Jonna Marie Lim<\/strong> Christian Go<\/strong> This paper explores the integration of ChatGPT into the L2 writing classroom as a tool for enhancing teacher feedback on student essays. Using a reflexive case study methodology, we examine how generative AI (GenAI) augments teacher feedback in areas such as thesis clarity, idea development, and grammatical accuracy. By combining ChatGPT’s rapid feedback with teachers\u2019 contextual insights, we propose a blended feedback model that leverages both AI capabilities and teachers\u2019 expertise for more personalized feedback. Results show that the teacher effectively utilizes ChatGPT by integrating AI-generated insights with her own to refine feedback, correct inaccuracies in ChatGPT’s feedback, and engage students in meaningful dialogues based on these AI-generated insights. These strategies highlight the blended feedback model\u2019s potential to provide comprehensive and personalized feedback, which could deepen student engagement and enhance the quality of their writing. While ChatGPT proves beneficial for formative feedback, its effectiveness is greatly enhanced when combined with the expert knowledge of teachers, ensuring that the feedback remains relevant and tailored to individual student needs. Thus, we recommend further exploration of blended feedback models and additional strategies for utilizing GenAI to enhance the quality of teacher feedback, particularly in formative assessments in L2 writing classes.<\/p>\n Keywords<\/em><\/strong>: GenAI, ChatGPT, teacher feedback, writing instruction<\/p>\n The integration of generative artificial intelligence (GenAI) in writing classrooms has sparked interest among scholars in education and applied linguistics. A significant strand of research on the subject revolves around concerns with GenAI\u2019s capacity to enhance and undermine the integrity of academic writing, the value of revision and editing in the writing process, and the potential for it to impede students\u2019 critical thinking skills (Sullivan et al., 2023; Walter, 2024). These concerns stem from the broader apprehension that relying heavily on AI could lead to an erosion of foundational writing skills and hinder students\u2019 ability to engage critically with their own work (Draxler et al., 2023). At the same time, attention is also turning to how GenAI is reshaping the pedagogical landscape\u2014not just for students, but for educators as well. A growing body of research is also investigating the ways in which GenAI can reshape the role and responsibilities of educators themselves (Crompton et al., 2024; Hockly, 2023). Scholars highlight that the integration of AI in teaching might be used to automate repetitive tasks like grammar checking or aid teachers in generating writing prompts, ultimately saving time and allowing educators to focus more on higher-level instructional tasks (Escalante et al., 2023).<\/p>\n Of particular interest to this paper is the role of GenAI as a tool for providing personalized feedback on L2 (second language) student writing. While AI-generated feedback systems like OpenAI\u2019s ChatGPT, Google\u2019s Bard, and Microsoft\u2019s Bing offer promising potential for reshaping feedback practices, they cannot replicate the nuanced understanding teachers bring to their students\u2019 individual needs and contexts. Still, their capacity for offering scalable critiques provides an external perspective that could be especially beneficial for L2 learners, who often require frequent, detailed feedback to address common writing errors and internalize writing conventions (Mollick & Mollick, 2023). This aligns with Escalante et al. (2023), who note that AI-generated feedback is useful not only for its time efficiency but also for offering focused, individualized comments that can be especially valuable in large classes where personalized attention is often limited. GenAI tools can identify recurring issues in student writing, such as lexical errors, structural problems, logical flow, and even style or tone\u2014areas where L2 learners frequently struggle.<\/p>\n While much of the current research has focused on the advantages of AI in providing generalized, scalable feedback, there is a gap in the literature regarding how teachers in L2 contexts can effectively incorporate AI-generated insights into their own feedback practices (Barrot, 2023; Mi et al., 2025). Existing studies have largely treated AI feedback as a standalone tool, but less attention has been paid to how teachers might blend AI-generated feedback with their own expertise to create a more personalized and nuanced feedback experience for students. This paper seeks to fill that gap by examining how ChatGPT, as a generative AI tool, can be integrated into the L2 writing classroom to enhance and supplement teacher feedback. Specifically, we propose a blended feedback model that combines the strengths of AI-generated insights with the contextual understanding and pedagogical expertise of teachers, offering a more comprehensive approach to feedback. This model aims to refine traditional feedback practices by using AI to highlight areas for improvement while still ensuring that feedback remains relevant, personalized, and aligned with the specific needs of each student. By focusing on how teachers can effectively adapt AI-generated insights to complement their own feedback, this paper contributes to the growing body of research on the pedagogical applications of GenAI in L2 education, offering practical strategies for educators seeking to enhance their feedback processes.