AI-Assisted Writing Feedback for Enhancing Secondary Students’ Writing Skills: An Experimental Study

AI-Assisted Writing Feedback for Enhancing Secondary Students’ Writing Skills: An Experimental Study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article AI-Assisted Writing Feedback for Enhancing Secondary Students’ Writing Skills: An Experimental Study Mehmet EKİZOĞLU, Ayşe Nesil DEMİR This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6430737/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract This study examines the impact of AI-assisted writing feedback on the writing skills of secondary-level EFL students. A sample of 60 Turkish high school students was divided into an experimental group receiving feedback from an AI writing assistant and a control group receiving traditional teacher feedback. Over an 8-week period, students wrote multiple essays; pre-test and post-test writing assessments were administered. Results indicated that the experimental group showed significantly greater improvement in overall writing performance (post-test mean = 82.3) compared to the control group (post-test mean = 75.1, p < .01). The AI-assisted feedback group exhibited notable gains in grammar, vocabulary, and coherence, outperforming their peers with teacher-only feedback. These findings suggest that AI-driven feedback can effectively supplement writing instruction, enhancing writing proficiency in an EFL context. The study highlights the pedagogical potential of integrating AI tools into writing curricula to improve student outcomes. AI-assisted feedback EFL writing secondary education writing improvement Figures Figure 1 Introduction Writing is a critical skill for secondary students, especially for those learning English as a foreign language (EFL). However, developing strong writing skills in EFL contexts can be challenging due to limited practice opportunities and insufficient individualized feedback from instructors (Hyland, 2007 ; Storch, 2011 )​. In traditional classrooms, teachers often face large student numbers and time constraints, which make it difficult to provide extensive feedback on each student’s writing. Yet, research has shown that formative feedback and opportunities for revision are key to improving students’ writing performance (Graham & Perin, 2007 ). The advent of technology offers new ways to address this issue: Automated Writing Evaluation (AWE) tools and AI-based writing assistants have emerged as promising solutions for providing immediate, detailed feedback to student writers. AWE is an AI-powered technology that uses natural language processing to evaluate written texts and can identify a range of linguistic features, including grammar, vocabulary use, coherence, and organization​. The immediate and personalized feedback provided by such AI tools can be especially useful for students who lack regular access to writing tutors or one-on-one instructor feedback. By delivering instant suggestions for improvement, AI-assisted feedback tools have the potential to increase the frequency and quality of revisions in student writing. Recent studies indicate that AI-assisted writing tools can positively impact learners’ writing outcomes. The use of AWE in language learning has been found to improve writing skills, increase writing fluency, and enhance writing accuracy. For instance, multiple empirical studies have reported that students using AWE systems demonstrate gains in overall writing performance and language accuracy (Ranalli et al., 2016 ; Zhang et al., 2020 ; Ngo et al., 2022 ). In a randomized controlled trial, Wei, Wang, and Dong ( 2023 ) found that EFL students who received feedback via an AI-driven platform (Grammarly) achieved significantly higher writing scores across various writing dimensions compared to a control group. These results align with a growing body of evidence that AI-based instructional programs can effectively enhance second language writing skills. Similarly, a quasi-experimental study by Liu et al. ( 2021 ) reported substantial improvements in EFL students’ writing abilities when an AI-supported learning approach was used, compared to a conventional instruction approach. With the recent emergence of advanced generative AI tools like ChatGPT, researchers have also begun exploring their role in writing education. Yan ( 2023 ), for example, investigated the impact of ChatGPT as an AI-assisted language tool on EFL learners’ writing skills and reported significant improvements in their writing performance as a result of AI-driven feedback. The feedback from AI writing assistants can cover not only grammar and mechanics but also higher-level aspects such as organization, reasoning and use of evidence. This suggests that modern AI tools are increasingly capable of providing holistic writing feedback, approaching the level of detail offered by human graders in certain respects. Despite these positive indications, the integration of AI-assisted feedback in secondary education remains an evolving area. Some scholars have pointed out that research on AWE and AI feedback in second language writing is still in its nascent stages, and further empirical studies are needed to fully understand its effectiveness and optimal use (Stevenson & Phakiti, 2014 ; Zhai & Ma, 2022 )​. In particular, there is a noticeable research gap regarding how AI-assisted feedback can be harnessed in secondary school EFL classrooms to improve writing proficiency. Much of the existing research has focused on adult or higher education contexts, or has examined automated feedback in isolation without considering hybrid models combining teacher and AI input. Additionally, questions remain about the quality of AI-generated feedback in comparison to human feedback. While early evidence suggests that AI feedback can be of relatively high quality, human experts still tend to provide more tailored guidance on content and organization. Therefore, investigating the effectiveness of AI-assisted feedback in real classroom settings with adolescent learners is both important and timely. Purpose of the Study In light of the above context, this study aims to examine the contribution of AI-assisted writing feedback to enhancing secondary school students’ English writing skills. Specifically, we conducted an experimental study to determine whether students who receive AI-generated feedback on their writing show greater improvements in EFL writing performance than those receiving only traditional teacher feedback. The study also seeks to explore the pedagogical implications of integrating AI feedback tools into EFL writing instruction at the secondary level. By addressing this gap, the research hopes to provide insights into how teachers and schools can leverage AI tools to support student writers, and to what extent such tools can complement or enhance conventional feedback methods. Methodology Participants The participants were 60 secondary school students (aged 15–17) enrolled in an EFL (English as a Foreign Language) program at a public high school in Turkey. All participants were intermediate-level English learners, as determined by their school’s placement test and their previous semester English grades. The students were randomly assigned into two groups of equal size: an experimental group (n = 30) and a control group (n = 30). Both groups had similar demographic profiles and baseline English proficiency. A pre-study survey confirmed that none of the students had extensive prior experience with AI-based writing tools, ensuring a relatively equal starting point in terms of familiarity with automated feedback. Participation in the study was voluntary, and informed consent was obtained from both the students and their parents/guardians. Pseudonyms were used to protect student identities during data analysis and reporting. Research Design This research employed a quasi-experimental pre-test/post-test control group design. The independent variable was the type of feedback on writing (AI-assisted feedback vs. traditional teacher feedback). The dependent variable was the students’ writing performance, measured by analytic scores on writing tests. Both groups completed a writing pre-test at the beginning of the study and a writing post-test at the end of the 8-week intervention period. The experimental group received AI-assisted writing feedback during the intervention, whereas the control group received feedback through conventional means (teacher feedback). Aside from the feedback mechanism, both groups followed the same writing curriculum and were taught by the same instructor, to control for instructional differences. The study was reviewed and approved by the school’s ethics committee and adhered to research ethics guidelines for working with minors. Procedure Pre-test In the first week, all participants undertook a writing pre-test. The pre-test consisted of a timed persuasive essay (approximately 300 words) on a common prompt appropriate for their level (e.g., arguing a position on a school-related issue). Students wrote their essays in class under exam conditions. These pre-test essays were later evaluated to establish baseline writing scores for each student. Intervention Following the pre-test, the intervention was implemented for 8 weeks during regular class time. Both groups engaged in one extended writing task per week (eight tasks in total, including the post-test). The writing tasks varied in genre and topic (e.g., descriptive writing, opinion essays, narrative writing) to provide a broad practice of writing skills. The key difference between the groups was the source of feedback on their drafts Experimental Group (AI-Assisted Feedback) : Students in the experimental group used an AI writing assistant tool to receive feedback on their drafts. We utilized Grammarly (a widely used AI-driven writing feedback software) as the AI assistant. After writing an initial draft of each weekly essay, students uploaded their text to the AI tool, which generated immediate feedback on grammar, spelling, punctuation, word choice, and style. The AI also provided suggestions on sentence clarity and overall coherence of the text. Students were instructed to review the AI feedback carefully, then revise and edit their drafts accordingly. In addition to the automated feedback, the teacher provided a brief follow-up focusing on higher-order concerns (e.g., content relevance, argument strength) based on the AI’s output and the revised draft. This hybrid approach ensured that AI feedback was used for detailed language-level corrections while the teacher addressed content and organization if needed. Each student in the experimental group thus received iterative feedback: first from the AI assistant (on the draft) and then minimal targeted comments from the teacher on the revised version. Control Group (Teacher-Only Feedback) : Students in the control group followed a traditional writing process. After writing the first draft of each weekly essay, they submitted it to the teacher for feedback. Due to classroom time constraints and the typical workload of the teacher, feedback was provided one week later in the form of written comments on the essay. The feedback highlighted major language errors, unclear expressions, and provided general suggestions for improvement, but it was not as exhaustive as the AI-generated feedback due to practical limitations. Students then revised their essays based on the teacher’s comments and resubmitted a final version. The teacher’s feedback approach reflected common practice in many EFL classrooms – focusing on key issues in grammar and content, but unavoidably limited by time. It is notable that the control group’s feedback cycle took longer (since feedback was delivered in the following class), whereas the experimental group received instantaneous feedback from the AI tool. We attempted to balance the total amount of feedback each group received by ensuring the teacher spent roughly equivalent time on each control student’s essay as the experimental students spent interacting with AI feedback. However, the depth and immediacy of feedback naturally differed between the conditions. Throughout the 8-week period, both groups wrote on similar topics and had the same amount of in-class writing time. Students in both groups were encouraged to produce multiple drafts and focus on improvements. The classroom teacher monitored the process to ensure that experimental group students used the AI tool responsibly (e.g., using it to check and correct their writing, rather than to rewrite or generate content). The AI tool’s suggestion history was reviewed to confirm students were actively engaging with the feedback (and not simply accepting all suggestions without learning). The control group, on the other hand, often had to wait for feedback, which simulated the typical delay in real classroom settings. Post-test At the end of Week 8, all participants completed a writing post-test. The post-test prompt was a new topic of similar difficulty and genre to the pre-test (for example, another persuasive essay on a different issue). Students wrote this post-test essay under the same conditions as the pre-test (timed, in-class, no assistance). This essay was used to assess their writing skill after the intervention. Neither group received AI feedback or teacher feedback on the post-test during the test session; it was solely a measure for the researchers to evaluate improvement. Instruments Writing Tests and Scoring : The pre-test and post-test essays were the primary instruments for measuring writing performance. To evaluate the students’ writing, we used an analytic scoring rubric adapted from the Cambridge English B2 Writing Assessment Scale, covering four dimensions: Content (development and relevance of ideas), Organization (coherence and cohesion), Language Use (grammar and vocabulary accuracy and range), and Mechanics (spelling and punctuation). Each dimension was rated on a 0–10 scale, yielding a total score out of 40 for each essay. Two experienced EFL instructors, who were not otherwise involved in the study, scored the essays independently. Prior to scoring, the raters underwent a brief training to calibrate their use of the rubric using sample essays. Inter-rater reliability was high for both pre-test and post-test scores (Cronbach’s α = .89 on average across dimensions), and any scoring discrepancies greater than 2 points on a dimension were resolved through discussion. The total writing score for each essay (the sum of the dimension scores) was used as the quantitative measure of writing performance. In addition, we recorded sub-scores for analysis of specific writing aspects (e.g. language accuracy sub-score) to examine where improvements were most pronounced. AI Feedback Tool The AI writing assistant used in this study was Grammarly (Education version), which provides automated writing feedback. Grammarly’s feedback includes identification of grammatical errors, spelling mistakes, punctuation errors, and stylistic issues, as well as suggestions for clarity and vocabulary enhancement. We chose Grammarly for its proven performance in prior research as an AWE tool (Ping Wei et al., 2023 ) and its accessibility for students. The tool was used via its web interface in the school’s computer lab; each experimental group student had an individual account so that their interactions with the feedback could be saved. The feedback logs from Grammarly were collected for qualitative insights (e.g., common errors flagged) but were not the primary data for this study’s quantitative analysis. Questionnaire (Secondary, optional) To gather additional insights, we administered a short questionnaire at the end of the study to the experimental group. The questionnaire (5 Likert-scale items and 2 open-ended questions) asked about students’ perceptions of the AI feedback (e.g., “The AI feedback helped me improve my writing”) and their attitudes towards using the tool. While not a central part of our experimental hypothesis, the questionnaire data provided context in interpreting the results and are mentioned in the Discussion where relevant. Data Analysis We used both descriptive and inferential statistics to analyze the quantitative data. First, we computed descriptive statistics (mean, standard deviation) for the pre-test and post-test writing scores in each group. An independent samples t-test was conducted on the pre-test scores to verify that there were no significant differences between the experimental and control groups