1. Introduction
Writing in English, particularly at an academic level, is a demanding and multifaceted process that requires the integration of various cognitive and linguistic skills (L. K. Allen & McNamara, 2017). English as a Foreign Language (EFL) learners often encounter significant challenges in this domain due to limited linguistic resources and difficulties in addressing multiple aspects of writing, such as organizing ideas, maintaining coherence, and ensuring grammatical accuracy (Chen & Cui, 2022; Wale & Bogale, 2021). To support EFL learners in navigating this complex process, personalized and formative feedback has become a cornerstone of effective pedagogy.
In the context of EFL writing, feedback plays a pivotal role (Crosthwaite et al., 2022). It involves providing learners with explicit or implicit written guidance on errors at local levels (e.g., grammar, vocabulary) or global levels (e.g., content, organization), with the aim of fostering linguistic accuracy and improving overall writing quality (Crosthwaite et al., 2022). Teacher-generated feedback, a primary form of feedback, has consistently been shown to positively influence students’ revision practices and subsequent writing performance (Leki, 1991). Teacher-generated feedback offers rich and context-sensitive insights that are critical for addressing both surface-level issues and deeper, global aspects of writing (Lin & Crosthwaite, 2024). However, challenges such as large class sizes, time constraints, and workload often limit teachers’ ability to deliver timely and personalized feedback to all students (Teng & Ma, 2024; Yu et al., 2021). To address these limitations, researchers and practitioners have turned to AI-generated writing technologies as a potential solution.
Over the years, AI-powered feedback tools have evolved significantly, transitioning from earlier rule-based systems and basic Natural Language Processing (NLP) algorithms to more sophisticated generative AI (GenAI) technologies (Yan et al., 2024). While traditional AI tools primarily offered grammar corrections or predefined responses (Hartshorn et al., 2010), modern GenAI, powered by large language models (LLMs) such as ChatGPT, Bing AI, and Gemini, provides personalized, context-aware, and nuanced feedback, thereby transforming the landscape of language education (Kasneci et al., 2023; Li et al., 2023; Mizumoto & Eguchi, 2023). AI-powered tools can provide immediate, automated feedback on learners’ writing, offering a scalable alternative to the labor-intensive nature of teacher-generated feedback (Yan et al., 2024). These tools are particularly effective in identifying and addressing surface-level issues, such as grammar, punctuation, and spelling, while enabling teachers to focus on higher-order writing skills like idea development and coherence (Link et al., 2022). However, the use of AI-generated feedback is not without its limitations. Critics argue that AI-generated feedback often lacks the contextual understanding needed to address deeper, global aspects of writing and may provide formulaic or redundant suggestions (Lin & Crosthwaite, 2024). Moreover, there is a risk of difficulty in understanding and comprehending responses as well as providing inaccurate feedback, which may confuse learners rather than support their writing development (Teng, 2024).
Given the respective strengths and limitations of both teacher- and AI-generated feedback, it is expected that integrating these two forms has the potential to enhance the writing process. By leveraging the complementary nature of AI’s scalability and automation with the contextual depth of teacher-generated feedback, educators can create a more balanced and effective feedback mechanism. However, existing research has paid limited attention to how these two feedback forms can be systematically integrated to optimize their impact on student writing. Specifically, little is known about the sequencing of AI- and teacher-generated feedback, whether providing AI-generated feedback before teacher-generated feedback (AIGF-TGF) or teacher-generated feedback before AI-generated feedback (TGF-AIGF) results in more comprehensive revisions.
To address this gap, this preliminary study aims to examine how the integration and sequencing of AI- and teacher-generated feedback influence EFL learners’ writing revision processes. It aims to investigate the extent to which these forms of feedback, as well as the two orders in which the two types of feedback have been executed, affect the processes of student revisions.
2. Literature Review
2.1. Feedback and Revisions
Feedback is broadly defined as information offered by an agent about specific aspects of an individual’s performance or understanding (Hattie & Timperley, 2007). Similarly, feedback provision refers to actions undertaken by external agents to supply information concerning certain facets of task performance (Kluger et al., 1996). These definitions emphasize two crucial elements, such as the “agent” delivering the feedback and the “information about performance”. The “agent” refers to the individual or system providing feedback, while the “information about performance” pertains to the gap between the learner’s current performance and the intended learning goals or target outcomes. The primary aim of feedback is to bridge this gap, underscoring its integral role in facilitating teaching and learning processes (Rollinson, 2005).
