Effectiveness of AI-based tutoring and assessment systems in mathematics education: a systematic review

Abstract

Abstract Artificial intelligence (AI) is increasingly transforming mathematics education through adaptive tutoring systems, automated feedback, and AI-supported assessment tools. However, evidence regarding the effectiveness of these technologies remains dispersed across diverse contexts and study designs. This systematic review synthesised empirical research on the effectiveness of AI-based tutoring and assessment systems in mathematics education published between 2015 and 2025. Guided by the PRISMA framework, a comprehensive search was conducted in Scopus and Web of Science using database-specific search strings. After screening 1,749 records and assessing 76 full-text articles for eligibility, 12 studies met the inclusion criteria and were included in the final synthesis. Data were extracted using a structured framework and study quality was appraised using the Mixed Methods Appraisal Tool (MMAT). Due to heterogeneity in interventions, populations, and outcome measures, a narrative synthesis was employed. Findings indicate that AI-based tutoring systems and adaptive learning platforms generally support improvements in mathematics achievement, particularly among lower-performing learners, although effectiveness varies depending on implementation fidelity, learner characteristics, and instructional context. Studies focusing on generative AI tools such as ChatGPT primarily reported positive perceptions and increased engagement, but evidence of direct achievement gains remains limited. Overall, the evidence base demonstrates moderate methodological quality, with stronger conclusions drawn from experimental and quasi-experimental designs. The review highlights the potential of AI-based tutoring and assessment systems to enhance mathematics learning while emphasising the need for rigorous, large-scale experimental research and clearer reporting of intervention mechanisms.