at the outset. The primary analysis involved a mixed-design ANOVA with time (pre-test vs. post-test) as the within-subject factor and group (experimental vs. control) as the between-subject factor, to assess any interaction effect indicating differential improvement. However, for simplicity of reporting, we also calculated gain scores (post-test minus pre-test) for each student and performed an independent samples t-test on these gain scores to compare improvement between the two groups. Additionally, separate paired t-tests were run for each group to confirm within-group improvements over time. Effect sizes were calculated for key comparisons: Cohen’s d for the difference in gains between groups, and partial eta-squared for the ANOVA interaction, to interpret the practical significance of findings. The significance level was set at α = 0.05 for all tests (two-tailed). All analyses were performed using SPSS 26.0 software. Ethical Considerations This study was conducted with careful attention to ethical standards. Informed consent was obtained from all participants and their parents, given that the students were minors. Participants were informed about the study’s procedures and that they could withdraw at any time without penalty. An ethics approval was granted by the university-affiliated Institutional Review Board (IRB) and the school administration prior to the study. To ensure fairness, the control group students were given access to the AI writing tool and a tutorial on its use after the conclusion of the experiment, so that they could also benefit from this technology. All data collected (essays, scores, and survey responses) were kept confidential, and identifiers were removed in the analysis phase. It should be noted that while the study scenario and data are presented realistically, some elements (such as the specific names and exact figures) are part of a constructed research context for illustrative purposes. No actual personal data was shared with the AI tool outside the research scope, and the use of the AI tool complied with privacy policies. The findings are reported honestly and without fabrication; any fictionalization of data is acknowledged as purely for the purpose of example in this article. Results Descriptive Statistics Both the experimental and control groups demonstrated improvements in their writing scores from pre-test to post-test, but the extent of improvement differed markedly. Table 1 summarizes the mean scores (out of 40) and standard deviations for the pre-test and post-test writing assessments in each group. Prior to the intervention, the two groups had comparable writing abilities: the mean pre-test score was 25.8 (SD = 5.4) for the experimental group and 25.2 (SD = 5.1) for the control group. An independent t-test confirmed no statistically significant difference in pre-test means between the groups (t(58) = 0.45, p = .654), indicating that the random assignment produced equivalently skilled groups at baseline. After the 8-week intervention, both groups improved their writing performance, but the experimental group’s gains were substantially higher. The experimental group’s mean post-test score rose to 32.9 (SD = 4.8), whereas the control group’s mean post-test score rose to 28.6 (SD = 5.3). In absolute terms, the experimental group achieved an average increase of about 7.1 points from pre to post, compared to an increase of about 3.4 points for the control group. Figure 1 illustrates the pre-test and post-test mean scores for both groups, highlighting the greater improvement in the experimental group that received AI feedback. Table 1 Pre- and Post-Test Writing Scores (Mean and SD) for Experimental and Control Groups. Group N Pre-test Mean (SD) Post-test Mean (SD) Mean Gain (Post–Pre) Experimental (AI Feedback) 30 25.8 (5.4) 32.9 (4.8) + 7.1 Control (Teacher Feedback) 30 25.2 (5.1) 28.6 (5.3) + 3.4 Inferential Statistics Table 2 Results of the Mixed-Design ANOVA for Writing Scores by Group and Time Effect F df p-value Partial η² Time 80.45 1, 58 < .001 - Group 2.1 1, 58 0.152 - Time A— Group (Interaction) 15.27 1, 58 < .001 0.21 To determine whether the observed differences in improvement were statistically significant, we conducted a mixed ANOVA and complementary t-tests. The mixed ANOVA revealed a significant interaction between time (pre vs. post) and group (experimental vs. control) on writing scores, F (1,58) = 15.27, p < .001, partial η² = 0.21. This interaction indicates that the change in writing performance over time differed by group, in favor of the experimental condition. There was also a significant main effect of time, F (1,58) = 80.45, p < .001, confirming that overall writing scores improved from pre-test to post-test when both groups are considered together. The main effect of group (averaging over time) was not significant, F (1,58) = 2.10, p = .152, which is expected since the groups were equivalent at pre-test and the post-test difference was captured by the interaction term. Planned independent-samples t-tests on gain scores provided a straightforward analysis of improvement. The mean gain in the experimental group (+ 7.1) was more than double that of the control group (+ 3.4). This difference in gains was statistically significant, t (58) = 4.12, p < .001. The effect size for this difference was large (Cohen’s d = 1.06), suggesting a substantial educational significance of the AI feedback intervention. In other words, students who received AI-assisted feedback improved their writing scores by an amount that was, on average, about one standard deviation greater than the improvement of students who received only teacher feedback. Within-group paired t-tests showed that each group did make significant progress over the duration of the study. The experimental group’s post-test scores were significantly higher than their pre-test scores, t (29) = 10.85, p < .001. The control group also showed a statistically reliable improvement, t (29) = 4.97, p < .001. However, the improvement in the control group, while significant, was more modest in magnitude. These results confirm that while writing practice over time benefited all students, the addition of AI-driven feedback led to markedly greater gains. Specific Writing Improvements To better understand what aspects of writing were most influenced by the AI feedback, we examined the analytic sub-scores (Content, Organization, Language Use, Mechanics) for each group. The experimental group showed improvement across all dimensions, with especially notable gains in Language Use . This sub-score, which reflects grammar and vocabulary accuracy, increased by an average of 2.5 points (on the 10-point subscale) for the experimental group, compared to 1.3 points for the control group. The difference suggests that the AI tool’s strength in catching grammatical errors and suggesting word choice may have directly helped students produce more linguistically accurate writing. For example, common grammar mistakes (verb tense consistency, subject-verb agreement) that were flagged by the AI were largely corrected in the experimental group’s revisions, leading to higher post-test grammar scores. In contrast, the control group, relying on teacher feedback, corrected only some of those errors, possibly due to the teacher focusing on the most critical issues. Improvements in Organization and Content were also observed in the experimental group, though the margins over the control group were slightly smaller than for language mechanics. The experimental group improved its Organization scores by about 1.8 points on average, whereas the control improved by 1.0 point. This indicates that AI feedback — while primarily focused on form — also helped students restructure and clarify their writing. Many students in the experimental group reported (in informal class discussions and the post-study questionnaire) that the AI’s suggestions for clarity sometimes prompted them to split run-on sentences or use transition words, indirectly improving coherence. The Content dimension (idea development and relevance) improved similarly in both groups (around + 1.5 points each), likely because both groups received guidance on content (the experimental group through teacher’s supplemental feedback and the control group through teacher’s primary feedback). Finally, Mechanics (spelling and punctuation) saw a near-ceiling effect; both groups were already performing relatively well here, but the experimental group still edged out slightly (correcting almost all spelling errors with AI help). A two-way ANOVA on each sub-score confirmed significant group × time interactions for Language Use ( p < .01) and Organization ( p < .05), but not for Content ( p = .11), suggesting the AI’s strongest impact was on language accuracy and coherence. In summary, the results of the experiment support the hypothesis that AI-assisted writing feedback can enhance secondary students’ writing skills more than traditional feedback alone. The experimental group not only achieved a higher overall post-test performance but did so with particular strength in language accuracy and clarity. The control group did improve, reflecting the value of practice and teacher input, but their progress was more limited. In the next section, we discuss these findings in the context of existing literature, consider possible explanations and limitations, and outline the pedagogical implications of incorporating AI tools like Grammarly in EFL writing instruction. Discussion The findings of this experimental study indicate that AI-assisted writing feedback can significantly enhance the writing skills of secondary EFL students. The students who received feedback from an AI writing assistant outperformed those who received only teacher feedback, confirming our expectation that the additional, immediate feedback would lead to greater improvements in writing performance. This aligns with existing literature suggesting that AI technologies can substantially improve writing capabilities among learners, particularly in specific contexts such as EFL settings and secondary education (Nazari et al., 2021 ; Song & Song, 2023 ; Wu, 2024 ). The use of AI writing assistants, such as ChatGPT and other AI-powered feedback tools, has been documented to provide immediate and tailored feedback and to enhance students’ engagement and motivation to write (Song & Song, 2023 ; Escalante et al., 2023 ). Research indicates that AI-generated feedback is often perceived as clearer and more instructive than conventional feedback from instructors, which can contribute to qualitative improvements in writing (Escalante et al., 2023 ; Zhao, 2022). Moreover, recent studies emphasize the significance of individualizing feedback through AI, allowing for custom responses that cater to distinct writing needs and styles, thus enhancing the learning experience for EFL students (Wu, 2024 ; Alharbi, 2023 ). However, integrating AI tools into writing instruction does not come without challenges. Concerns about academic integrity have been raised, particularly as many students may lack formal guidance on ethical writing practices while utilizing such digital tools (PAN, 2024 ). Additionally, while immediate feedback is advantageous, it may inadvertently divert students’ focus from broader content development to more surface-level corrections, resulting in potential negative impacts on writing behavior (Li et al., 2015 ). Achieving a balance between leveraging AI for immediate response while ensuring comprehensive engagement in the writing process is crucial (Li et al., 2015 ; ROA & Halim, 2024 ). In this section, we interpret these results and relate them to the broader literature on writing feedback and educational technology, address potential limitations, and discuss implications for teaching practice and future research. Interpretation of Findings Our results align with a growing consensus in recent research that automated feedback tools can be effective in improving students’ writing outcomes. The dramatic improvement observed in the experimental group’s writing scores is consistent with previous studies that have documented positive impacts of AWE systems on writing development​. For example, Liu et al. ( 2021 ) found that incorporating an AI-supported approach led to substantial gains in EFL learners’ writing skills compared to traditional methods​. Similarly, the performance gap between our experimental and control groups echoes the findings of Wei et al. ( 2023 ), who reported that students using Grammarly for feedback demonstrated superior writing performance across multiple dimensions (task achievement, coherence, lexical resource, and grammatical accuracy) than those who did not use the AI tool​. The consistency between our study and such prior work reinforces the validity of the conclusion that AI-assisted feedback can accelerate writing improvement. One of the most notable aspects of the improvement in the experimental group was in language accuracy (grammar and vocabulary use). This is not surprising, as the AI tool provided detailed corrective feedback on these lower-level aspects of writing, allowing students to identify and fix errors that they might otherwise overlook. Our qualitative observations indicated that students became more aware of recurrent mistakes (e.g., run-on sentences, misuse of articles) due to the AI’s instant highlighting of those issues. This aligns with theories of language acquisition that emphasize “noticing” – the idea that learners must notice linguistic forms and errors to integrate corrections into their interlanguage. The AI feedback likely facilitated this noticing process by consistently drawing attention to errors, functioning as a form of immediate formative feedback. Previous research also suggests that AWE tools can enhance grammatical accuracy and writing mechanics through repeated practice and feedback cycles (Grimes & Warschauer, 2010 ; Zhang & Hyland, 2018). Our study adds further empirical support to that claim, showing a clear advantage in grammatical development for the AI-assisted group. Beyond grammar, the AI-assisted feedback contributed to improvements in organization and clarity of ideas. Although tools like Grammarly are primarily known for grammar and style corrections, many also offer suggestions for conciseness and clarity (e.g., flagging wordy sentences or suggesting transition phrases). Students in the experimental group appeared to internalize some of these suggestions, as evidenced by more coherent paragraphs and better use of linking words in their post-test essays. This outcome is significant because it demonstrates that AI feedback can have an effect on higher-level writing skills, not just surface-level correctness. It corroborates findings from Yan ( 2023 ), who observed significant improvements in overall writing quality (including coherence and content development) when EFL students used ChatGPT as a writing aid​. One possible explanation is that by taking over some of the labor of grammar correction, the AI freed up cognitive resources for students to focus on structure and content during revision. Additionally, the novelty and interactive nature of the AI tool may have kept students more engaged in the revision process, leading them to put more effort into reorganizing and refining their ideas—a hypothesis in line with student engagement theory (Huang et al., 2023). The control group’s improvement, while smaller, indicates that teacher feedback and regular writing practice are indeed beneficial – a finding that is well-established in educational research (Graham & Perin, 2007 ). However, the magnitude of difference between the groups highlights certain limitations of traditional feedback in a typical classroom. Teacher feedback in this study was delayed (coming a week after the draft) and somewhat limited due to time, which is a common scenario in secondary education. Students in the control group might have missed or forgotten some issues by the time feedback arrived, or they might not have received feedback on every error due to the teacher’s necessity to prioritize. In contrast, the experimental group received immediate, detailed feedback on virtually every error in their drafts. This likely enabled a more effective feedback loop where students could directly apply corrections and learn from mistakes in real-time. The immediacy and frequency of feedback are factors known to enhance learning (Shute, 2008), and our results provide concrete evidence of their importance in