Effects of AI-Powered Feedback on English Writing Development: A Systematic Review of Meta-Analytic and Empirical Evidence

The recent proliferation of AI-based technology, in particular generative AI applications such as ChatGPT, and automated writing evaluation tools such as Grammarly, Pigai, and Criterion, has transformed how writing is taught in education. Nevertheless, synthesized evidence on their effectiveness relative to traditional modes of feedback, across diverse instructional contexts, remains limited. This study presents a systematic review that synthesizes existing meta-analytic and empirical evidence on AI-based feedback systems and their influence on English writing outcomes among EFL/ESL students, with particular attention to writing quality, feedback literacy development, and self-regulated learning. We searched Scopus, Web of Science, ERIC, and PsycINFO for publications from January 2018 to December 2024. Inclusion was restricted to peer-reviewed experimental, quasi-experimental, or rigorous mixed-method studies involving adult learners in higher education settings. The review adopts a narrative synthesis approach, integrating findings from seven existing meta-analyses with thematic analysis of 137 empirical studies. This design is best characterized as a systematic review with meta-analytic integration, rather than an original meta-analysis. Meta-analytic evidence from Zhai and Ma (2023) indicates that AWE has a large overall effect on writing quality (g = 0.861, 95% CI [0.65, 1.07]), with medium-sized effects on language quality (g = 0.61) and content quality (g = 0.57). Sun and Zhou’s (2024) comparative meta-analysis of 41 studies (N = 4,813 participants) revealed no significant difference between AI and human feedback (g = 0.25, 95% CI [−0.11, 0.60]). Deng et al. (2024) reported moderate but statistically significant effects on academic performance (g = 0.53). Across the empirical literature, hybrid feedback systems combining AI and teacher feedback yielded superior results compared to AI-only conditions. AI-based feedback represents a promising complement to traditional writing instruction; however, its impact depends on learner proficiency, feedback type, intervention duration, and pedagogical design. Future research should prioritize long-term learning outcomes, the development of feedback literacy, and optimal human-AI collaboration models.