Discover Education Education

Artificial intelligence tools as adaptive feedback systems in higher education a systematic review and meta analysis

Washahi, Maryam Al
Springer Nature · 2026-11-21 · DOI 10.1007/s44217-026-02436-3

Abstract

not applicable. Context-aware adaptive and personalized digital learning environments increasingly rely on artificial intelligence (AI) to deliver automated assessment and adaptive feedback at scale, yet the evidence base for their effects on student learning outcomes remains fragmented. This paper reports a systematic review and meta-analysis of empirical studies examining the impact of AI tools, conceptualized as adaptive feedback systems, on learning outcomes in higher education between 2015 and 2024. Following PRISMA 2020 guidelines, five databases (Scopus, Web of Science, ERIC, APA PsycINFO, and Google Scholar) were searched in January 2024, with an update in January 2025, yielding 6,086 records. Fifty reports met the eligibility criteria. Of these, 44 were primary empirical studies, which formed the meta-analytic sample (approximately 28,400 students across 19 countries); the remaining six were prior systematic reviews or meta-analyses, which were used for narrative context only and contributed no effect sizes. Random-effects meta-analysis found a moderate overall effect on learning outcomes (Hedges’ g = 0.42, 95% CI [0.34, 0.50]). Effects were largest for surface-level accuracy (g = 0.74) and smallest for argumentation quality (g = 0.09, ns). Moderator analyses showed that task-specific automated writing evaluation tools outperformed general-purpose generative AI, that English-as-a-foreign-language contexts produced larger effects than first-language English contexts, and that longer interventions produced larger effects. High heterogeneity (I² = 58–84%) indicates that instructional design and implementation quality substantially condition effectiveness. The findings are synthesized into an adaptive AI feedback framework that positions instructor scaffolding, learner trust, and self-regulation as the mechanisms through which adaptive feedback translates into learning, with implications for the design of AI-integrated curricula and adaptive learning environments.

Read at publisher →