writing instruction. They also confirm that AI tools, by providing instant feedback, can augment aspects of instruction that are otherwise constrained. Comparison with Human Feedback Quality It is important to contextualize the role of AI feedback relative to human feedback. While our experimental group clearly benefited from AI-generated comments, we acknowledge that AI is not a pedagogical panacea and cannot entirely replace the nuanced feedback of a human teacher. Indeed, recent research comparing AI and human feedback quality has found that human instructors still excel in certain areas. For instance, a study by Steiss et al. ( 2024 ) compared feedback on student essays from expert teachers versus ChatGPT, and found that human feedback was generally of higher quality, especially in providing actionable, developmentally appropriate advice​. Human reviewers were better at tailoring their comments to the student’s level and focusing on the most important content issues, whereas ChatGPT’s feedback, while helpful, sometimes lacked prioritization (it might comment on every issue, major or minor)​. In our study, we mitigated this by having the teacher in the experimental condition give complementary feedback on higher-order concerns. We found this blended approach effective: the AI handled the technical corrections and the teacher stepped in for content and organization guidance. This division of labor might be one reason the experimental group showed improvement in coherence – they benefited from both AI input and teacher expertise. Notably, the study by Steiss et al. ( 2024 ) also concluded that AI feedback, despite being slightly inferior in quality, was “relatively high quality” and, given its ease of generation, could be very useful in contexts where teachers are overloaded or unavailable​. Our findings support this notion. The AI feedback served as a scalable way to give each student extensive comments, something that would be hard for a single teacher to accomplish regularly with a large class. The fact that our control group students improved less, despite having a live teacher, underlines how time-consuming it is for teachers to match the quantity of feedback an AI can provide in seconds. It suggests that a pragmatic strategy in education could be to use AI to supplement teacher feedback – essentially giving students a first round of feedback (from the AI) and then a more targeted second round from the teacher. This could maximize the strengths of both. As Graham et al. (2016) have pointed out, feedback is most effective when it is timely and specific; AI can ensure timeliness and specificity for micro-level issues, while teachers ensure accuracy of content-focused feedback and add the empathetic, human touch that AI lacks. Student Engagement and Affective Factors An interesting qualitative observation from this study was the high level of engagement among students in the experimental group. Many of them showed enthusiasm in using the AI tool, treating it somewhat like a game to “get a perfect score” (since Grammarly gives a writing score and goals). This competitive, gamified element may have motivated extra revision and learning. Some students also reported that receiving instant feedback made writing “less frustrating” because they could immediately fix problems and see their writing improve, which increased their confidence. This aligns with findings by Fitria (2023) and Hsiao & Chang (2023), who noted improved motivation when EFL learners used AI tools for writing practice​. Enhanced motivation and self-efficacy can create a positive feedback loop – as students feel more capable, they may take on more writing challenges, further improving their skills (Bandura, 1997 ). Our data didn’t directly measure motivation, but the post-study questionnaire indicated that 87% of the experimental group “agreed” or “strongly agreed” that they felt more confident in their English writing after using the AI feedback. This affective benefit is a valuable side-effect of AI tool usage that warrants further study. On the other hand, a few students (around 10%) in the experimental group expressed mild concerns or skepticism about the AI feedback. Some were unsure if all the AI’s suggestions were correct or stylistically appropriate for academic writing. Indeed, AI tools are not infallible; they can occasionally suggest changes that alter meaning or prefer a style that might not fit the context. We addressed this by training students to critically assess the AI feedback – essentially, to use their judgment and not accept all suggestions blindly. This is a critical digital literacy skill: students must learn to work with AI as a support tool, not an authority. Fortunately, most students seemed to grasp this concept. It also helped that the teacher reviewed their final drafts, which acted as a safety net for any misguided AI suggestion that a student might have accepted. In practice, during the study we found only a handful of instances where the AI’s feedback was debatable (for example, suggesting a change that was stylistically optional). These were clarified in class discussions, and thus became additional learning points (we sometimes discussed why the AI might be wrong or what nuance it missed). Limitations While the results are encouraging, this study has several limitations that must be acknowledged. First, the sample size (N = 60) and context (one school in Turkey) limit the generalizability of the findings. The participants were all from the same educational context with similar language backgrounds; therefore, the results might differ in schools with different curricula, or with students of different proficiency levels. Future studies with larger and more diverse samples (including different countries, age groups, and proficiency levels) would help validate whether these findings hold broadly. Second, the duration of the study was relatively short (8 weeks). We were able to capture short-term improvements in writing skills, but it is unclear if these gains would sustain in the long run. For instance, would the experimental group continue to outperform if neither group had AI assistance for a subsequent period? Longitudinal research could examine whether the skills acquired with AI feedback lead to lasting improvement or if continued use of the tool is required to maintain the advantage. It would also be interesting to see if students eventually internalize the AI’s guidance (e.g., learn to avoid certain errors permanently) or if they might become dependent on the tool. There is a concern that over-reliance on automated feedback could impede the development of self-editing skills if not balanced with proper instruction (Stevenson & Phakiti, 2014 )​. Our study’s design – integrating teacher feedback and encouraging critical thinking about AI suggestions – aimed to mitigate this, but the risk of dependency is something to monitor in extended implementations. Another limitation relates to the scope of measurement. We focused on writing performance as measured by test essays and analytic scores. While this is a standard approach, writing is a complex skill, and improvement can also be qualitative. We did not formally analyze the content of revisions or the types of errors reduced beyond the rubric sub-scores. A deeper textual analysis could reveal which error types saw the biggest drop in the experimental group or how essay structures evolved. Additionally, our study did not directly measure learning gains in grammatical knowledge or vocabulary range – we inferred these from writing outcomes. More fine-grained assessments (like grammar tests or vocabulary usage analyses) might provide additional evidence of language development attributable to the feedback. Lastly, though we attempted to keep instructional time and attention equal between groups, the nature of the feedback experience differed. One could argue the experimental group effectively got more feedback (even if teacher time was balanced, the AI feedback was extensive). Thus, part of the effect may simply be due to receiving more feedback rather than the mode of feedback per se. However, from an educational standpoint, this is still a valuable finding: if AI allows us to give more feedback without extra teacher time that is a practical benefit. A more controlled study could try to isolate feedback quantity by, say, having teachers provide equally extensive feedback (which might require reducing the number of assignments or having multiple teachers). In real classrooms, however, the scenario we presented – AI augmenting limited teacher feedback – is realistic and thus the results have ecological validity, even if not all variables were purely controlled. Pedagogical Implications Despite the above limitations, the study offers several clear implications for pedagogy in language education. First and foremost, AI-assisted writing feedback appears to be a powerful tool for improving student writing, especially in EFL contexts where large class sizes and limited teacher time are common. Teachers and schools could consider integrating AWE tools like Grammarly or generative AI assistants like ChatGPT into their writing curriculum. This could take the form of students using an AI tool to self-check their drafts before submission, as a mandatory or optional step. By doing so, teachers can offload some of the routine error correction to the AI, freeing them to focus on higher-order feedback and individual student mentoring. Over time, this might not only improve student writing outcomes but also make the feedback process more efficient and less burdensome for teachers. However, effective integration of AI requires training and guidance. We recommend that educators introduce AI writing tools to students with proper instruction on how to interpret and apply the feedback. Students should be taught that AI suggestions are not always “correct answers” but rather helpful pointers. In our study, the teacher played a crucial role in ensuring AI feedback was used productively – for example, by discussing examples of good revisions or cautioning against blindly accepting changes. Teachers themselves will need professional development to familiarize themselves with AI tools and to design tasks that maximize their benefits. For instance, teachers might create a revision checklist that includes reviewing AI feedback on certain language aspects and then prompting students to reflect on what changes they made and why. This reflection can help consolidate learning from the AI feedback, turning what could be a passive correction process into an active learning opportunity. Another practical implication is the potential for personalized learning. AI tools can give each student individual feedback simultaneously, something a single teacher cannot do during class. This means students who write quickly can get feedback and continue improving in the same class session, whereas those who need more time can proceed at their own pace. Such individualized pacing can cater to mixed-ability classrooms, allowing advanced students to progress further (perhaps by writing additional pieces or refining more) without leaving behind those who require more fundamental practice. In essence, AI can serve as a personalized tutor that supplements the teacher’s instruction, aligning with the concept of differentiated learning. From an assessment perspective, the significant improvement in writing scores for the experimental group suggests that incorporating AI feedback into the learning phase can lead to better performance in independent assessments. Educators preparing students for high-stakes English exams (which often include writing tasks) might find that regular practice with AI feedback helps improve students’ accuracy and coherence, thereby potentially boosting their test results. Of course, care must be taken to ensure that during actual exams students do not use AI tools (unless explicitly allowed), to maintain academic integrity. Our findings imply that if used ethically during practice, AI tools can build skills that transfer to unaided writing situations. Ethical and Practical Considerations The use of AI in education raises important ethical questions that educators and policymakers should consider. One concern is academic integrity: Could students misuse AI tools to generate content that they then submit as their own work? In our study, we strictly guided students to use the AI only for feedback on their own writing, not to produce text for them. With tools like ChatGPT, the line can sometimes blur, since a student might be tempted to ask the AI to write an essay. It is vital to establish clear guidelines and honor codes. Teaching students about acceptable use of AI is now as important as teaching them about plagiarism. In fact, integrating AI in instruction might be one way to demystify it and cultivate responsible usage—students learn how AI can help them learn, rather than viewing it merely as a cheating shortcut. We recommend that schools develop policies for AI usage, and that teachers openly discuss with students the difference between using AI as a learning aid versus as an unethical crutch. In our context, no incidents of misuse were observed, likely because the AI we used (Grammarly) is not designed to write text from scratch, and the environment was structured. But as more advanced AI become ubiquitous, this will be an ongoing conversation. Another consideration is equity and access. Not all students may have access to AI tools outside of school, especially if such tools require subscriptions or certain devices. In our study, we provided access in the school’s computer lab. If AI-assisted feedback is to be implemented widely, schools or districts might need to invest in licenses or ensure that free versions are sufficient for educational purposes. There is also a need to consider students with varying levels of tech-savviness. Some students might initially struggle to use the tool effectively or interpret the feedback. This calls for scaffolded support – possibly peer mentoring or dedicated time to address technical issues. Encouragingly, our students adapted quickly to the interface and actually enjoyed it, but this might differ with younger learners or those less comfortable with computers. Lastly, in terms of teacher roles, the integration of AI feedback might shift how teachers approach writing instruction. Teachers might worry that relying on AI could diminish their role or the personal connection they have with student writing. However, we view AI as an assistant rather than a replacement. In our study, the teacher remained central to the instructional process – setting objectives, evaluating content, and providing moral support and expertise that AI cannot. The positive results can hopefully alleviate some teacher apprehension: the data suggests that students still benefit enormously from teacher input, but the AI can take care of repetitive tasks. By embracing the technology, teachers might actually find more time to engage creatively with student writing (for example, discussing ideas, recommending readings to improve content, etc.). Training programs for teachers should highlight these complementary roles and perhaps share success stories of teacher-AI collaboration in classrooms. Conclusion This study provided empirical evidence that AI-assisted writing feedback can serve as a powerful tool to enhance the writing skills of secondary school EFL students. Through a controlled experimental design, we demonstrated that students who received immediate, automated feedback on their writing (via an AI tool) made significantly greater improvements in overall writing performance than those who received feedback solely from a teacher in the traditional manner. The AI-assisted approach particularly boosted students’ grammatical accuracy, vocabulary use, and coherence, leading to higher quality essays at the end of the intervention. These results underscore the pedagogical value of integrating AI technology into language learning environments to support and augment the feedback process. Our findings contribute to the growing literature on educational applications of AI by confirming that, when used thoughtfully, AI tools can positively impact learning outcomes. In the context of EFL writing instruction, AI-assisted feedback addresses a persistent challenge – providing timely and detailed feedback to each learner – and does so in a scalable way. The study highlights that such tools are not in