<\/p>\n The following section outlines current scholarship on teacher feedback and automated forms of evaluation. Afterward, we present the methodology and findings based on one of the authors\u2019 self-reflective practice vis-\u00e0-vis GenAI-use in providing feedback on students\u2019 writing.<\/p>\n Teacher feedback has been viewed as a crucial component in second-language writing research. Scholars note the importance of teacher feedback in strengthening student\u2019s language skills development (e.g., grammar) and their willingness to revise their work (Hyland & Hyland, 2006). Previous research on teacher feedback involved providing descriptions of the ways in which feedback is conveyed (Chaudron, 1986) and insights on designing feedback processes (Olsen & Hunnes, 2024). These studies note the importance of effective feedback strategies, such as error correction, multiple feedback modalities, and timely and personalized feedback. This immediacy is particularly valuable, as feedback is most effective when delivered promptly, ideally soon after the completion of a writing task (Rahimi et al., 2024). However, teacher feedback has also been subject to critique as a potentially problematic practice. Several studies investigating the nature of written responses of instructors to L2 student writing reveal teachers\u2019 tendencies to prioritize technical concerns or surface features, such as grammar, over rhetorical ones (Furneaux et al., 2007; Lee, 2019; Stern & Solomon, 2006). This type of feedback that focuses disproportionately on error correction may lead L2 students to perceive writing as a purely technical skill rather than a communicative or expressive one (Ferris, 2006; Nguyen, 2024). Additionally, such feedback that emphasizes deficits or adheres to monolingual norms can discourage linguistic diversity thereby limiting students\u2019 ability to engage with their multilingual identities in their writing (Canagarajah, 2013). As such, there is a need for feedback practices that are inclusive, context-sensitive, and aligned with fostering students\u2019 broader communicative competence.<\/p>\n Additionally, studies on feedback for L2 students show that both the amount and type of feedback affect the quality of their final drafts (Dawson et al., 2019; Hyland & Hyland, 2019; Matsumura et al., 2002). Ellis (2009) distinguishes between direct feedback, which provides explicit corrections, and indirect feedback, which encourages students to self-edit. While direct feedback benefits lower-proficiency learners, indirect feedback fosters deeper cognitive engagement by prompting students to reflect on their linguistic choices (Bitchener & Ferris, 2012; Ferris, 2006; Gu\u00e9nette, 2007). Put in another way, the efficacy of feedback strategies varies depending on the specific writing situation and learner needs. Furthermore, feedback timing is crucial\u2014studies indicate that immediate feedback enhances learning outcomes, as students can integrate corrections more effectively (Rahimi et al., 2024). Bock et al. (2024) add that providing effective feedback requires teachers to draw on a variety of knowledge: understanding how writing works (content knowledge), what good writing looks like in a specific context (pedagogical knowledge), identifying areas for improvement (diagnostic knowledge), and explaining those areas to the student in a way that promotes improvement (instructional knowledge). Moreover, teacher-student interaction influences feedback effectiveness. Storch and Wigglesworth (2010) suggest that dialogic feedback, such as teacher-student conferences, enhances students\u2019 ability to internalize suggestions and apply them in subsequent writing tasks. Thus, the effectiveness of feedback is closely tied to the teacher\u2019s ability to synthesize content, pedagogical, diagnostic, and instructional knowledge while also considering individual learner characteristics and the broader instructional context (Lee, 2019; Yu & Hu, 2017).