Revision, as defined by Reid (1993, p. 233), means “seeing again” and refers to the process of evaluating and reworking text at any stage in the writing process (Piolat, 1997). It is an iterative and problem-solving process through which writers reshape their ideas, refine meaning, and improve the quality of their texts (Faigley & Witte, 1981; Sommers, 1996). Faigley and Witte (1981) further classify revisions into two categories, including surface-level changes, which include adjustments to spelling, punctuation, and grammar that do not alter the meaning of the text, and text-based changes, which involve modifications to the structure, content, or overall meaning. This distinction underpins the analysis of revisions in the current study, providing a framework to evaluate the origin and focus of the changes made during the feedback sequences.
To ensure terminological clarity, this study distinguishes between the areas of revisions (local vs. global) and the focus of feedback (surface-level vs. higher-order). Local revisions refer to modifications at the sentence or word level, such as grammar, vocabulary, and punctuation, while global revisions address broader writing aspects, such as content development, organization, and coherence. In contrast, surface-level feedback focuses on form-related issues (e.g., grammatical accuracy or lexical choices), whereas higher-order feedback targets content-related concerns, rhetorical strategies, and overall cohesion. By distinguishing these terms, the study aims to provide a nuanced understanding of the feedback–revision process in academic writing.
2.2. Teacher-Generated Feedback
In EFL language teaching, feedback provided by teachers has long been a cornerstone of writing instruction (Rollinson, 2005). Previous research has highlighted its pivotal role in enhancing EFL writing skills and supporting the broader process of second language acquisition (K. Hyland & F. Hyland, 2006). The nature, focus, and effectiveness of teacher-generated feedback continue to be subjects of considerable academic discussion and inquiry in both English as a Second Language (ESL) contexts (Montgomery & Baker, 2007) and English as a Foreign Language (EFL) settings (Mao & Crosthwaite, 2019).
One key consideration in teacher-generated feedback is achieving an appropriate balance between direct and indirect written corrective feedback (WCF), tailored to the pedagogical objectives of writing instruction (K. Hyland & F. Hyland, 2006). Direct WCF, which explicitly identifies and corrects errors, is particularly effective in drawing learners’ attention to discrepancies between their interlanguage and target language forms, thus helping them notice linguistic gaps (Bonilla López, 2021). In contrast, indirect WCF requires learners to interpret error codes or markings to identify and correct their mistakes. This approach demands greater cognitive engagement from learners, encouraging them to solve problems independently, develop self-regulation, and acquire long-term editing and revision skills that contribute to meaningful second language acquisition (Bozorgian & Yazdani, 2021; Ferris & Roberts, 2001).
Another critical aspect of teacher-generated feedback involves its focus on local and global issues. Excessive emphasis on local issues, such as grammar and spelling, may reduce the teacher’s role to that of a proofreader rather than an instructor and may lead students to equate successful writing solely with linguistic accuracy (Lee, 2008). Conversely, global feedback focuses on the content, coherence, and organization of the writing from the reader’s perspective, highlighting areas where the writing fails to convey the intended message and offering strategies for improvement (Goldstein, 2004). A balanced approach to addressing both local and global aspects is considered essential for fostering comprehensive improvements in student writing (Lee, 2009).
Despite its recognized benefits, teacher-generated feedback practices often vary widely, with challenges persisting in their implementation. Writing instructors, particularly those working in large classes, frequently struggle to provide timely, individualized, and comprehensive feedback. Furthermore, teachers often exhibit preferences for specific feedback types—such as direct error correction over indirect feedback—and tend to focus on local issues in an unfocused manner (Abdullah & Aziz, 2020; Lee, 2009). These practices may stem from contextual constraints, insufficient training in feedback delivery, and time limitations, all of which hinder the provision of tailored feedback that meets individual student needs (Mao & Crosthwaite, 2019; Er et al., 2021). As a result, discrepancies often arise between the feedback provided and the pedagogical principles underlying effective writing instruction.
