Mapping the research landscape of AI-supported professional development for special education teachers in science education: A bibliometric and content analysis

This study maps the global research landscape (2023-2026) regarding artificial intelligence (AI)-supported learning and teacher professional development (TPD) in inclusive science education. Records of any document type were retrieved from Scopus, Web of Science, and ERIC (N = 271) and screened in accordance with the PRISMA 2020 guidelines, resulting in 27 full-text studies that addressed AI, special education, and science/STEM contexts with implications for teacher practice. Descriptive bibliometric indicators (publication year and country) were combined with a thematic content analysis of a structured synthesis matrix covering AI technologies, target populations, and study designs. The results indicate a marked increase in publications (55.6% of the corpus in 2026, a partial year) with geographical contributions from the USA, Indonesia, and Greece. The thematic synthesis of literature has led to the following four themes: (1) teacher competence, AI literacy, and professional development, (2) accessibility, universal design for learning (UDL), and personalization, (3) immersive, multisensory, and robotic AI technologies, and (4) AI assessment, auto-scoring, and learning analytics. In terms of methodology, the corpus comprises systematic reviews and conceptual frameworks (37.0%), quantitative (25.9%), and qualitative research (25.9%), whereas experimental studies are scarce (11.1%). The synthesis suggests that one-off orientation sessions are unlikely to build sustainable classroom integration capacity, whereas experiential TPD models combining ‘learner-first’ prompt engineering, technological pedagogical content knowledge extended for AI, and UDL principles were associated with higher teaching self-efficacy and instructional adaptation. Multisensory AI environments such as virtual reality and educational robotics were reported to support conceptual learning among students with special educational needs (SEND), particularly those with autism spectrum disorder, although this evidence rests on few studies. Furthermore, the implementation of AI machine learning scoring showed substantial agreement with human scorers in one study (quadratic weighted kappa = 0.72) and may reduce teachers’ workload. However, the issues associated with algorithmic bias, web content accessibility guidelines accessibility compliance, and data privacy necessitate that AI be viewed as pedagogical co-mediator not as a substitute for teachers.