<\/p>\n Complementary to scholarly inquiries on the means of providing feedback to students are explorations of the most efficient and pedagogically sound ways to deliver these. The use of Automated Writing Evaluation (AWE) tools for formative feedback on student writing has engendered an ongoing conversation about its utility. Proponents highlight the potential for AWEs to streamline the grading process, freeing up teacher time to address higher-level writing skills like organization and argument development (Wilson & Roscoe, 2020). Moreover, AWEs offer consistency in feedback delivery, reducing potential bias in teacher evaluations and ensuring all students receive structured guidance (Dikli, 2006; Stevenson & Phakiti, 2014). Some studies also suggest that AWE-generated feedback can enhance student engagement by providing immediate responses that facilitate iterative revisions (Ranalli, 2018). That said, while automated writing evaluation (AWE) systems provide constant, personalized feedback, this feedback often focuses on surface-level comments rather than propositional content, thus restricting students\u2019 creativity, hindering the development of ideas, and encouraging the adoption of shortcuts (Link et al., 2022; Y. Wang et al., 2013). AWE tools also struggle to effectively assess rhetorical elements such as audience awareness, argument structure, and coherence, which are central to academic writing (Grimes & Warschauer, 2010; Roscoe et al., 2017). Link et al. (2022) highlight several limitations in current discussions on AWE: (1) a reliance on theoretical rather than substantial empirical evidence, (2) inconclusive findings on AWE program effectiveness, and (3) a focus on short-term impacts, neglecting potential long-term effects on student writing skills. Thus, while AWE systems hold promise as supplementary tools, scholars emphasize the need for hybrid approaches that integrate automated feedback with human guidance to foster deeper writing development (Liu & Yu, 2022; Stevenson & Phakiti, 2014).<\/p>\n The introduction of chatbots like ChatGPT has significantly influenced Automated Writing Evaluation (AWE) and feedback mechanisms in writing instruction (Barrot, 2023; Gayed et al., 2022). As research on ChatGPT\u2019s role in writing classrooms continues to grow, studies have examined its affordances in areas such as automated essay evaluation, AI-generated feedback, writing efficacy, and its practical applications in enhancing writing skills (Kohnke et al., 2023; Mizumoto & Eguchi, 2023; Teng, 2024a). While these studies highlight ChatGPT\u2019s utility in improving student writing, ongoing research continues to explore its specific strengths and limitations as a feedback tool. In their study examining the impact of ChatGPT as a formative feedback tool on the academic writing skills of undergraduate ESL students, Mahapatra (2024) employed a mixed-methods intervention approach. The study involved 35 participants in an experimental group and 37 in a control group, with qualitative data collected through focus group discussions. The findings indicated that ChatGPT significantly enhanced students’ writing organization, grammatical accuracy, and overall engagement in the writing process. Furthermore, students reported increased collaboration and focus while using ChatGPT, which helped them adhere to topic sentences and improve the coherence of their writing (see also Su et al., 2023; Teng, 2024b). L. Wang et al. (2024) assessed the feedback accuracy of ChatGPT in evaluating undergraduate students\u2019 argumentation skills and found that it demonstrated a precision rate of 91.8% and a recall rate of 63.2%. The accuracy of ChatGPT’s feedback was significantly influenced by the length of the arguments, with longer arguments yielding lower precision and recall rates, and by the presence of discourse markers, which served as indicators of argument structure. Further, they noted that although ChatGPT could identify the structural components of arguments, its qualitative assessments\u2014such as evaluating the adequacy of evidence\u2014were less robust. Meanwhile, M. Wang & Guo (2023) found that ChatGPT provided more directive feedback (i.e., specifying what changes need to be implemented), offered more positive reinforcements, and gave more summaries when assessing the organization of writing. These affordances may offer a distinct advantage in that students can re-examine their work from a holistic perspective (global level) and grasp organizational shortcomings within their writing. Taken together, these studies suggest that ChatGPT can enhance students\u2019 writing by providing structured, rubric-based feedback and facilitating engagement, though its effectiveness varies depending on writing task complexity and the type of feedback required.