2.3. AI-Generated Feedback
Owing to the dilemmas teachers face with feedback provision against rapid advances in technology, research towards feedback practices now covers the integration of computer-generated feedback (K. Hyland & F. Hyland, 2006). In the current writing teaching context, AI-generated feedback has been proposed as a potential solution to enhance the quality and value of feedback received by EFL students. Although ethical concerns—such as high-tech plagiarism (Stewart, 2023), threats to human intelligence and academic integrity (Weissman, 2023), and the risk of generating incorrect feedback (Koltovskaia, 2023)—have been raised, the use of AI in education continues to grow. Researchers are increasingly exploring AI’s role in providing writing feedback (Guo & Wang, 2024). Given AI’s ability to analyze language patterns (Rudolph et al., 2023), it holds promising potential for supporting writing tasks such as correcting grammar and rephrasing sentences.
In terms of advantages, AI-generated feedback can produce personalized, meaningful, timely, and constantly manageable feedback, improving writing accuracy (Ghafouri et al., 2024; Mizumoto & Eguchi, 2023). In addition, it can be immediately provided without limitations and allow writing instructors to spend more time on global errors. However, compared to teacher-generated feedback, it expressed some limitations regarding provision for individual differences (Yoon et al., 2023). While teacher-generated feedback can take various forms and scopes, AI-generated feedback typically reproduces the explicitness of its feedback (direct corrections) in an unfocused manner, potentially negatively affecting learners’ uptake (Woodworth & Barkaoui, 2020).
The nature and quality of teacher- and AI-generated feedback demonstrate significant discrepancies, particularly in terms of detail, consistency, and focus. Research by Guo and Wang (2024) and Yoon et al. (2023) highlights that AI-generated feedback tends to be lengthier and more overwhelming, often containing redundant or overly generic suggestions. While this extensive feedback covers a wide range of issues (content, language, and organization), it can lack context-specific insights tailored to the individual student’s work. In contrast, teacher-generated feedback is typically concise, formative, and context-sensitive, focusing on critical areas for improvement in a more targeted manner. For instance, Guo and Wang (2024) and Yoon et al. (2023) compared teacher- and AI-generated feedback with respect to feedback quantity and focus. Their findings revealed that while AI-generated feedback was greater in volume and primarily direct, it was evenly distributed across content, language, and organizational issues. In contrast, teacher-generated feedback tended to focus selectively on content-specific and language-specific aspects, with a preference for indirect feedback types, such as “informative” or “querying” remarks. Furthermore, Lin and Crosthwaite’s (2024) comparative study on teacher- and ChatGPT-generated feedback sheds light on these distinctions. Teacher-generated feedback typically combines direct correction and indirect prompts addressing both local (e.g., grammar and mechanics) and global (e.g., content and organization) issues.
However, it is not without limitations, often displaying inconsistencies or inaccuracies. Meanwhile, AI-generated feedback frequently adopts metalinguistic explanations or reformulations of the original text but demonstrates variability in its approach to feedback provision. Even when identical prompts are used on the same text, AI’s feedback may vary significantly, raising concerns about consistency and reliability.
These findings underscore the potential of integrating teacher- and AI-generated feedback to harness their complementary strengths. While teacher-generated feedback provides nuanced and contextualized insights, AI-generated feedback offers volume and immediacy, making it a promising supplement to traditional feedback practices. This raises the question of whether AI-generated feedback and human-generated feedback should be utilized in a complementary manner to capitalize on their respective strengths while addressing their shortcomings. Nonetheless, careful consideration is needed to address the limitations of both forms and explore how they can best work together to enhance writing development. Although we cannot fully control the feedback produced by AI technologies, we can look forward to ongoing updates and improvements in these tools over time. This study aims to examine how the teacher- and AI-generated feedback complement each other by analyzing the impact of two forms of feedback (teacher- and AI-generated) on student revisions and the orders in which the two types of feedback have been executed as a solution to challenges in feedback provision.