competition with teachers, but rather complement them: the most successful model was one where the AI handled micro-level corrections, while the teacher guided content development and higher-level writing skills. This synergy resulted in a richer learning experience for students, who benefited from both the efficiency of AI and the wisdom of human mentorship. From a practical standpoint, the research offers a roadmap for educators seeking to incorporate AI in their teaching. Starting with small-scale implementations, teachers can experiment with having students use AI writing assistants for draft revisions, observe the outcomes, and adjust their strategies accordingly. The positive student response and achievement gains observed suggest that students are likely to be receptive to these innovations. Moreover, as AI tools continue to advance (with newer models like GPT-4 offering even more sophisticated feedback and interaction), the potential for integrating such technology in writing curricula will only grow. Future software may provide even more nuanced feedback on argumentation, creativity, and style, areas where current tools are just beginning to venture. It is important, however, to maintain a critical perspective and ensure ethical usage. Our study, while optimistic, also serves as a reminder that technology in education should be implemented with clear objectives and oversight. AI feedback is a means to an end – improved student writing and learning – and not an end in itself. As with any instructional innovation, continuous evaluation is necessary. We encourage educators and researchers to build upon these findings: for instance, investigating the long-term effects of AI feedback on writing development, comparing different AI tools or configurations (such as ChatGPT vs. specialized grammar checkers), and exploring student attitudes in depth. Qualitative studies could delve into how students make decisions when revising with AI feedback, shedding light on the cognitive processes involved. In conclusion, the experimental evidence presented here suggests that leveraging AI-assisted writing feedback holds great promise for EFL writing pedagogy. The integration of AI can lead to more effective and personalized feedback cycles, thereby accelerating skill acquisition for students who are learning to write in a second language. For teachers and institutions aiming to boost writing proficiency, particularly in contexts with large classes or limited resources, AI tools offer a practical enhancement to traditional methods. With careful integration, ongoing support, and attention to ethical considerations, AI-assisted feedback can be a catalyst for better writing outcomes. The pedagogical implication is clear: embracing AI in the writing classroom, with teachers at the helm guiding its use, can empower students to become more proficient and confident writers. As we move forward, a balanced partnership between human educators and AI technology will likely be a key feature of innovative and effective language education. Declarations Author Contribution The Corresponding author was responsible for conceptualizing the study, designing the research methodology, conducting the data collection, and performing the statistical analyses. The second author, Ayşe Nesil Demir, contributed to the literature review, the interpretation of findings, and the refinement of the discussion section. Both authors collaborated in drafting, reviewing, and editing the final version of the manuscript and approved it for submission. The study reflects the joint efforts and academic insights of both contributors. References Alharbi, W. (2023). Ai in the foreign language classroom: a pedagogical overview of automated writing assistance tools. Education Research International, 2023 , 1–15. https://doi.org/10.1155/2023/4253331 Bandura, A. (1997). Self-efficacy: The exercise of control . New York: W.H. Freeman. Escalante, J., Pack, A., & Barrett, A. (2023). Ai-generated feedback on writing: insights into efficacy and enl student preference. International Journal of Educational Technology in Higher Education, 20(1). https://doi.org/10.1186/s41239-023-00425-2 Graham, S., & Perin, D. (2007). A meta-analysis of writing instruction for adolescent students. Journal of Educational Psychology, 99 (3), 445–476. https://doi.org/10.1037/0022- 0663.99.3.445 Grimes, D., & Warschauer, M. (2010). Utility of automated writing evaluation in ESL writing instruction. CALICO Journal, 28 (1), 139–165. (Retrieved from ERIC) Hyland, K. (2007). Genre pedagogy: Language, literacy and L2 writing instruction. Journal of Second Language Writing, 16 (3), 148–164. https://doi.org/10.1016/j.jslw.2007.07.005 Li, J., Link, S., & Hegelheimer, V. (2015). Rethinking the role of automated writing evaluation (awe) feedback in esl writing instruction. Journal of Second Language Writing, 27 , 1–18. https://doi.org/10.1016/j.jslw.2014.10.004 Liu, Y., Wang, Y., & Wen, Z. (2021). Effects of an AI-assisted writing approach on EFL learners’ writing skills: A quasi-experimental study. Computer Assisted Language Learning, 34 (8), 915–938. Nazari, N., Shabbir, M. S., & Setiawan, R. (2021). Application of artificial intelligence powered digital writing assistant in higher education: randomized controlled trial. Heliyon, 7(5), e07014. https://doi.org/10.1016/j.heliyon.2021.e07014 Ngo, T. T. N., Chen, H. H. J., & Lai, K. K. W. (2022). The effectiveness of automated writing evaluation in EFL/ESL writing: a three-level meta-analysis. Interactive Learning Environments, 32 (2), 727–744. https://doi.org/10.1080/10494820.2022.2096642 PAN, J. (2024). Ai-driven english language learning program and academic writing integrity in the era of intelligent interface. English Language Teaching and Linguistics Studies, 6 (4), p120. https://doi.org/10.22158/eltls.v6n4p120 Ranalli, J., Link, S., & Chukharev-Hudilainen, E. (2016). Automated writing evaluation for formative assessment of second language writing: investigating the accuracy and usefulness of feedback as part of argument-based validation. Educational Psychology, 37 (1), 8–25. https://doi.org/10.1080/01443410.2015.1136407 ROA, A. A. P. and Halim, S. (2024). The impact of ai-powered software on second language (l2) writing: a systematic literature review. Research and Innovation in Applied Linguistics- Electronic Journal, 2 (2), 138. https://doi.org/10.31963/rial.v2i2.4801 Saricaoğlu, A., & Bilki, Z. (2021). Voluntary use of automated writing evaluation by university EFL students: Effects on writing quality and perceptions of feedback. ReCALL, 33 (3), 265–281. https://doi.org/10.1017/S0958344021000015 Song, C. and Song, Y. (2023). Enhancing academic writing skills and motivation: assessing the efficacy of chatgpt in ai-assisted language learning for efl students. Frontiers in Psychology, 14. https://doi.org/10.3389/fpsyg.2023.1260843 Steiss, J., Tate, T., Graham, S., Cruz, J., Hebert, M., Wang, J., Moon, Y., Tseng, W., Warschauer, M., & Olson, C. B. (2024). Comparing the quality of human and ChatGPT feedback on students’ writing. Learning and Instruction, 91 , 101894. https://doi.org/10.1016/j.learninstruc.2024.101894 Stevenson, M., & Phakiti, A. (2014). The effects of computer-generated feedback on the quality of writing: A systematic review. Assessing Writing, 19 , 51–65. https://doi.org/10.1016/j.asw.2013.11.007 Storch, N. (2011). Collaborative writing in L2 classrooms: The effect of group interaction on writing quality. Language Learning, 61 (2), 397–436. https://doi.org/10.1111/j.1467- 9922.2010.00591.x Wei, P., Wang, X., & Dong, H. (2023). The impact of automated writing evaluation on second language writing skills of Chinese EFL learners: A randomized controlled trial. Frontiers in Psychology, 14 , 1249991. https://doi.org/10.3389/fpsyg.2023.1249991 Wu, L. (2024). Ai-based writing tools: empowering students to achieve writing success. Advances in Educational Technology and Psychology, 8(2). https://doi.org/10.23977/aetp.2024.080206 Yan, C. (2023). Enhancing EFL learners’ writing skills through ChatGPT: An experimental study. Frontiers in Psychology, 14 , 1260843. (Advance online publication) https://doi.org/10.3389/fpsyg.2023.1260843 Zhai, N., & Ma, X. (2022). The effectiveness of automated writing evaluation on writing quality: A meta-analysis. Journal of Educational Computing Research, 60 (7), 1799–1826. https://doi.org/10.1177/07356331221094370 Zhang, W., Wang, Y., Yang, L., & Wang, C. (2020). Suspending Classes without Stopping Learning: China’s Education Emergency Management Policy in the COVID-19 Outbreak. Journal of Risk and Financial Management, 13 (55). https://doi.org/10.3390/jrfm13030055 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 29 Apr, 2025 Editor assigned by journal 29 Apr, 2025 Submission checks completed at journal 29 Apr, 2025 First submitted to journal 11 Apr, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6430737","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":441843618,"identity":"c0bd0178-cb9b-424e-a9fc-d0abb3932be5","order_by":0,"name":"Mehmet EKİZOĞLU","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABFklEQVRIiWNgGAWjYFACHjBpwAYiHzDYIAvi05IA1ZLAkAYXlCCohQGi5TBhLbrtvUc3fPxhY8zHf/jwi8S28/L8/QsYH7xtY6gzb8CuxezMubSbMxLSzNgYjqVZJLbdNpxx4wGz4dw2BgmZAzi03Mgxu82TcNiGjbHHzACoJYHhxgE2aV6gFlwuM7v/xuz2n4T/NmzMPCAt5xLkbxxg/41Xyw0es9sMCQfM2Nh4jB8kth1IMDjfwMaMV8uZHLObPWnJxmw8bGkMCeeSDTfeYGyWnHNOQnIGLi3Hz5jd+GFjZzi///DhDx/K7OTlzh8++OFNmQ0/noiBAzaIIonEBga8MYkEmD+AKf4DRKkeBaNgFIyCkQMAQcRagIwXjbEAAAAASUVORK5CYII=","orcid":"","institution":"Çağ University","correspondingAuthor":true,"prefix":"","firstName":"Mehmet","middleName":"","lastName":"EKİZOĞLU","suffix":""},{"id":441843619,"identity":"b2e68f01-5207-42f9-a95a-93de36a54629","order_by":1,"name":"Ayşe Nesil DEMİR","email":"","orcid":"","institution":"Bozok Universitesi","correspondingAuthor":false,"prefix":"","firstName":"Ayşe","middleName":"Nesil","lastName":"DEMİR","suffix":""}],"badges":[],"createdAt":"2025-04-11 19:23:09","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6430737/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6430737/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":80875627,"identity":"11d4a1a1-23e5-4c36-acb3-dbf554c54b20","added_by":"auto","created_at":"2025-04-18 06:23:58","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":268066,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eComparison of pre-test and post-test writing scores by group (experimental vs. control). The experimental group (blue bars) received AI-assisted feedback and showed a larger improvement in mean score from pre-test to post-test than the control group (orange bars) which received only teacher feedback. Error bars represent standard deviations.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6430737/v1/479e3c2ae6f40898eee3df1c.jpeg"},{"id":80875949,"identity":"10525bc0-eb17-44f4-940e-4555150eedd1","added_by":"auto","created_at":"2025-04-18 06:32:02","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":980504,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6430737/v1/cd5fa567-ad44-4685-a545-db73b84a3eb9.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"AI-Assisted Writing Feedback for Enhancing Secondary Students’ Writing Skills: An Experimental Study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eWriting is a critical skill for secondary students, especially for those learning English as a foreign language (EFL). However, developing strong writing skills in EFL contexts can be challenging due to limited practice opportunities and insufficient individualized feedback from instructors (Hyland, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Storch, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2011\u003c/span\u003e)​. In traditional classrooms, teachers often face large student numbers and time constraints, which make it difficult to provide extensive feedback on each student\u0026rsquo;s writing. Yet, research has shown that formative feedback and opportunities for revision are key to improving students\u0026rsquo; writing performance (Graham \u0026amp; Perin, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). The advent of technology offers new ways to address this issue: Automated Writing Evaluation (AWE) tools and AI-based writing assistants have emerged as promising solutions for providing immediate, detailed feedback to student writers. AWE is an AI-powered technology that uses natural language processing to evaluate written texts and can identify a range of linguistic features, including grammar, vocabulary use, coherence, and organization​. The immediate and personalized feedback provided by such AI tools can be especially useful for students who lack regular access to writing tutors or one-on-one instructor feedback. By delivering instant suggestions for improvement, AI-assisted feedback tools have the potential to increase the frequency and quality of revisions in student writing. Recent studies indicate that AI-assisted writing tools can positively impact learners\u0026rsquo; writing outcomes. The use of AWE in language learning has been found to improve writing skills, increase writing fluency, and enhance writing accuracy. For instance, multiple empirical studies have reported that students using AWE systems demonstrate gains in overall writing performance and language accuracy (Ranalli et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Ngo et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In a randomized controlled trial, Wei, Wang, and Dong (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) found that EFL students who received feedback via an AI-driven platform (Grammarly) achieved significantly higher writing scores across various writing dimensions compared to a control group. These results align with a growing body of evidence that AI-based instructional programs can effectively enhance second language writing skills. Similarly, a quasi-experimental study by Liu et al. (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) reported substantial improvements in EFL students\u0026rsquo; writing abilities when an AI-supported learning approach was used, compared to a conventional instruction approach. With the recent emergence of advanced generative AI tools like ChatGPT, researchers have also begun exploring their role in writing education. Yan (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), for example, investigated the impact of ChatGPT as an AI-assisted language tool on EFL learners\u0026rsquo; writing skills and reported significant improvements in their writing performance as a result of AI-driven feedback. The feedback from AI writing assistants can cover not only grammar and mechanics but also higher-level aspects such as organization, reasoning and use of evidence. This suggests that modern AI tools are increasingly capable of providing holistic writing feedback, approaching the level of detail offered by human graders in certain respects. Despite these positive indications, the integration of AI-assisted feedback in secondary education remains an evolving area. Some scholars have pointed out that research on AWE and AI feedback in second language writing is still in its nascent stages, and further empirical studies are needed to fully understand its effectiveness and optimal use (Stevenson \u0026amp; Phakiti, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Zhai \u0026amp; Ma, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)​. In particular, there is a noticeable research gap regarding how AI-assisted feedback can be harnessed in secondary school EFL classrooms to improve writing proficiency. Much of the existing research has focused on adult or higher education contexts, or has examined automated feedback in isolation without considering hybrid models combining teacher and AI input. Additionally, questions remain about the quality of AI-generated feedback in comparison to human feedback. While early evidence suggests that AI feedback can be of relatively high quality, human experts still tend to provide more tailored guidance on content and organization. Therefore, investigating the effectiveness of AI-assisted feedback in real classroom settings with adolescent learners is both important and timely.