Hello everyone. This is Tanaka from the doctoral program. In this article, I will introduce a paper I read at the recent English seminar.
Paper Title: Can GenAI-empowered feedback promote L2 learners’ self-regulation strategic behavior and writing performance?
Can GenAI-empowered feedback promote L2 learners’ self-regulation strategic behavior and writing performance?
Authors: Lin Sophie Teng, Xinjie Deng, Jing Yang
Journal: System (2026)

1. Overview
This study, based on Self-Regulated Learning (SRL) theory, examines the impact of different types of feedback on EFL learners’ use of SRL strategies and their writing performance. The authors divided 84 learners at a Chinese university into three groups: a “comparison group (no feedback),” a “Bingo group (AWE: automated writing evaluation tool),” and an “ERNIE group (GenAI-powered feedback).” Over a three-month period involving three English writing tasks and revision processes, the study conducted analyses including a mixed factorial ANOVA. The results indicate that the effectiveness of GenAI-powered feedback varies depending on the learner’s English proficiency.

2. Introduction
In English writing instruction, feedback is an essential element for learners’ language development and motivation maintenance. However, it is difficult for teachers to provide personalized, high-quality feedback to all students immediately. To address this challenge, AWE tools have been primarily used; however, it has been pointed out that these tend to focus on formal corrections such as grammar and spelling, and have limitations in promoting learners’ critical thinking and deep self-regulation. The authors focus on the fact that Generative AI based on Large Language Models (LLMs) can generate more nuanced, context-aware feedback through an interactive interface. This study aims to clarify how such “interactive feedback” influences learners’ SRL beyond mere error correction.

3. Literature Review
・The role of SRL and feedback
In Zimmerman’s model, SRL is a dynamic process in which learners control their cognition, emotion, and behavior to achieve their goals, consisting of three cyclical phases (forethought, performance, and self-reflection). The authors state that effective feedback in writing serves as an important external stimulus that allows learners to recognize the gap between their current status and their goals, and encourages the selection of appropriate strategies.

・Comparison between AWE and GenAI
Conventional AWE tools (such as Bingo English or Pigai) are mechanisms that point out errors through pattern recognition and statistical methods; while they contribute to improving linguistic accuracy, the feedback tends to be uniform and is considered less likely to elicit deep learner engagement. In contrast, the authors state that Generative AI can provide specific suggestions for higher-order aspects such as logical structure, idea development, and style improvement. This interactivity is thought to stimulate learners’ metacognition and support the execution of more advanced SRL strategies.
It is also stated that the effectiveness of feedback depends on the learner’s English proficiency; lower-proficiency learners are more likely to feel cognitive load from detailed feedback, while higher-proficiency learners tend to receive GenAI suggestions more critically and are better able to select and reject them.

・Research Gap
The authors point out that while it has already been shown that AWE contributes to improving writing accuracy, there is insufficient research directly comparing GenAI with AWE regarding their impact on both SRL strategies and performance. This study aims to verify this point. The research questions in this study are the following three:
RQ1: To what extent do AWE feedback and GenAI-powered feedback influence Chinese EFL learners’ use of SRL strategies?
RQ2: To what extent do AWE feedback and GenAI-powered feedback influence Chinese EFL learners’ writing performance?
RQ3: Does second language (L2) proficiency moderate the impact of feedback type on the development of Chinese EFL learners’ writing ability?

4. Research Method
4.1 Participants
The study targeted 84 first-year non-English major students at a Chinese university, divided into the following three groups:
・GenAI group: Used ERNIE Bot
・AWE group: Used Bingo English
・Comparison group: No feedback

4.2 Experimental Design and Procedure
This study consists of a pre-test, a three-month intervention, and a post-test.
・In the pre-test, a writing task and an SRL questionnaire (WSSLQ) were administered.
・During training, instruction on how to use each tool was provided.
・During the intervention period, learners worked on three essay tasks and performed multiple revisions based on feedback.
・In the post-test, the same measurements were conducted again.

4.3 Measurement Methods
SRL strategies were measured using the WSSLQ, and writing ability was evaluated based on four aspects: content, organization, language use, and mechanics. Automated evaluation by GPT-4 was used for grading, and high reliability with human evaluation was confirmed. Proficiency was classified based on the EF SET.