<\/p>\n However, research also highlights ChatGPT\u2019s limitations compared to human evaluators. Steiss et al. (2024), for instance, compared ChatGPT\u2019s feedback to that of human evaluators and found that human evaluators generally provided higher-quality feedback in terms of accuracy, prioritization of essential features, clarity of directions, and supportive tone. That said, ChatGPT performed better than human evaluators in criteria-based feedback, as it consistently referenced evaluation rubrics. These studies underscore the potential for ChatGPT to serve as a preliminary feedback tool, helping students address fundamental issues before seeking human evaluation, while also highlighting the need for AI literacy and strategic implementation in writing instruction. In this sense ChatGPT\u2019s role as a useful preliminary feedback tool, is contingent on human factors (e.g., oversight, engagement).<\/p>\n Research on ChatGPT as a feedback tool also highlights perceptions about its accessibility and effectiveness in enhancing student writing, while also underscoring concerns about its limitations in providing personalized and context-sensitive feedback. In Mai et al.\u2019s (2024) systematic review of ChatGPT\u2019s use in education. They analyzed studies from various countries (including the United States, Canada, the United Kingdom, Germany, China, Saudi Arabia, and Vietnam) and noted that educators in these contexts generally perceive ChatGPT as a useful tool in accomplishing teaching tasks which includes providing feedback. For students, while ChatGPT enhances feedback accessibility\u2014particularly in writing tasks, as evidenced by studies in Hong Kong and Turkey\u2014its responses tend to be generic, often necessitating human oversight to foster deeper critical engagement (Mai et al., 2024).<\/p>\n The effectiveness of ChatGPT as a feedback tool has been further explored in research comparing its performance to that of human evaluators. Building on perceptions of AI’s role as a feedback tool, Teng (2024b) examines how ChatGPT influences EFL learners\u2019 writing development in a classroom setting by focusing on its impact on motivation, self-efficacy, and collaborative writing. Specifically, Teng (2024b) investigated the impact of ChatGPT on the writing skills of 45 EFL learners in Macau, utilizing a mixed-methods approach that included quantitative data from questionnaires and qualitative insights from interviews after a writing course. Findings revealed significant positive effects of AI assistance on students\u2019 writing motivation, self-efficacy, engagement, and collaborative writing tendencies. While participants acknowledged the benefits of ChatGPT in providing timely and specific feedback, some expressed concerns regarding the lack of personal touch and the machine\u2019s limitations in understanding nuanced writing issues (e.g., students noted that ChatGPT sometimes struggled with understanding the more nuanced aspects of writing, such as tone, context, and specific writing challenges unique to individual learners.).<\/p>\n Expanding on these findings, Han and Li (2024) examined ChatGPT-assisted teacher feedback in a Chinese university EFL course, proposing an \u201cAI + Teacher\u201d model where instructors refined ChatGPT\u2019s indirect corrective and rhetorical feedback before sharing it with students. Their study of 102 learners found that while AI-generated feedback improved efficiency, teachers still played a crucial role in adapting the feedback for clarity and relevance.<\/p>\n Building upon the previously discussed scholarly explorations of GenAI\u2019s potential within writing instruction, this paper advances these conversations by examining the ways in which ChatGPT, as a tool for AWE, may be utilized in conjunction with the teacher\u2019s feedback. Furthermore, according to L. Yang and Li (2024), most studies on ChatGPT and language learning have primarily examined the contexts of EFL college learners. As such, this study aims to expand the investigation into how AI-generated feedback might be tailored to meet the needs of L2 writers and support educators in providing more targeted guidance and instruction.<\/p>\n With these insights, this paper explores how ChatGPT’s capabilities can be strategically utilized to optimize the effectiveness and efficiency of feedback delivery in L2 writing classes. To achieve this, it focuses on ChatGPT\u2019s utility as a tool for refining the quality of feedback on student writing. In this inquiry, we seek to answer the following research question: In what ways can teachers integrate and adapt ChatGPT\u2019s feedback to enhance the quality of their own feedback on L2 student writing? We suggest that the incorporation of ChatGPT\u2019s analytical and editing capabilities with the more nuanced, contextual expert knowledge of teachers can lead to a blended model of feedback. This approach leverages AI-generated insights as a springboard for deeper engagement with students\u2019 work, fostering more in-depth, personalized teacher feedback and potentially leading to improved writing outcomes for L2 students.