AI-Driven Feedback for University Students’ ESL/EFL Academic Writing: A Scoping Review of Pedagogical Implementation and Learning Outcomes

How Well Does AI-Generated Feedback Work? Intrinsic and Extrinsic Evaluation across more than 20,000 EFL Essay Drafts

This study examines feedback in English as a Foreign Language (EFL) writing contexts, focusing on written corrective feedback (WCF). Large language models (LLMs) can provide WCF at scale, but aligning them with pedagogical best practices remains an ongoing challenge. WCF meeting criteria like factuality or relevance may still be unsuitable for learning contexts, highlighting the need for extrinsic evaluation based on the learner's perspective. We deployed WCF systems in a university-level EFL class with nearly 2,000 students, collecting over 20,000 drafts. We evaluated the generated WCF from two perspectives: intrinsic evaluation by experienced English teachers using a rubric, and extrinsic evaluation via student feedback and engagement metrics. Results revealed low alignment between teacher expert ratings and student feedback. These findings suggest that traditional expert evaluation alone may not fully capture WCF's usability or helpfulness from the learner's perspective, highlighting the importance of learner-centered evaluation frameworks for AI-based applications in language education.

Evidence on AI Tools in Second Language Writing: A Systematic Review of SLA Outcomes, Learner Experiences, and Pedagogical Dynamics

Artificial intelligence (AI) tools have transformed second language (L2) writing instruction, yet empirical evidence spanning traditional Automated Writing Evaluation (AWE) and modern Generative AI (GenAI) remains fragmented. This systematic literature review synthesized 22 primary empirical studies published between 2018 and 2025 to evaluate the effects of AI tools on L2 writing performance and map associated pedagogical benefits, operational challenges, and ethical concerns within Second Language Acquisition (SLA) frameworks. Guided by PRISMA 2020 protocols, literature was gathered across five academic databases and analyzed using SLA performance constructs and thematic synthesis. Findings reveal that most included studies focusing on accuracy (11 out of 14) reported improvements in surface-level linguistic error reduction, whereas effects on syntactic complexity were variable across 8 studies depending on tool type (AWE vs. GenAI). GenAI also showed promising effects on coherence and structural organization when accompanied by explicit instructional scaffolding. While AI tools provide immediate scaffolding and reduce writing anxiety, operational challenges, such as declining engagement, feedback overload, and cognitive passivity, and ethical risks concerning academic integrity and loss of learner voice persist. The review concludes that AI tools are most productive when used as supportive writing companions within explicitly scaffolded and collaborative learning environments. Future research should employ longitudinal designs, include more diverse learner populations, and empirically examine critical AI literacy interventions and the transfer of AI-assisted gains to unassisted L2 writing. Educators should integrate AI critically while fostering learner agency, ethical awareness, and independent writing development.

AI-Mediated Writing Instruction in Higher Education: A Systematic Review of Empirical Evidence

It is suggested that AI can enhance drafting, revision, and feedback processes, improving coherence, metacognition, and writing confidence, however, these benefits are accompanied by persistent concerns regarding ethical ambiguity, inconsistent policy guidance, and insufficient faculty training.

What Error Analysis can still offer TESOL classrooms: A scoping review of research trends and pedagogical implications

Error Analysis (EA) has long been used in English language teaching to identify recurring learner difficulties and inform decisions about instruction and feedback. However, the research literature on EA has developed unevenly, making it difficult for TESOL practitioners to judge what this work can usefully offer in classroom settings. This scoping review examines how EA has been used in English-language research and what it can reasonably contribute to TESOL classrooms. Guided by Arksey and O’Malley’s framework and reported in line with PRISMA-ScR guidelines, the review included 106 peer-reviewed studies published up to February 2025. Analysis of the literature showed three recurring patterns: a strong emphasis on grammatical and lexical errors, continued reliance on small datasets and manual coding, and uneven connections between error description, theoretical explanation, and pedagogical application. Taken together, the findings suggest that EA remains most useful when treated as a diagnostic rather than a prescriptive classroom resource.  

When AI-Mediated Feedback Writing Works and When It Does Not: Lessons Learned from Two Classroom Cases

A Systematic Review of Empirical Research about Beyond Automated Correction by Human–AI Feedback Partnership for Developing L2 Writing

This paper synthesizes empirical studies published from 2023 to 2026 to discuss the growing position of AI-aided writing feedback for L2 students. Drawing upon the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) statement, this study organizes and critically reviews past findings in the context of instructional transformation from existing AWE systems to LLM-empowered collaborative Human-AI writing feedback in L2 education; influence of learner engagement, trust, and AI writing feedback literacy on the adoptability and applicability of AI feedback; and (3) instructional and research limitations that need to be addressed in subsequent research. Our systematic review demonstrates that generative AI facilitates writing quality improvement, revision processes, learner autonomy and increased writing access by providing on-time, tailored and interactive support.  Moreover, this review reveals that the contribution of tool-aided feedback to learning lies not just in the features of AI feedback alone, but more in the users’ ability to critically understand, judge and utilize the suggested pieces of writing to advance writing competence within the teacher-supervised writing environment. Thus, the results would recommend the Human-AI Partnership framework, in which LLMs would substitute or at least assist the role of the human tutor rather than directly replace; at the same time, it would also shed light upon issues of conducting longitudinal, methodologically robust, and theoretically grounded studies regarding generative AI for L2 writing pedagogy.