5. Results and Discussion
・Impact on SRL strategies (RQ1): Results of the mixed factorial ANOVA
The analysis showed that while the effect of SRL strategies over time was observed, no major differences were confirmed based on the type of feedback alone. However, an interaction was observed in some strategies (e.g., Goal monitoring: η² = 0.09), indicating that growth patterns differ depending on the type of feedback. Also, in the group that received GenAI-powered feedback, improvements in multiple cognitive strategies were confirmed as shown below, while declines in aspects such as feedback processing and emotional regulation were reported.
• ERNIE (GenAI):
・Significantly improved items: Idea planning (d = 1.78), Text processing (d = 0.76), Knowledge rehearsal (d = 0.64), Interest enhancement (d = 0.48)
・Significantly declined items: Feedback handling (d = 0.45), Emotional control (d = 0.52)
The authors mention the possibility of dependence on external tools and the externalization of cognitive load regarding these results.

・Impact on writing performance (RQ2)
Improvement in writing performance was observed in all groups, but there were differences in the rate of improvement. As shown below, the GenAI group showed the highest improvement in performance, and it was reported that improvement in content was particularly notable. On the other hand, it was suggested that AWE contributes to improvements in grammar and mechanics.
• Content ERNIE > Bingo (MD = 0.97, p < .05) > Comparison (significant) (Table 5) → Certain effect on accuracy as well

・Impact of writing proficiency (RQ3)
The effectiveness of feedback differed depending on the learner’s proficiency, and a significant interaction was confirmed between the groups.
・Regarding advanced learners, a greater improvement was confirmed in the Bingo (AWE) condition than in intermediate learners (t(24) = 2.82, MD = 4.68, p = .010), and this result suggests the possibility that AWE feedback shows a greater effect on advanced learners.
The reasons are:
1) Intermediate learners may not possess sufficient language ability to process AWE feedback, resulting in lower engagement in revision activities.
2) High-proficiency learners have higher SRL and can utilize diverse writing strategies, such as cognitive and metacognitive strategies, to internalize AWE feedback.

・For intermediate learners, ERNIE showed a significantly greater improvement than Bingo (MD = 5.67, p < .01). The reason is that intermediate learners may need more support in areas such as content, and it is stated that personalized support can be realized through dialogue (such as follow-up questions) with ERNIE Bot. 6. Conclusion
This study shows that feedback is effective for EFL writing development, and that GenAI is particularly useful for writing content (development of ideas). On the other hand, it has also become clear that its effectiveness varies depending on the learner’s proficiency. Specifically, it is suggested that GenAI is effective for intermediate learners, while AWE may be relatively more suitable for advanced learners. The authors state that GenAI and AWE are not competing but should be used complementarily, and that the selection of feedback according to learner characteristics such as proficiency is important.
It was also suggested that the use of GenAI may affect some aspects of SRL. In particular, it is concluded that educationally careful design is required because there is a possibility that the learner’s active engagement may weaken in aspects such as feedback processing and emotional regulation.

Impressions
I selected this paper from the perspective of a review as a preceding study related to my own research theme, “GenAIxSRLxEFL writing.” In particular, I would like to verify the use of SRL strategies by proficiency level this semester, and the result shown in this study that “the effects on performance and SRL differed according to proficiency depending on the type of feedback” was a very useful finding.
In the background of this study, the issue was pointed out that “conventional AWE tools focus on formal corrections such as spelling and cannot sufficiently promote learners’ critical thinking and SRL.” However, as a result of the verification, it was also shown that GenAI may lower some of the SRL strategies in writing, and I reaffirmed the importance of feedback design utilizing GenAI. The authors stated that “GenAI is effective for intermediate learners, while AWE may be relatively more suitable for advanced learners,” but I believe it is important to integrate the advantages of both and design feedback that can accommodate learners of various proficiency and SRL levels in the future. In particular, I would like to design feedback that serves as scaffolding to more effectively promote SRL awareness and strategies.

Report by: Sayo Tanaka