<\/p>\n The present study employs a reflexive case study approach to analyze qualitative data derived from the lead author’s annotations and reflections on ChatGPT\u2019s feedback on the 22 problem-solution essay drafts written by her first-year college students in a communication course. The analysis focuses on evaluating the effectiveness of ChatGPT’s feedback, with the lead author\u2013hereafter referred to as the teacher\u2013making critical decisions on whether to modify, retain, or discard the AI-generated feedback before it is finalized and forwarded to the students. By integrating self-reflection, a fundamental aspect of effective teaching practice (Yin, 2014), this study aims to deepen understanding of how emerging technologies like ChatGPT can complement and enhance teacher feedback on student writing by analyzing the teacher\u2019s interactions with AI-generated feedback on student essays.<\/p>\n This study examines the feedback annotations and reflections of a female university instructor who brings her extensive experience to teaching writing, with 18 years at the secondary level and 6 years at the tertiary level. She has a specialized certification in assessment and has an MA in English Language Teaching and a PhD in Applied Linguistics. Currently, she works at a university in Manila where she teaches Purposive Communication, a General Education course required for first-year college students in the Philippines, which also serves as the data collection site for this study. These experiences greatly shaped her decisions on ChatGPT\u2019s feedback utility and effectiveness in creating quality, more personalized feedback on her students\u2019 essays. To mitigate potential bias given her dual role as a participant and a researcher, peer debriefing was done with the co-researcher who provided an external review to ensure that the data inferences were accurate and within the bounds of the available data.<\/p>\n The primary data for this study were the feedback annotations made by the teacher on 22 problem-solution essays written by first-year college students in their Purposive Communication class. Purposive Communication is a mandatory General Education course for all Filipino college students where they are required to develop their abilities to communicate ethically, effectively, and professionally for diverse audiences, and purposes in various modes. One of the major requirements of the course is the problem-solution essay where students engage critically with a localized issue linked to a Sustainable Development Goal (SDG).<\/p>\n Data collection procedures<\/em> Shown in Figure 1 is a sample of ChatGPT\u2019s generated feedback, categorized according to the specific feedback areas mentioned earlier.<\/p>\n The teacher then reviewed ChatGPT\u2019s generated feedback, annotating at the margins whether she would retain, modify, or remove each feedback. She also added her initial insights on the quality of the feedback generated by ChatGPT, as illustrated in Figure 2. In the analysis, the texts enclosed in brackets [] and italicized represent the feedback of the teacher. Each feedback is tagged with a code like \u201cA57G15\u201d, where \u201cA57\u201d specifies the class section and \u201cG15\u201d identifies the group number within that section.<\/p>\n The final feedback given to the students was compiled in a separate document. This document included feedback added and modified by the teacher while removing any incorrect or unnecessary comments from ChatGPT.<\/p>\n The teacher’s annotations and reflections underwent reflexive thematic analysis (Braun & Clarke, 2013) to uncover the underlying strategies\u2013 these teacher strategies emerged as the four key themes of the data. The analysis began with data familiarization, where all marginal comments by the teacher on the ChatGPT-generated feedback were consolidated into a single document. These comments reflected the teacher\u2019s decisions to adopt, remove, or modify the AI feedback and included her initial thoughts and insights during the review process.<\/p>\n Next, initial codes were generated where the coding focused on the strategies the teacher employed in response to the AI-generated feedback. These initial codes were then analyzed to identify patterns that formed the themes, in this case, the teacher’s strategies for using or building upon ChatGPT’s feedback. These themes were ultimately finalized and defined as teacher strategies. The reflexive thematic analysis proved useful in identifying how the teacher effectively integrated her expertise with AI-generated comments to achieve the desired learning outcomes of the essays. It highlighted her extensive experience in providing feedback and her nuanced understanding of her students\u2019 actual needs.<\/p>\n Since the data came from the lead author’s personal annotations and reflections, traditional consent from external participants was not necessary. However, the analysis ensured integrity in the handling and presentation of the reflective data by acknowledging the intrinsic subjectivity of a reflexive thematic analysis.