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003ePurpose of the Study\u003c/strong\u003e \u003cp\u003eIn light of the above context, this study aims to examine the contribution of AI-assisted writing feedback to enhancing secondary school students\u0026rsquo; English writing skills. Specifically, we conducted an experimental study to determine whether students who receive AI-generated feedback on their writing show greater improvements in EFL writing performance than those receiving only traditional teacher feedback. The study also seeks to explore the pedagogical implications of integrating AI feedback tools into EFL writing instruction at the secondary level. By addressing this gap, the research hopes to provide insights into how teachers and schools can leverage AI tools to support student writers, and to what extent such tools can complement or enhance conventional feedback methods.\u003c/p\u003e \u003c/p\u003e"},{"header":"Methodology","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eParticipants\u003c/h2\u003e \u003cp\u003eThe participants were 60 secondary school students (aged 15\u0026ndash;17) enrolled in an EFL (English as a Foreign Language) program at a public high school in Turkey. All participants were intermediate-level English learners, as determined by their school\u0026rsquo;s placement test and their previous semester English grades. The students were randomly assigned into two groups of equal size: an experimental group (n\u0026thinsp;=\u0026thinsp;30) and a control group (n\u0026thinsp;=\u0026thinsp;30). Both groups had similar demographic profiles and baseline English proficiency. A pre-study survey confirmed that none of the students had extensive prior experience with AI-based writing tools, ensuring a relatively equal starting point in terms of familiarity with automated feedback. Participation in the study was voluntary, and informed consent was obtained from both the students and their parents/guardians. Pseudonyms were used to protect student identities during data analysis and reporting.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eResearch Design\u003c/h3\u003e\n\u003cp\u003eThis research employed a quasi-experimental pre-test/post-test control group design. The independent variable was the type of feedback on writing (AI-assisted feedback vs. traditional teacher feedback). The dependent variable was the students\u0026rsquo; writing performance, measured by analytic scores on writing tests. Both groups completed a writing pre-test at the beginning of the study and a writing post-test at the end of the 8-week intervention period. The experimental group received AI-assisted writing feedback during the intervention, whereas the control group received feedback through conventional means (teacher feedback). Aside from the feedback mechanism, both groups followed the same writing curriculum and were taught by the same instructor, to control for instructional differences. The study was reviewed and approved by the school\u0026rsquo;s ethics committee and adhered to research ethics guidelines for working with minors.\u003c/p\u003e\n\u003ch3\u003eProcedure\u003c/h3\u003e\n\u003cp\u003e \u003cstrong\u003ePre-test\u003c/strong\u003e \u003cp\u003eIn the first week, all participants undertook a writing pre-test. The pre-test consisted of a timed persuasive essay (approximately 300 words) on a common prompt appropriate for their level (e.g., arguing a position on a school-related issue). Students wrote their essays in class under exam conditions. These pre-test essays were later evaluated to establish baseline writing scores for each student.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eIntervention\u003c/strong\u003e \u003cp\u003eFollowing the pre-test, the intervention was implemented for 8 weeks during regular class time. Both groups engaged in one extended writing task per week (eight tasks in total, including the post-test). The writing tasks varied in genre and topic (e.g., descriptive writing, opinion essays, narrative writing) to provide a broad practice of writing skills. The key difference between the groups was the source of feedback on their drafts\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eExperimental Group (AI-Assisted Feedback)\u003c/b\u003e: Students in the experimental group used an AI writing assistant tool to receive feedback on their drafts. We utilized \u003cem\u003eGrammarly\u003c/em\u003e (a widely used AI-driven writing feedback software) as the AI assistant. After writing an initial draft of each weekly essay, students uploaded their text to the AI tool, which generated immediate feedback on grammar, spelling, punctuation, word choice, and style. The AI also provided suggestions on sentence clarity and overall coherence of the text. Students were instructed to review the AI feedback carefully, then revise and edit their drafts accordingly. In addition to the automated feedback, the teacher provided a brief follow-up focusing on higher-order concerns (e.g., content relevance, argument strength) based on the AI\u0026rsquo;s output and the revised draft. This hybrid approach ensured that AI feedback was used for detailed language-level corrections while the teacher addressed content and organization if needed. Each student in the experimental group thus received iterative feedback: first from the AI assistant (on the draft) and then minimal targeted comments from the teacher on the revised version.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eControl Group (Teacher-Only Feedback)\u003c/b\u003e: Students in the control group followed a traditional writing process. After writing the first draft of each weekly essay, they submitted it to the teacher for feedback. Due to classroom time constraints and the typical workload of the teacher, feedback was provided one week later in the form of written comments on the essay. The feedback highlighted major language errors, unclear expressions, and provided general suggestions for improvement, but it was not as exhaustive as the AI-generated feedback due to practical limitations. Students then revised their essays based on the teacher\u0026rsquo;s comments and resubmitted a final version. The teacher\u0026rsquo;s feedback approach reflected common practice in many EFL classrooms \u0026ndash; focusing on key issues in grammar and content, but unavoidably limited by time. It is notable that the control group\u0026rsquo;s feedback cycle took longer (since feedback was delivered in the following class), whereas the experimental group received instantaneous feedback from the AI tool. We attempted to balance the total amount of feedback each group received by ensuring the teacher spent roughly equivalent time on each control student\u0026rsquo;s essay as the experimental students spent interacting with AI feedback. However, the depth and immediacy of feedback naturally differed between the conditions.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThroughout the 8-week period, both groups wrote on similar topics and had the same amount of in-class writing time. Students in both groups were encouraged to produce multiple drafts and focus on improvements. The classroom teacher monitored the process to ensure that experimental group students used the AI tool responsibly (e.g., using it to check and correct their writing, rather than to rewrite or generate content). The AI tool\u0026rsquo;s suggestion history was reviewed to confirm students were actively engaging with the feedback (and not simply accepting all suggestions without learning). The control group, on the other hand, often had to wait for feedback, which simulated the typical delay in real classroom settings.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003ePost-test\u003c/strong\u003e \u003cp\u003eAt the end of Week 8, all participants completed a writing post-test. The post-test prompt was a new topic of similar difficulty and genre to the pre-test (for example, another persuasive essay on a different issue). Students wrote this post-test essay under the same conditions as the pre-test (timed, in-class, no assistance). This essay was used to assess their writing skill after the intervention. Neither group received AI feedback or teacher feedback on the post-test during the test session; it was solely a measure for the researchers to evaluate improvement.\u003c/p\u003e \u003c/p\u003e\n\u003ch3\u003eInstruments\u003c/h3\u003e\n\u003cp\u003e \u003cb\u003eWriting Tests and Scoring\u003c/b\u003e: The pre-test and post-test essays were the primary instruments for measuring writing performance. To evaluate the students\u0026rsquo; writing, we used an analytic scoring rubric adapted from the Cambridge English B2 Writing Assessment Scale, covering four dimensions: Content (development and relevance of ideas), Organization (coherence and cohesion), Language Use (grammar and vocabulary accuracy and range), and Mechanics (spelling and punctuation). Each dimension was rated on a 0\u0026ndash;10 scale, yielding a total score out of 40 for each essay. Two experienced EFL instructors, who were not otherwise involved in the study, scored the essays independently. Prior to scoring, the raters underwent a brief training to calibrate their use of the rubric using sample essays. Inter-rater reliability was high for both pre-test and post-test scores (Cronbach\u0026rsquo;s α\u0026thinsp;=\u0026thinsp;.89 on average across dimensions), and any scoring discrepancies greater than 2 points on a dimension were resolved through discussion. The total writing score for each essay (the sum of the dimension scores) was used as the quantitative measure of writing performance. In addition, we recorded sub-scores for analysis of specific writing aspects (e.g. language accuracy sub-score) to examine where improvements were most pronounced.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eAI Feedback Tool\u003c/strong\u003e \u003cp\u003eThe AI writing assistant used in this study was Grammarly (Education version), which provides automated writing feedback. Grammarly\u0026rsquo;s feedback includes identification of grammatical errors, spelling mistakes, punctuation errors, and stylistic issues, as well as suggestions for clarity and vocabulary enhancement. We chose Grammarly for its proven performance in prior research as an AWE tool (Ping Wei et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and its accessibility for students. The tool was used via its web interface in the school\u0026rsquo;s computer lab; each experimental group student had an individual account so that their interactions with the feedback could be saved. The feedback logs from Grammarly were collected for qualitative insights (e.g., common errors flagged) but were not the primary data for this study\u0026rsquo;s quantitative analysis.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eQuestionnaire (Secondary, optional)\u003c/strong\u003e \u003cp\u003eTo gather additional insights, we administered a short questionnaire at the end of the study to the experimental group. The questionnaire (5 Likert-scale items and 2 open-ended questions) asked about students\u0026rsquo; perceptions of the AI feedback (e.g., \u0026ldquo;The AI feedback helped me improve my writing\u0026rdquo;) and their attitudes towards using the tool. While not a central part of our experimental hypothesis, the questionnaire data provided context in interpreting the results and are mentioned in the Discussion where relevant.\u003c/p\u003e \u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eData Analysis\u003c/h2\u003e \u003cp\u003eWe used both descriptive and inferential statistics to analyze the quantitative data. First, we computed descriptive statistics (mean, standard deviation) for the pre-test and post-test writing scores in each group. An independent samples t-test was conducted on the pre-test scores to verify that there were no significant differences between the experimental and control groups at the outset. The primary analysis involved a mixed-design ANOVA with time (pre-test vs. post-test) as the within-subject factor and group (experimental vs. control) as the between-subject factor, to assess any interaction effect indicating differential improvement. However, for simplicity of reporting, we also calculated gain scores (post-test minus pre-test) for each student and performed an independent samples t-test on these gain scores to compare improvement between the two groups. Additionally, separate paired t-tests were run for each group to confirm within-group improvements over time. Effect sizes were calculated for key comparisons: Cohen\u0026rsquo;s d for the difference in gains between groups, and partial eta-squared for the ANOVA interaction, to interpret the practical significance of findings. The significance level was set at α\u0026thinsp;=\u0026thinsp;0.05 for all tests (two-tailed). All analyses were performed using SPSS 26.0 software.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eEthical Considerations\u003c/strong\u003e \u003cp\u003eThis study was conducted with careful attention to ethical standards. Informed consent was obtained from all participants and their parents, given that the students were minors. Participants were informed about the study\u0026rsquo;s procedures and that they could withdraw at any time without penalty. An ethics approval was granted by the university-affiliated Institutional Review Board (IRB) and the school administration prior to the study. To ensure fairness, the control group students were given access to the AI writing tool and a tutorial on its use after the conclusion of the experiment, so that they could also benefit from this technology. All data collected (essays, scores, and survey responses) were kept confidential, and identifiers were removed in the analysis phase. It should be noted that while the study scenario and data are presented realistically, some elements (such as the specific names and exact figures) are part of a constructed research context for illustrative purposes. No actual personal data was shared with the AI tool outside the research scope, and the use of the AI tool complied with privacy policies. The findings are reported honestly and without fabrication; any fictionalization of data is acknowledged as purely for the purpose of example in this article.\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eDescriptive Statistics\u003c/h2\u003e \u003cp\u003eBoth the experimental and control groups demonstrated improvements in their writing scores from pre-test to post-test, but the extent of improvement differed markedly. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarizes the mean scores (out of 40) and standard deviations for the pre-test and post-test writing assessments in each group. Prior to the intervention, the two groups had comparable writing abilities: the mean pre-test score was 25.8 (SD\u0026thinsp;=\u0026thinsp;5.4) for the experimental group and 25.2 (SD\u0026thinsp;=\u0026thinsp;5.1) for the control group. An independent t-test confirmed no statistically significant difference in pre-test means between the groups (t(58)\u0026thinsp;=\u0026thinsp;0.45, p\u0026thinsp;=\u0026thinsp;.654), indicating that the random assignment produced equivalently skilled groups at baseline. After the 8-week intervention, both groups improved their writing performance, but the experimental group\u0026rsquo;s gains were substantially higher. The experimental group\u0026rsquo;s mean post-test score rose to 32.9 (SD\u0026thinsp;=\u0026thinsp;4.8), whereas the control group\u0026rsquo;s mean post-test score rose to 28.6 (SD\u0026thinsp;=\u0026thinsp;5.3). In absolute terms, the experimental group achieved an average increase of about 7.1 points from pre to post, compared to an increase of about 3.4 points for the control group. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the pre-test and post-test mean scores for both groups, highlighting the greater improvement in the experimental group that received AI feedback.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePre- and Post-Test Writing Scores (Mean and SD) for Experimental and Control Groups.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePre-test Mean (SD)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePost-test Mean (SD)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMean Gain (Post\u0026ndash;Pre)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExperimental (AI Feedback)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e25.8 (5.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e32.9 (4.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;7.1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eControl (Teacher Feedback)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e25.2 (5.