Exploring the effects of generative AI (GenAI) on learning outcomes: A scoping review

Abstract

Generative artificial intelligence (GenAI) is being rapidly incorporated into educational settings; however, empirical evidence regarding its effects on learning remains fragmented and conceptually uneven. This review synthesises evidence from 31 empirical studies, identified through a structured search of Scopus and Web of Science, to examine how GenAI tools shape learning outcomes across educational levels and disciplinary contexts. Thematic analysis of the included studies identified four outcome domains through which GenAI’s educational impact has been examined: achievements, skills, dispositions, and learning processes. Among these, achievement outcomes were the most frequently measured (71.0%), followed by dispositions and skills (58.1% each), whereas learning processes were comparatively under-examined (32.3%). Although less frequently examined, learners’ engagement patterns offered important insight into how GenAI supported or constrained learning across different contexts. Specifically, GenAI appeared to be educationally productive when learners engaged in cycles of explanation, feedback uptake, monitoring, and iterative refinement, but less productive when AI outputs were used to bypass cognitive effort, defer judgement, or produce superficially improved work without deeper understanding. Based on these findings, the review proposes a cognitive–affective–behavioural model that conceptualises GenAI-supported learning as unfolding through productive scaffolding and risk-oriented offloading pathways. These findings underscore the need for AI literacy development, reflective learning tasks, process-oriented assessment, and opportunities for social interaction to ensure that GenAI supports meaningful learning rather than superficial task completion.