2.4. Previous Studies on AI-Assisted Feedback in Teaching EFL Writing
In EFL writing classrooms, AI has demonstrated significant potential in supporting various learning tasks, such as argumentative writing (Su et al., 2023) and digital multimodal composition (Liu et al., 2024). Moreover, research has explored the multifaceted role of AI tools in providing feedback on EFL writing (Guo & Wang, 2024). These studies collectively underscore the transformative potential of AI-generated feedback in enhancing students’ language learning experiences.
Previous studies have highlighted the promise of AI-generated feedback in improving language learning outcomes. For instance, Kasneci et al. (2023) emphasized the potential of incorporating ChatGPT into education, noting the advancements of large language models in natural language processing. However, they also stressed the importance of ethical considerations, as the effective use of ChatGPT requires educators to guide learners in leveraging this tool responsibly. Similarly, Kohnke et al. (2023) explored the challenges and opportunities of ethically integrating ChatGPT into language learning. They emphasized the need for teacher guidance to ensure that learners can maximize the educational benefits of AI tools.
In the context of writing instruction, Mizumoto and Eguchi (2023) examined ChatGPT’s role in automated essay scoring for L2 learners. Their findings demonstrated that AI-driven features could enhance scoring accuracy and provide effective corrective feedback, establishing AI tools as valuable resources for classroom instruction. Building on these findings, Mizumoto et al. (2024) validated ChatGPT’s capability in assessing linguistic accuracy by comparing its performance against human evaluators and Grammarly. The results revealed a strong correlation between ChatGPT’s assessments and human evaluations, underscoring its reliability in language accuracy assessment.
AI tools also play a significant role in facilitating feedback processes. T. J. Allen and Mizumoto (2024) investigated how Japanese EFL learners utilized ChatGPT and AI tools in writing groups for editing and proofreading. Their study found that AI-assisted feedback improved the clarity and cohesion of learners’ writing. Additionally, ChatGPT provided learners with tailored prompts, encouraging deeper engagement with the revision process. Similarly, Ghafouri et al. (2024) explored the impact of ChatGPT on self-efficacy in a 10-week writing program. The study reported significant improvements in the writing performance of EFL learners, further demonstrating the potential of AI-generated feedback to enhance writing skills.
To summarize, these studies highlight the multifaceted benefits of AI-generated feedback in EFL writing classrooms. AI tools, such as ChatGPT, have shown the potential to significantly enhance students’ writing performance across various dimensions, including linguistic accuracy, clarity, cohesion, and self-efficacy. However, there remains a need for further research to explore the complementary use of AI- and teacher-generated feedback and to investigate effective strategies for integrating these feedback sources in writing instruction.
2.5. Gemini AI as Feedback Provider
Generative Artificial Intelligence (GenAI) chatbots have increasingly supported students in the academic writing process, providing particular benefits for EFL learners by enhancing their writing skills (Kim et al., 2024). GenAI refers to intelligent algorithms capable of interacting with users through natural language exchanges, producing new content, refining responses based on user needs, and simulating human-like intelligence (Choudhuri et al., 2023). For instance, GenAI systems offer continuous support throughout various stages of the writing process—from brainstorming and idea generation to revising and proofreading, including error correction, grammar checks, and stylistic adjustments (Rowland, 2023).
One example of advanced GenAI technology is Gemini, a next-generation chatbot developed by Google DeepMind. Built upon Google’s extensive research in artificial intelligence, Gemini uses advanced natural language processing and machine learning to engage users responsively (Google DeepMind, 2023). Among its functions, Gemini provides real-time feedback on language tasks, including creative writing support, grammar and stylistic corrections, and assistance with complex summarization and rephrasing (Rudolph et al., 2023). Gemini’s flexibility enables it to generate insightful suggestions across various fields, making it a versatile tool in educational and professional contexts (Guo & Wang, 2024).
2.6. Rationale and Research Gaps
In the context of EFL writing, limited research has focused on GenAI tools, particularly Gemini, in the context of feedback provision for writing. Even fewer studies have explored the integration of Gemini-generated feedback with teacher-generated feedback, leaving significant gaps in understanding how these two forms of feedback can be combined effectively.