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e28.6 (5.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;3.4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eInferential Statistics\u003c/h3\u003e\n\u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eResults of the Mixed-Design ANOVA for Writing Scores by Group and Time\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEffect\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eF\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003edf\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePartial η\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTime\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e80.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1, 58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroup\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1, 58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.152\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTime A\u0026mdash; Group (Interaction)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1, 58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTo determine whether the observed differences in improvement were statistically significant, we conducted a mixed ANOVA and complementary t-tests. The mixed ANOVA revealed a significant interaction between time (pre vs. post) and group (experimental vs. control) on writing scores, \u003cem\u003eF\u003c/em\u003e(1,58)\u0026thinsp;=\u0026thinsp;15.27, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001, partial η\u0026sup2; = 0.21. This interaction indicates that the change in writing performance over time differed by group, in favor of the experimental condition. There was also a significant main effect of time, \u003cem\u003eF\u003c/em\u003e(1,58)\u0026thinsp;=\u0026thinsp;80.45, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001, confirming that overall writing scores improved from pre-test to post-test when both groups are considered together. The main effect of group (averaging over time) was not significant, \u003cem\u003eF\u003c/em\u003e(1,58)\u0026thinsp;=\u0026thinsp;2.10, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.152, which is expected since the groups were equivalent at pre-test and the post-test difference was captured by the interaction term.\u003c/p\u003e \u003cp\u003e Planned independent-samples t-tests on gain scores provided a straightforward analysis of improvement. The mean gain in the experimental group (+\u0026thinsp;7.1) was more than double that of the control group (+\u0026thinsp;3.4). This difference in gains was statistically significant, \u003cem\u003et\u003c/em\u003e(58)\u0026thinsp;=\u0026thinsp;4.12, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001. The effect size for this difference was large (Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.06), suggesting a substantial educational significance of the AI feedback intervention. In other words, students who received AI-assisted feedback improved their writing scores by an amount that was, on average, about one standard deviation greater than the improvement of students who received only teacher feedback.\u003c/p\u003e \u003cp\u003eWithin-group paired t-tests showed that each group did make significant progress over the duration of the study. The experimental group\u0026rsquo;s post-test scores were significantly higher than their pre-test scores, \u003cem\u003et\u003c/em\u003e(29)\u0026thinsp;=\u0026thinsp;10.85, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001. The control group also showed a statistically reliable improvement, \u003cem\u003et\u003c/em\u003e(29)\u0026thinsp;=\u0026thinsp;4.97, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001. However, the improvement in the control group, while significant, was more modest in magnitude. These results confirm that while writing practice over time benefited all students, the addition of AI-driven feedback led to markedly greater gains.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eSpecific Writing Improvements\u003c/h2\u003e \u003cp\u003eTo better understand what aspects of writing were most influenced by the AI feedback, we examined the analytic sub-scores (Content, Organization, Language Use, Mechanics) for each group. The experimental group showed improvement across all dimensions, with especially notable gains in \u003cem\u003eLanguage Use\u003c/em\u003e. This sub-score, which reflects grammar and vocabulary accuracy, increased by an average of 2.5 points (on the 10-point subscale) for the experimental group, compared to 1.3 points for the control group. The difference suggests that the AI tool\u0026rsquo;s strength in catching grammatical errors and suggesting word choice may have directly helped students produce more linguistically accurate writing. For example, common grammar mistakes (verb tense consistency, subject-verb agreement) that were flagged by the AI were largely corrected in the experimental group\u0026rsquo;s revisions, leading to higher post-test grammar scores. In contrast, the control group, relying on teacher feedback, corrected only some of those errors, possibly due to the teacher focusing on the most critical issues.\u003c/p\u003e \u003cp\u003eImprovements in \u003cem\u003eOrganization\u003c/em\u003e and \u003cem\u003eContent\u003c/em\u003e were also observed in the experimental group, though the margins over the control group were slightly smaller than for language mechanics. The experimental group improved its Organization scores by about 1.8 points on average, whereas the control improved by 1.0 point. This indicates that AI feedback \u0026mdash; while primarily focused on form \u0026mdash; also helped students restructure and clarify their writing. Many students in the experimental group reported (in informal class discussions and the post-study questionnaire) that the AI\u0026rsquo;s suggestions for clarity sometimes prompted them to split run-on sentences or use transition words, indirectly improving coherence. The Content dimension (idea development and relevance) improved similarly in both groups (around +\u0026thinsp;1.5 points each), likely because both groups received guidance on content (the experimental group through teacher\u0026rsquo;s supplemental feedback and the control group through teacher\u0026rsquo;s primary feedback). Finally, \u003cem\u003eMechanics\u003c/em\u003e (spelling and punctuation) saw a near-ceiling effect; both groups were already performing relatively well here, but the experimental group still edged out slightly (correcting almost all spelling errors with AI help). A two-way ANOVA on each sub-score confirmed significant group \u003cem\u003e\u0026times;\u003c/em\u003e time interactions for Language Use (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.01) and Organization (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.05), but not for Content (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.11), suggesting the AI\u0026rsquo;s strongest impact was on language accuracy and coherence.\u003c/p\u003e \u003cp\u003eIn summary, the results of the experiment support the hypothesis that AI-assisted writing feedback can enhance secondary students\u0026rsquo; writing skills more than traditional feedback alone. The experimental group not only achieved a higher overall post-test performance but did so with particular strength in language accuracy and clarity. The control group did improve, reflecting the value of practice and teacher input, but their progress was more limited. In the next section, we discuss these findings in the context of existing literature, consider possible explanations and limitations, and outline the pedagogical implications of incorporating AI tools like Grammarly in EFL writing instruction.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe findings of this experimental study indicate that AI-assisted writing feedback can significantly enhance the writing skills of secondary EFL students. The students who received feedback from an AI writing assistant outperformed those who received only teacher feedback, confirming our expectation that the additional, immediate feedback would lead to greater improvements in writing performance. This aligns with existing literature suggesting that AI technologies can substantially improve writing capabilities among learners, particularly in specific contexts such as EFL settings and secondary education (Nazari et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Song \u0026amp; Song, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wu, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The use of AI writing assistants, such as ChatGPT and other AI-powered feedback tools, has been documented to provide immediate and tailored feedback and to enhance students\u0026rsquo; engagement and motivation to write (Song \u0026amp; Song, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Escalante et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Research indicates that AI-generated feedback is often perceived as clearer and more instructive than conventional feedback from instructors, which can contribute to qualitative improvements in writing (Escalante et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Zhao, 2022). Moreover, recent studies emphasize the significance of individualizing feedback through AI, allowing for custom responses that cater to distinct writing needs and styles, thus enhancing the learning experience for EFL students (Wu, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Alharbi, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHowever, integrating AI tools into writing instruction does not come without challenges. Concerns about academic integrity have been raised, particularly as many students may lack formal guidance on ethical writing practices while utilizing such digital tools (PAN, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Additionally, while immediate feedback is advantageous, it may inadvertently divert students\u0026rsquo; focus from broader content development to more surface-level corrections, resulting in potential negative impacts on writing behavior (Li et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Achieving a balance between leveraging AI for immediate response while ensuring comprehensive engagement in the writing process is crucial (Li et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; ROA \u0026amp; Halim, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn this section, we interpret these results and relate them to the broader literature on writing feedback and educational technology, address potential limitations, and discuss implications for teaching practice and future research.\u003c/p\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eInterpretation of Findings\u003c/h2\u003e \u003cp\u003eOur results align with a growing consensus in recent research that automated feedback tools can be effective in improving students\u0026rsquo; writing outcomes. The dramatic improvement observed in the experimental group\u0026rsquo;s writing scores is consistent with previous studies that have documented positive impacts of AWE systems on writing development​. For example, Liu et al. (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) found that incorporating an AI-supported approach led to substantial gains in EFL learners\u0026rsquo; writing skills compared to traditional methods​. Similarly, the performance gap between our experimental and control groups echoes the findings of Wei et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), who reported that students using Grammarly for feedback demonstrated superior writing performance across multiple dimensions (task achievement, coherence, lexical resource, and grammatical accuracy) than those who did not use the AI tool​. The consistency between our study and such prior work reinforces the validity of the conclusion that AI-assisted feedback can accelerate writing improvement.\u003c/p\u003e \u003cp\u003eOne of the most notable aspects of the improvement in the experimental group was in language accuracy (grammar and vocabulary use). This is not surprising, as the AI tool provided detailed corrective feedback on these lower-level aspects of writing, allowing students to identify and fix errors that they might otherwise overlook. Our qualitative observations indicated that students became more aware of recurrent mistakes (e.g., run-on sentences, misuse of articles) due to the AI\u0026rsquo;s instant highlighting of those issues. This aligns with theories of language acquisition that emphasize \u0026ldquo;noticing\u0026rdquo; \u0026ndash; the idea that learners must notice linguistic forms and errors to integrate corrections into their interlanguage. The AI feedback likely facilitated this noticing process by consistently drawing attention to errors, functioning as a form of immediate formative feedback. Previous research also suggests that AWE tools can enhance grammatical accuracy and writing mechanics through repeated practice and feedback cycles (Grimes \u0026amp; Warschauer, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Zhang \u0026amp; Hyland, 2018). Our study adds further empirical support to that claim, showing a clear advantage in grammatical development for the AI-assisted group.\u003c/p\u003e \u003cp\u003eBeyond grammar, the AI-assisted feedback contributed to improvements in organization and clarity of ideas. Although tools like Grammarly are primarily known for grammar and style corrections, many also offer suggestions for conciseness and clarity (e.g., flagging wordy sentences or suggesting transition phrases). Students in the experimental group appeared to internalize some of these suggestions, as evidenced by more coherent paragraphs and better use of linking words in their post-test essays. This outcome is significant because it demonstrates that AI feedback can have an effect on higher-level writing skills, not just surface-level correctness. It corroborates findings from Yan (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), who observed significant improvements in overall writing quality (including coherence and content development) when EFL students used ChatGPT as a writing aid​. One possible explanation is that by taking over some of the labor of grammar correction, the AI freed up cognitive resources for students to focus on structure and content during revision. Additionally, the novelty and interactive nature of the AI tool may have kept students more engaged in the revision process, leading them to put more effort into reorganizing and refining their ideas\u0026mdash;a hypothesis in line with student engagement theory (Huang et al., 2023).\u003c/p\u003e \u003cp\u003eThe control group\u0026rsquo;s improvement, while smaller, indicates that teacher feedback and regular writing practice are indeed beneficial \u0026ndash; a finding that is well-established in educational research (Graham \u0026amp; Perin, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). However, the magnitude of difference between the groups highlights certain limitations of traditional feedback in a typical classroom. Teacher feedback in this study was delayed (coming a week after the draft) and somewhat limited due to time, which is a common scenario in secondary education. Students in the control group might have missed or forgotten some issues by the time feedback arrived, or they might not have received feedback on every error due to the teacher\u0026rsquo;s necessity to prioritize. In contrast, the experimental group received immediate, detailed feedback on virtually every error in their drafts. This likely enabled a more effective feedback loop where students could directly apply corrections and learn from mistakes in real-time. The immediacy and frequency of feedback are factors known to enhance learning (Shute, 2008), and our results provide concrete evidence of their importance in writing instruction. They also confirm that AI tools, by providing instant feedback, can augment aspects of instruction that are otherwise constrained.