DESIGN OVER HYPE: EMPIRICAL EVIDENCE ON EFFECTIVE AI INTEGRATION IN EDUCATION

Abstract

Artificial Intelligence (AI) has grown rapidly from optional experiments to core infrastructure in the current learning environment while it’s evidence of real impact remains inconclusive. This study synthesizes contemporary empirical research on AI-integrated learning, teaching and assessment in the era of generative systems and large language models. A comprehensive search on popular reputed databases resulted in 612 records which was further filtered to include only 35 peer reviewed studies after performing PRISMA 2020 reporting standards. The corpus ranges from K-12, higher education, and professional learning, encompassing AI-powered personalization, smart tutoring systems, generative AI assistants, AI-enhanced assessment, chatbots and adaptive feedback environments. Across various study results, AI-powered interventions consistently show positive influence on academic performance, with an average impact from small to large (g/d ≈ 0.43–0.70), while promoting growth in metacognitive regulation, learner engagement and self- efficacy. The most effective outcomes occur when AI functions as structured, feedback-oriented and pedagogically-aligned co-instructor, whereas peripheral or unstructured usage results in modest or minimal impact. Moderator patterns reflect the necessity of instructional design, institutional capacity, educator readiness, and student AI literacy, as well as growing concerns on equity, transparency, academic integrity and over-reliance on generative outputs. The review outlines that AI-enabled learning environments can improvise both cognitive and affective results when coupled with supervised, ethically governed and context-sensitive scenarios, and identifies necessity for longitudinal, cross-disciplinary and multi-site assessments of generative and adaptive AI in education.