Surveys and case studies dominate research designs (37.7%), while experimental approaches represent only 4.0% of the corpus. Green skills and climate literacy remain conceptually peripheral despite their policy urgency, institutional collaboration is fragmented (20.3% of institutions isolated), and faculty populations are understudied (7.0%) relative to students (32.7%). Fifty-six knowledge gaps were identified across structural, declared, and emergent categories, with methods-by-concepts coverage reaching 64.0%. Conclusions The field would benefit from a methodological transition toward longitudinal and experimental designs, enhanced visibility of Global South scholarship (contextualizing database indexing constraints), systematic pedagogical theorization of simulation-based approaches, and deliberate cross-institutional consortia to transition from conceptual proliferation to evidence-based pedagogical science.