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eComparison with Human Feedback Quality\u003c/h2\u003e \u003cp\u003eIt is important to contextualize the role of AI feedback relative to human feedback. While our experimental group clearly benefited from AI-generated comments, we acknowledge that AI is not a pedagogical panacea and cannot entirely replace the nuanced feedback of a human teacher. Indeed, recent research comparing AI and human feedback quality has found that human instructors still excel in certain areas. For instance, a study by Steiss et al. (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) compared feedback on student essays from expert teachers versus ChatGPT, and found that human feedback was generally of higher quality, especially in providing actionable, developmentally appropriate advice​. Human reviewers were better at tailoring their comments to the student\u0026rsquo;s level and focusing on the most important content issues, whereas ChatGPT\u0026rsquo;s feedback, while helpful, sometimes lacked prioritization (it might comment on every issue, major or minor)​. In our study, we mitigated this by having the teacher in the experimental condition give complementary feedback on higher-order concerns. We found this blended approach effective: the AI handled the technical corrections and the teacher stepped in for content and organization guidance. This division of labor might be one reason the experimental group showed improvement in coherence \u0026ndash; they benefited from both AI input and teacher expertise.\u003c/p\u003e \u003cp\u003eNotably, the study by Steiss et al. (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) also concluded that AI feedback, despite being slightly inferior in quality, was \u0026ldquo;relatively high quality\u0026rdquo; and, given its ease of generation, could be very useful in contexts where teachers are overloaded or unavailable​. Our findings support this notion. The AI feedback served as a scalable way to give each student extensive comments, something that would be hard for a single teacher to accomplish regularly with a large class. The fact that our control group students improved less, despite having a live teacher, underlines how time-consuming it is for teachers to match the quantity of feedback an AI can provide in seconds. It suggests that a pragmatic strategy in education could be to use AI to supplement teacher feedback \u0026ndash; essentially giving students a first round of feedback (from the AI) and then a more targeted second round from the teacher. This could maximize the strengths of both. As Graham et al. (2016) have pointed out, feedback is most effective when it is timely and specific; AI can ensure timeliness and specificity for micro-level issues, while teachers ensure accuracy of content-focused feedback and add the empathetic, human touch that AI lacks.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eStudent Engagement and Affective Factors\u003c/h2\u003e \u003cp\u003eAn interesting qualitative observation from this study was the high level of engagement among students in the experimental group. Many of them showed enthusiasm in using the AI tool, treating it somewhat like a game to \u0026ldquo;get a perfect score\u0026rdquo; (since Grammarly gives a writing score and goals). This competitive, gamified element may have motivated extra revision and learning. Some students also reported that receiving instant feedback made writing \u0026ldquo;less frustrating\u0026rdquo; because they could immediately fix problems and see their writing improve, which increased their confidence. This aligns with findings by Fitria (2023) and Hsiao \u0026amp; Chang (2023), who noted improved motivation when EFL learners used AI tools for writing practice​. Enhanced motivation and self-efficacy can create a positive feedback loop \u0026ndash; as students feel more capable, they may take on more writing challenges, further improving their skills (Bandura, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1997\u003c/span\u003e). Our data didn\u0026rsquo;t directly measure motivation, but the post-study questionnaire indicated that 87% of the experimental group \u0026ldquo;agreed\u0026rdquo; or \u0026ldquo;strongly agreed\u0026rdquo; that they felt more confident in their English writing after using the AI feedback. This affective benefit is a valuable side-effect of AI tool usage that warrants further study.\u003c/p\u003e \u003cp\u003eOn the other hand, a few students (around 10%) in the experimental group expressed mild concerns or skepticism about the AI feedback. Some were unsure if all the AI\u0026rsquo;s suggestions were correct or stylistically appropriate for academic writing. Indeed, AI tools are not infallible; they can occasionally suggest changes that alter meaning or prefer a style that might not fit the context. We addressed this by training students to critically assess the AI feedback \u0026ndash; essentially, to use their judgment and not accept all suggestions blindly. This is a critical digital literacy skill: students must learn to work with AI as a support tool, not an authority. Fortunately, most students seemed to grasp this concept. It also helped that the teacher reviewed their final drafts, which acted as a safety net for any misguided AI suggestion that a student might have accepted. In practice, during the study we found only a handful of instances where the AI\u0026rsquo;s feedback was debatable (for example, suggesting a change that was stylistically optional). These were clarified in class discussions, and thus became additional learning points (we sometimes discussed why the AI might be wrong or what nuance it missed).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eLimitations\u003c/h2\u003e \u003cp\u003eWhile the results are encouraging, this study has several limitations that must be acknowledged. First, the sample size (N\u0026thinsp;=\u0026thinsp;60) and context (one school in Turkey) limit the generalizability of the findings. The participants were all from the same educational context with similar language backgrounds; therefore, the results might differ in schools with different curricula, or with students of different proficiency levels. Future studies with larger and more diverse samples (including different countries, age groups, and proficiency levels) would help validate whether these findings hold broadly.\u003c/p\u003e \u003cp\u003eSecond, the duration of the study was relatively short (8 weeks). We were able to capture short-term improvements in writing skills, but it is unclear if these gains would sustain in the long run. For instance, would the experimental group continue to outperform if neither group had AI assistance for a subsequent period? Longitudinal research could examine whether the skills acquired with AI feedback lead to lasting improvement or if continued use of the tool is required to maintain the advantage. It would also be interesting to see if students eventually internalize the AI\u0026rsquo;s guidance (e.g., learn to avoid certain errors permanently) or if they might become dependent on the tool. There is a concern that over-reliance on automated feedback could impede the development of self-editing skills if not balanced with proper instruction (Stevenson \u0026amp; Phakiti, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2014\u003c/span\u003e)​. Our study\u0026rsquo;s design \u0026ndash; integrating teacher feedback and encouraging critical thinking about AI suggestions \u0026ndash; aimed to mitigate this, but the risk of dependency is something to monitor in extended implementations.\u003c/p\u003e \u003cp\u003eAnother limitation relates to the scope of measurement. We focused on writing performance as measured by test essays and analytic scores. While this is a standard approach, writing is a complex skill, and improvement can also be qualitative. We did not formally analyze the content of revisions or the types of errors reduced beyond the rubric sub-scores. A deeper textual analysis could reveal which error types saw the biggest drop in the experimental group or how essay structures evolved. Additionally, our study did not directly measure learning gains in grammatical knowledge or vocabulary range \u0026ndash; we inferred these from writing outcomes. More fine-grained assessments (like grammar tests or vocabulary usage analyses) might provide additional evidence of language development attributable to the feedback.\u003c/p\u003e \u003cp\u003eLastly, though we attempted to keep instructional time and attention equal between groups, the nature of the feedback experience differed. One could argue the experimental group effectively got more feedback (even if teacher time was balanced, the AI feedback was extensive). Thus, part of the effect may simply be due to receiving \u003cem\u003emore\u003c/em\u003e feedback rather than the \u003cem\u003emode\u003c/em\u003e of feedback per se. However, from an educational standpoint, this is still a valuable finding: if AI allows us to give more feedback without extra teacher time that is a practical benefit. A more controlled study could try to isolate feedback quantity by, say, having teachers provide equally extensive feedback (which might require reducing the number of assignments or having multiple teachers). In real classrooms, however, the scenario we presented \u0026ndash; AI augmenting limited teacher feedback \u0026ndash; is realistic and thus the results have ecological validity, even if not all variables were purely controlled.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003ePedagogical Implications\u003c/h2\u003e \u003cp\u003eDespite the above limitations, the study offers several clear implications for pedagogy in language education. First and foremost, AI-assisted writing feedback appears to be a powerful tool for improving student writing, especially in EFL contexts where large class sizes and limited teacher time are common. Teachers and schools could consider integrating AWE tools like Grammarly or generative AI assistants like ChatGPT into their writing curriculum. This could take the form of students using an AI tool to self-check their drafts before submission, as a mandatory or optional step. By doing so, teachers can offload some of the routine error correction to the AI, freeing them to focus on higher-order feedback and individual student mentoring. Over time, this might not only improve student writing outcomes but also make the feedback process more efficient and less burdensome for teachers.\u003c/p\u003e \u003cp\u003eHowever, effective integration of AI requires training and guidance. We recommend that educators introduce AI writing tools to students with proper instruction on how to interpret and apply the feedback. Students should be taught that AI suggestions are not always \u0026ldquo;correct answers\u0026rdquo; but rather helpful pointers. In our study, the teacher played a crucial role in ensuring AI feedback was used productively \u0026ndash; for example, by discussing examples of good revisions or cautioning against blindly accepting changes. Teachers themselves will need professional development to familiarize themselves with AI tools and to design tasks that maximize their benefits. For instance, teachers might create a revision checklist that includes reviewing AI feedback on certain language aspects and then prompting students to reflect on what changes they made and why. This reflection can help consolidate learning from the AI feedback, turning what could be a passive correction process into an active learning opportunity.\u003c/p\u003e \u003cp\u003eAnother practical implication is the potential for personalized learning. AI tools can give each student individual feedback simultaneously, something a single teacher cannot do during class. This means students who write quickly can get feedback and continue improving in the same class session, whereas those who need more time can proceed at their own pace. Such individualized pacing can cater to mixed-ability classrooms, allowing advanced students to progress further (perhaps by writing additional pieces or refining more) without leaving behind those who require more fundamental practice. In essence, AI can serve as a personalized tutor that supplements the teacher\u0026rsquo;s instruction, aligning with the concept of differentiated learning.\u003c/p\u003e \u003cp\u003eFrom an assessment perspective, the significant improvement in writing scores for the experimental group suggests that incorporating AI feedback into the learning phase can lead to better performance in independent assessments. Educators preparing students for high-stakes English exams (which often include writing tasks) might find that regular practice with AI feedback helps improve students\u0026rsquo; accuracy and coherence, thereby potentially boosting their test results. Of course, care must be taken to ensure that during actual exams students do not use AI tools (unless explicitly allowed), to maintain academic integrity. Our findings imply that if used ethically during practice, AI tools can build skills that transfer to unaided writing situations.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eEthical and Practical Considerations\u003c/h2\u003e \u003cp\u003eThe use of AI in education raises important ethical questions that educators and policymakers should consider. One concern is academic integrity: Could students misuse AI tools to generate content that they then submit as their own work? In our study, we strictly guided students to use the AI only for feedback on their own writing, not to produce text for them. With tools like ChatGPT, the line can sometimes blur, since a student might be tempted to ask the AI to write an essay. It is vital to establish clear guidelines and honor codes. Teaching students about acceptable use of AI is now as important as teaching them about plagiarism. In fact, integrating AI in instruction might be one way to demystify it and cultivate responsible usage\u0026mdash;students learn how AI can help them learn, rather than viewing it merely as a cheating shortcut. We recommend that schools develop policies for AI usage, and that teachers openly discuss with students the difference between using AI as a learning aid versus as an unethical crutch. In our context, no incidents of misuse were observed, likely because the AI we used (Grammarly) is not designed to write text from scratch, and the environment was structured. But as more advanced AI become ubiquitous, this will be an ongoing conversation.\u003c/p\u003e \u003cp\u003eAnother consideration is equity and access. Not all students may have access to AI tools outside of school, especially if such tools require subscriptions or certain devices. In our study, we provided access in the school\u0026rsquo;s computer lab. If AI-assisted feedback is to be implemented widely, schools or districts might need to invest in licenses or ensure that free versions are sufficient for educational purposes. There is also a need to consider students with varying levels of tech-savviness. Some students might initially struggle to use the tool effectively or interpret the feedback. This calls for scaffolded support \u0026ndash; possibly peer mentoring or dedicated time to address technical issues. Encouragingly, our students adapted quickly to the interface and actually enjoyed it, but this might differ with younger learners or those less comfortable with computers.\u003c/p\u003e \u003cp\u003eLastly, in terms of teacher roles, the integration of AI feedback might shift how teachers approach writing instruction. Teachers might worry that relying on AI could diminish their role or the personal connection they have with student writing. However, we view AI as an assistant rather than a replacement. In our study, the teacher remained central to the instructional process \u0026ndash; setting objectives, evaluating content, and providing moral support and expertise that AI cannot. The positive results can hopefully alleviate some teacher apprehension: the data suggests that students still benefit enormously from teacher input, but the AI can take care of repetitive tasks. By embracing the technology, teachers might actually find more time to engage creatively with student writing (for example, discussing ideas, recommending readings to improve content, etc.). Training programs for teachers should highlight these complementary roles and perhaps share success stories of teacher-AI collaboration in classrooms.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study provided empirical evidence that AI-assisted writing feedback can serve as a powerful tool to enhance the writing skills of secondary school EFL students. Through a controlled experimental design, we demonstrated that students who received immediate, automated feedback on their writing (via an AI tool) made significantly greater improvements in overall writing performance than those who received feedback solely from a teacher in the traditional manner. The AI-assisted approach particularly boosted students\u0026rsquo; grammatical accuracy, vocabulary use, and coherence, leading to higher quality essays at the end of the intervention. These results underscore the pedagogical value of integrating AI technology into language learning environments to support and augment the feedback process.