\u00a0 The teacher, a university instructor with extensive experience in secondary and tertiary education, utilizes her background in assessment and advanced degrees in English Language Teaching and Applied Linguistics to enrich her analysis.<\/p>\n In the sample, four key strategies emerged from the teacher\u2019s annotations and reflections on ChatGPT\u2019s feedback. These strategies relate to the dynamic between ChatGPT-generated comments and teacher expertise, specifically how the teacher negotiates the generated output: (1) retaining ChatGPT\u2019s feedback and enhancing it with additional comments; (2) omitting ChatGPT\u2019s inaccurate feedback; (3) creating more nuanced feedback; and (4) engaging the students in a dialogue using ChatGPT\u2019s feedback. This section discusses each of these key strategies in detail, highlighting both the strengths and limitations of ChatGPT\u2019s feedback and underscoring the critical role of the teacher in refining and contextualizing AI-generated feedback to better meet students’ learning needs.<\/p>\n ChatGPT was valuable in identifying specific areas for student improvement, which served as a starting point for teachers to expand and offer more targeted guidance. By analyzing student work, ChatGPT could pinpoint weaknesses in the quality of thesis statements, development of ideas, and use of transitions. This pinpointed feedback allowed teachers to focus their efforts on these crucial aspects of writing, ensuring their interventions were directly addressing student needs.<\/p>\n In the first type of feedback, ChatGPT comments on the quality of the student’s thesis statement. It pinpoints the strengths by acknowledging the focus of the thesis\u2019 central dilemma: balancing marine conservation with livelihoods (Sample 1 below). Furthermore, the feedback constructively criticizes the weakness by suggesting the inclusion of a proposed solution or recommendation. Additionally, the provided sample demonstrates how incorporating a potential solution, such as stricter enforcement alongside community engagement initiatives, can enhance the thesis. The provided feedback thus provides instructions on how to make the thesis more argumentative and outlines the paper\u2019s intended direction. While the feedback doesn\u2019t explicitly state that including a solution is mandatory for all thesis statements, its suggestion and sample illustrate how following the feedback can significantly improve the focus and argumentative nature of the thesis.<\/p>\n Sample 1<\/em><\/strong><\/p>\n Feedback on the quality of the thesis statement: <\/strong>The thesis statement provided in the student’s writing does address the writing prompt by highlighting the dilemma between preserving marine biodiversity and supporting the livelihoods of people. However, it could be improved by explicitly mentioning the proposed solution or recommendations for practical action. For example, an improved thesis statement could be:<\/p>\n “While addressing the challenge of balancing marine conservation with supporting local livelihoods poses a dilemma, implementing stricter enforcement of marine biodiversity laws alongside community engagement initiatives can effectively preserve marine biodiversity while also sustaining the livelihoods of coastal communities.” (A57DG8)<\/p><\/blockquote>\n As can be seen in Sample 2 below, both the ChatGPT-generated and teacher feedback focus on improving the development of ideas in the essay and complement each other. The ChatGPT feedback highlights the need for further elaboration on a specific solution related to the IFN (Ideal Flow Network)\u2013 its workings, feasibility, and potential challenges. This technical focus strengthens the argument\u2019s foundation. This comment is further enhanced through the teacher\u2019s clarification where she emphasizes the integration and flow of ideas. Rather than listing these benefits, the teacher advises integrating them seamlessly into the paragraphs, thereby enhancing the earlier recommendations from ChatGPT more clearly. Thus, while each feedback emphasizes the need for a deeper and more nuanced discussion of the IFN, the teacher\u2019s feedback specifically focuses on enhancing the coherence and flow of the essay by improving how the benefits are presented.<\/p>\n Sample 2.