Artificial intelligence in STEM education: applications, learning outcomes, and methodological gaps (2016–2025)

Purpose This study analyzes the evolution, applications, and methodological gaps of artificial intelligence (AI) in STEM education between 2016 and 2025. It aims to identify scientific productivity patterns, leading countries and journals, thematic clusters, and emerging trends. Additionally, it examines how AI technologies—such as machine learning, virtual/augmented reality, and generative language models—are being integrated into STEM learning. The purpose is to provide a comprehensive understanding of pedagogical, technological, and methodological developments, while revealing existing limitations and opportunities for future research in STEM education supported by AI. Design/methodology/approach A mixed-method design combining bibliometric and qualitative analysis was applied. Data were extracted from Scopus and Web of Science using a structured search strategy and inclusion criteria, yielding 826 articles. Bibliometric indicators (productivity, citations, collaboration networks, keyword co-occurrence) were analyzed using R and VOSviewer. The 50 most cited and methodologically robust studies underwent qualitative review, focusing on technological, pedagogical, and methodological dimensions. This approach enabled mapping scientific evolution, identifying conceptual clusters, and evaluating AI applications in STEM education from an evidence-based perspective. Findings Results reveal exponential growth of AI in STEM education, with strong increases in publications and citations, particularly since 2020. The United States, China, and the United Kingdom lead global production and collaboration networks. Five thematic clusters dominate the field: learning analytics, engineering education, AI literacy, immersive technologies, and generative AI. Machine learning and VR/AR remain foundational, while ChatGPT and large language models drive recent research impact. Persistent gaps include geographical asymmetry, limited longitudinal studies, and insufficient integration of ethical and equity frameworks. Originality/value This study provides one of the most comprehensive examinations of AI in STEM education by combining large-scale bibliometric mapping with an in-depth qualitative analysis of the most influential articles. It integrates technological, pedagogical, and methodological perspectives to reveal emerging paradigms—particularly the rapid rise of generative AI and AI literacy. The work highlights structural gaps and proposes research directions to strengthen ethical, inclusive, and context-sensitive integration of AI in STEM learning. Its findings offer valuable guidance for researchers, educators, and policymakers seeking to understand and shape this rapidly evolving field.