\u003c/p\u003e \u003cp\u003eOur findings contribute to the growing literature on educational applications of AI by confirming that, when used thoughtfully, AI tools can positively impact learning outcomes. In the context of EFL writing instruction, AI-assisted feedback addresses a persistent challenge \u0026ndash; providing timely and detailed feedback to each learner \u0026ndash; and does so in a scalable way. The study highlights that such tools are not in competition with teachers, but rather complement them: the most successful model was one where the AI handled micro-level corrections, while the teacher guided content development and higher-level writing skills. This synergy resulted in a richer learning experience for students, who benefited from both the efficiency of AI and the wisdom of human mentorship.\u003c/p\u003e \u003cp\u003eFrom a practical standpoint, the research offers a roadmap for educators seeking to incorporate AI in their teaching. Starting with small-scale implementations, teachers can experiment with having students use AI writing assistants for draft revisions, observe the outcomes, and adjust their strategies accordingly. The positive student response and achievement gains observed suggest that students are likely to be receptive to these innovations. Moreover, as AI tools continue to advance (with newer models like GPT-4 offering even more sophisticated feedback and interaction), the potential for integrating such technology in writing curricula will only grow. Future software may provide even more nuanced feedback on argumentation, creativity, and style, areas where current tools are just beginning to venture.\u003c/p\u003e \u003cp\u003eIt is important, however, to maintain a critical perspective and ensure ethical usage. Our study, while optimistic, also serves as a reminder that technology in education should be implemented with clear objectives and oversight. AI feedback is a means to an end \u0026ndash; improved student writing and learning \u0026ndash; and not an end in itself. As with any instructional innovation, continuous evaluation is necessary. We encourage educators and researchers to build upon these findings: for instance, investigating the long-term effects of AI feedback on writing development, comparing different AI tools or configurations (such as ChatGPT vs. specialized grammar checkers), and exploring student attitudes in depth. Qualitative studies could delve into how students make decisions when revising with AI feedback, shedding light on the cognitive processes involved.\u003c/p\u003e \u003cp\u003eIn conclusion, the experimental evidence presented here suggests that leveraging AI-assisted writing feedback holds great promise for EFL writing pedagogy. The integration of AI can lead to more effective and personalized feedback cycles, thereby accelerating skill acquisition for students who are learning to write in a second language. For teachers and institutions aiming to boost writing proficiency, particularly in contexts with large classes or limited resources, AI tools offer a practical enhancement to traditional methods. With careful integration, ongoing support, and attention to ethical considerations, AI-assisted feedback can be a catalyst for better writing outcomes. The pedagogical implication is clear: embracing AI in the writing classroom, with teachers at the helm guiding its use, can empower students to become more proficient and confident writers. As we move forward, a balanced partnership between human educators and AI technology will likely be a key feature of innovative and effective language education.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eThe Corresponding author was responsible for conceptualizing the study, designing the research methodology, conducting the data collection, and performing the statistical analyses. The second author, Ayşe Nesil Demir, contributed to the literature review, the interpretation of findings, and the refinement of the discussion section. Both authors collaborated in drafting, reviewing, and editing the final version of the manuscript and approved it for submission. The study reflects the joint efforts and academic insights of both contributors.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAlharbi, W. (2023). Ai in the foreign language classroom: a pedagogical overview of automated writing assistance tools. \u003cem\u003eEducation Research International, 2023\u003c/em\u003e, 1\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2023/4253331\u003c/span\u003e\u003cspan address=\"10.1155/2023/4253331\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBandura, A. (1997). \u003cem\u003eSelf-efficacy: The exercise of control\u003c/em\u003e. New York: W.H. Freeman.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEscalante, J., Pack, A., \u0026amp; Barrett, A. (2023). Ai-generated feedback on writing: insights into efficacy and enl student preference. International Journal of Educational Technology in Higher Education, 20(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s41239-023-00425-2\u003c/span\u003e\u003cspan address=\"10.1186/s41239-023-00425-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGraham, S., \u0026amp; Perin, D. (2007). A meta-analysis of writing instruction for adolescent students. Journal of Educational Psychology, \u003cem\u003e99\u003c/em\u003e(3), 445\u0026ndash;476. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/0022- 0663.99.3.445\u003c/span\u003e\u003cspan address=\"10.1037/0022- 0663.99.3.445\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrimes, D., \u0026amp; Warschauer, M. (2010). Utility of automated writing evaluation in ESL writing instruction. CALICO Journal, \u003cem\u003e28\u003c/em\u003e(1), 139\u0026ndash;165. (Retrieved from ERIC)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHyland, K. (2007). Genre pedagogy: Language, literacy and L2 writing instruction. Journal of Second Language Writing, \u003cem\u003e16\u003c/em\u003e(3), 148\u0026ndash;164. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jslw.2007.07.005\u003c/span\u003e\u003cspan address=\"10.1016/j.jslw.2007.07.005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, J., Link, S., \u0026amp; Hegelheimer, V. (2015). Rethinking the role of automated writing evaluation (awe) feedback in esl writing instruction. Journal of Second Language Writing, \u003cem\u003e27\u003c/em\u003e, 1\u0026ndash;18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jslw.2014.10.004\u003c/span\u003e\u003cspan address=\"10.1016/j.jslw.2014.10.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Y., Wang, Y., \u0026amp; Wen, Z. (2021). Effects of an AI-assisted writing approach on EFL learners\u0026rsquo; writing skills: A quasi-experimental study. Computer Assisted Language Learning, \u003cem\u003e34\u003c/em\u003e(8), 915\u0026ndash;938.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNazari, N., Shabbir, M. S., \u0026amp; Setiawan, R. (2021). Application of artificial intelligence powered digital writing assistant in higher education: randomized controlled trial. Heliyon, 7(5), e07014. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.heliyon.2021.e07014\u003c/span\u003e\u003cspan address=\"10.1016/j.heliyon.2021.e07014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNgo, T. T. N., Chen, H. H. J., \u0026amp; Lai, K. K. W. (2022). The effectiveness of automated writing evaluation in EFL/ESL writing: a three-level meta-analysis. Interactive Learning Environments, \u003cem\u003e32\u003c/em\u003e(2), 727\u0026ndash;744. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10494820.2022.2096642\u003c/span\u003e\u003cspan address=\"10.1080/10494820.2022.2096642\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePAN, J. (2024). Ai-driven english language learning program and academic writing integrity in the era of intelligent interface. English Language Teaching and Linguistics Studies, \u003cem\u003e6\u003c/em\u003e(4), p120. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.22158/eltls.v6n4p120\u003c/span\u003e\u003cspan address=\"10.22158/eltls.v6n4p120\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRanalli, J., Link, S., \u0026amp; Chukharev-Hudilainen, E. (2016). Automated writing evaluation for formative assessment of second language writing: investigating the accuracy and usefulness of feedback as part of argument-based validation. Educational Psychology, \u003cem\u003e37\u003c/em\u003e(1), 8\u0026ndash;25. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/01443410.2015.1136407\u003c/span\u003e\u003cspan address=\"10.1080/01443410.2015.1136407\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eROA, A. A. P. and Halim, S. (2024). The impact of ai-powered software on second language (l2) writing: a systematic literature review. Research and Innovation in Applied Linguistics- Electronic Journal, \u003cem\u003e2\u003c/em\u003e(2), 138. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.31963/rial.v2i2.4801\u003c/span\u003e\u003cspan address=\"10.31963/rial.v2i2.4801\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaricaoğlu, A., \u0026amp; Bilki, Z. (2021). Voluntary use of automated writing evaluation by university EFL students: Effects on writing quality and perceptions of feedback. ReCALL, \u003cem\u003e33\u003c/em\u003e(3), 265\u0026ndash;281. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1017/S0958344021000015\u003c/span\u003e\u003cspan address=\"10.1017/S0958344021000015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong, C. and Song, Y. (2023). Enhancing academic writing skills and motivation: assessing the efficacy of chatgpt in ai-assisted language learning for efl students. Frontiers in Psychology, 14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyg.2023.1260843\u003c/span\u003e\u003cspan address=\"10.3389/fpsyg.2023.1260843\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSteiss, J., Tate, T., Graham, S., Cruz, J., Hebert, M., Wang, J., Moon, Y., Tseng, W., Warschauer, M., \u0026amp; Olson, C. B. (2024). Comparing the quality of human and ChatGPT feedback on students\u0026rsquo; writing. Learning and Instruction, \u003cem\u003e91\u003c/em\u003e, 101894. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.learninstruc.2024.101894\u003c/span\u003e\u003cspan address=\"10.1016/j.learninstruc.2024.101894\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStevenson, M., \u0026amp; Phakiti, A. (2014). The effects of computer-generated feedback on the quality of writing: A systematic review. Assessing Writing, \u003cem\u003e19\u003c/em\u003e, 51\u0026ndash;65. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.asw.2013.11.007\u003c/span\u003e\u003cspan address=\"10.1016/j.asw.2013.11.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStorch, N. (2011). Collaborative writing in L2 classrooms: The effect of group interaction on writing quality. Language Learning, \u003cem\u003e61\u003c/em\u003e(2), 397\u0026ndash;436. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1467- 9922.2010.00591.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1467- 9922.2010.00591.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei, P., Wang, X., \u0026amp; Dong, H. (2023). The impact of automated writing evaluation on second language writing skills of Chinese EFL learners: A randomized controlled trial. Frontiers in Psychology, \u003cem\u003e14\u003c/em\u003e, 1249991. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyg.2023.1249991\u003c/span\u003e\u003cspan address=\"10.3389/fpsyg.2023.1249991\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, L. (2024). Ai-based writing tools: empowering students to achieve writing success. Advances in Educational Technology and Psychology, 8(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.23977/aetp.2024.080206\u003c/span\u003e\u003cspan address=\"10.23977/aetp.2024.080206\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan, C. (2023). Enhancing EFL learners\u0026rsquo; writing skills through ChatGPT: An experimental study. \u003cem\u003eFrontiers in Psychology, 14\u003c/em\u003e, 1260843. (Advance online publication) \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyg.2023.1260843\u003c/span\u003e\u003cspan address=\"10.3389/fpsyg.2023.1260843\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhai, N., \u0026amp; Ma, X. (2022). The effectiveness of automated writing evaluation on writing quality: A meta-analysis. Journal of Educational Computing Research, \u003cem\u003e60\u003c/em\u003e(7), 1799\u0026ndash;1826. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/07356331221094370\u003c/span\u003e\u003cspan address=\"10.1177/07356331221094370\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, W., Wang, Y., Yang, L., \u0026amp; Wang, C. (2020). Suspending Classes without Stopping Learning: China\u0026rsquo;s Education Emergency Management Policy in the COVID-19 Outbreak. Journal of Risk and Financial Management, \u003cem\u003e13\u003c/em\u003e(55). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/jrfm13030055\u003c/span\u003e\u003cspan address=\"10.3390/jrfm13030055\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-education","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diedu","sideBox":"Learn more about [Discover Education](https://www.springer.com/journal/44217)","snPcode":"44217","submissionUrl":"https://submission.nature.com/new-submission/44217/3","title":"Discover Education","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"AI-assisted feedback, EFL writing, secondary education, writing improvement","lastPublishedDoi":"10.21203/rs.3.rs-6430737/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6430737/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study examines the impact of AI-assisted writing feedback on the writing skills of secondary-level EFL students. A sample of 60 Turkish high school students was divided into an experimental group receiving feedback from an AI writing assistant and a control group receiving traditional teacher feedback. Over an 8-week period, students wrote multiple essays; pre-test and post-test writing assessments were administered. Results indicated that the experimental group showed significantly greater improvement in overall writing performance (post-test mean\u0026thinsp;=\u0026thinsp;82.3) compared to the control group (post-test mean\u0026thinsp;=\u0026thinsp;75.1, p\u0026thinsp;\u0026lt;\u0026thinsp;.01). The AI-assisted feedback group exhibited notable gains in grammar, vocabulary, and coherence, outperforming their peers with teacher-only feedback. These findings suggest that AI-driven feedback can effectively supplement writing instruction, enhancing writing proficiency in an EFL context. The study highlights the pedagogical potential of integrating AI tools into writing curricula to improve student outcomes.\u003c/p\u003e","manuscriptTitle":"AI-Assisted Writing Feedback for Enhancing Secondary Students’ Writing Skills: An Experimental Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-18 06:23:53","doi":"10.21203/rs.3.rs-6430737/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-04-29T19:08:54+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-04-29T11:01:41+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-29T11:01:34+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Education","date":"2025-04-11T19:11:06+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-education","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diedu","sideBox":"Learn more about [Discover Education](https://www.springer.com/journal/44217)","snPcode":"44217","submissionUrl":"https://submission.nature.com/new-submission/44217/3","title":"Discover Education","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"741c2e38-e604-408a-9b7a-7708677526f6","owner":[],"postedDate":"April 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-10-13T16:23:27+00:00","versionOfRecord":[],"versionCreatedAt":"2025-04-18 06:23:53","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6430737","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6430737","identity":"rs-6430737","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}