<\/em><\/strong><\/p>\n Feedback on the development of ideas throughout the paragraph and essay:<\/strong> Your essay effectively develops the problem of traffic congestion in Manila, providing ample evidence and statistics to support the argument. However, the discussion on the proposed solution using the Ideal Flow Network could be further elaborated. Provide more details on how this model works, its feasibility, and potential challenges in its implementation. Additionally, consider addressing potential counterarguments to strengthen your argument. [Can you re-write the part where you presented the benefits of the IFN. Instead of enumerating them, integrate them in your paragraphs. Better yet, use these input to address the recommendations stated in the previous paragraph.] <\/em>(A57DG7)<\/p><\/blockquote>\n Meanwhile in Sample 3, the feedback concerns the use of transitional phrases, but from different angles. ChatGPT first provides positive reinforcement, commending the appropriate use of transitions for guiding the reader. At the same time, it recommends creating variety by replacing repetitive phrases like \u201cMoreover\u201d with alternatives like \u201cIn addition\u201d or \u201cConsequently.\u201d The teacher\u2019s feedback, on the other hand, enhances the feedback by providing personalized commentary such as integrating bulleted points from an outline into a cohesive essay, and raises a concern about exceeding the word limit. In this sense, the teacher\u2019s feedback points to specific structural issues in the student\u2019s paper and offers strategies for improving the overall organization and coherence of the essay as well as contextualized feedback that fits with the specifics of the assignment.<\/p>\n Sample 3.<\/em><\/strong> ChatGPT Feedback with the Teacher\u2019s Personalized Commentary on Transitions<\/strong><\/p>\n Feedback on the transitional phrases: <\/strong>Your essay uses transitional phrases appropriately to guide the reader through the various sections and ideas. However, you could improve the flow by incorporating more diverse transitional phrases. For example, instead of repeatedly using phrases like “Moreover” or “Furthermore,” you could vary your transitions with alternatives like “In addition” or “Consequently” to add variety and sophistication to your writing. [When you remove the headings and integrate the bullet points in paragraphs, make sure to use effective transitions to link your ideas. Btw, did you not go beyond the 2500 word count? Your draft looks like an outline. You revise it in a way that all your key ideas are presented in one continuous essay.]<\/em> (A58G7)<\/p><\/blockquote>\n These examples highlight how ChatGPT\u2019s comments provided a foundation for improvement by focusing on essential elements that contribute to effective essays.<\/p>\n Although ChatGPT-generated feedback provides a useful starting point for personalized feedback, it also has limitations that necessitate teacher intervention and correction. These constraints stem from the nature of AI language models, which can struggle to replicate the nuanced judgment of an instructor. The examples below are instances where the generated feedback may be inaccurate or misleading. This prompted the researchers to flag them as inaccurate and omit them. This problematic feedback includes the following issues: deviations from the essay\u2019s central topic, provision of irrelevant advice, inclusion of factually incorrect information, and generation of unhelpful suggestions that do little to improve the student’s writing.<\/p>\n Sample 4<\/em><\/strong>.<\/em> Sample of ChatGPT\u2019s Incorrect Feedback on Comma Splice<\/strong><\/p>\n
\nDe La Salle University-Manila
\n<jonna.lim![\"atmark\"](\"http:\/\/www.tesl-ej.org\/atmark.png\")
dlsu.edu.ph><\/p>\n
\nUniversity of the Philippines-Diliman
\n<cggoup.edu.ph><\/p>\nAbstract<\/h3>\n
Teacher feedback, automation and their interface<\/h3>\n
ChatGPT feedback on writing<\/h3>\n
Methodology<\/h3>\n
Research design<\/h4>\n
The participant<\/h4>\n
\u00a0The context<\/h4>\n
\nIn integrating ChatGPT into the feedback process, the following steps were taken. The teacher informed the students that ChatGPT will be used in creating more personalized feedback on their essays and she explained the process for generating the final feedback for revisions. She then read each essay and uploaded the draft to ChatGPT 3.5, using a detailed prompt instructing it to give feedback on the following feedback areas: quality of the thesis statement, clarity of problem, solution and call to action, development of ideas throughout the paragraph and essay, academic quality of the language, transitional phrases, use of sources and evidence, and grammatical accuracy. The prompt was adapted from Escalante et al. (2023) to align with the grading criteria for the course’s problem-solution essay.<\/p>\n![\"Sample](\"https:\/\/tesl-ej.org\/ej115\/intpix\/image1.png\")
\nFigure 1. Sample ChatGPT-generated Feedback on an Essay Draft<\/strong><\/p>\n![\"Sample](\"https:\/\/tesl-ej.org\/ej115\/intpix\/image2.png\")
\nFigure 2. Sample Teacher Annotation on ChatGPT-generated Feedback<\/strong><\/p>\nData analysis<\/h4>\n
Findings<\/h3>\n
Retaining and enhancing ChatGPT-generated feedback strategy<\/h4>\n
Omitting ChatGPT\u2019s inaccurate feedback strategy<\/h4>\n