Firstly, a review of the literature on teacher- and AI-generated feedback reveals several gaps that warrant further investigation. While previous studies have highlighted the benefits and limitations of both feedback types (Dai et al., 2023; Guo & Wang, 2024; Lin & Crosthwaite, 2024), researchers have consistently emphasized the potential advantages of integrating teacher-generated and AI-generated feedback to maximize their strengths and mitigate their respective weaknesses (Camarata & Slieman, 2020; Zainurrahman & Rodliyah, 2024). However, there is insufficient empirical evidence addressing the practicalities of such integration. Specifically, it remains unclear how to optimize the use of both feedback types to ensure the best outcomes for student writing development.
Secondly, one key issue concerns providing feedback. When integrating teacher- and AI-generated feedback, previous research has not adequately explored whether the order in which the feedback is delivered—AI-generated feedback followed by teacher-generated feedback or vice versa—affects the quality of student revisions. Given the assumption that different the order in which the two types of feedback have been executed may yield distinct outcomes, more evidence is necessary to identify the most effective order for enhancing writing performance. Such findings could offer valuable insights for both writing instructors and researchers seeking to improve feedback practices in EFL contexts.
Finally, another notable gap pertains to the technologies employed for AI-generated feedback. While much of the existing research has centered on widely used tools such as ChatGPT, relatively little attention has been given to newer generative AI platforms like Gemini. This lack of focus limits our understanding of how Gemini-generated feedback compares to other AI tools in terms of impact on the students’ revisions, especially when integrated with teacher-generated feedback.
To address these gaps, the present study examines the impact of two feedback forms (teacher-based and AI-based feedback) and two orders in which the two types of feedback have been executed (AIGF-TGF and TGF-AIGF) on student writing revisions. This research seeks to provide empirical evidence on how the integration of teacher- and AI-generated feedback can be optimized for the best learning outcomes. The study is guided by the following research questions:
To what extent do the two forms (AIGF and TGF) impact student revisions?
To what extent do the two orders in which the two types of feedback have been executed (AIGF-TGF and TGF-AIGF) impact student revisions?
3. Methodology
3.1. Participants
A total of 14 Vietnamese students who were in the second year of a four-year bachelor’ program enrolled in an essay writing course and participated in this study. The participants were all English majors. According to a school-administered English test, these students demonstrated low to intermediate English proficiency. All participants consented to participate. They were assured that their participation was voluntary and that their work would be anonymized.
The course, designed to improve students’ essay-writing skills, covered areas such as proofreading, editing, composing, and providing peer feedback. Structured over a 15-week semester, the course incorporated multiple assignments focused on enhancing students’ writing abilities. Specifically, four major assignments were assigned: two opinion essays and two cause-and-effect essays, approached through a process-based method involving stages of pre-writing, drafting, revising, editing, and publishing. The first task, two opinion essays, was assigned in the fourth and eighth weeks, followed by the two cause-and-effect essays in the tenth and thirteenth weeks.
Despite efforts to recruit a larger number of participants, fulfilling the study’s protocols required a high level of commitment, as students were expected to complete four essays and follow two distinct feedback sequences. Only participants who demonstrated a strong willingness to comply with these guidelines were included. These requirements, alongside the need for a controlled environment, contributed to the relatively small sample size. However, such focused sampling is common in exploratory studies that aim to provide foundational insights into the application of emerging technologies in educational contexts.
Similarly, previous studies exploring the impact of technology-based feedback in EFL writing instruction have also employed small sample sizes due to similar methodological constraints. For instance, Yang et al. (2024) examined in detail how five EFL students interacted with technology-based feedback, highlighting the challenges of maintaining participant engagement throughout the study. Along the same lines, Utami et al. (2023) conducted a case study with three Indonesian students from three senior high schools in Central Java, mapping perceptions, obstacles, and recommendations for optimizing AI use in academic writing instruction. The combination of qualitative and quantitative methods necessitated a focused participant group to ensure rich data collection. Supporting this pattern, Woo et al. (2024) explored the effectiveness of AI-generated feedback in enhancing writing quality among 23 EFL students, demonstrating that even with a relatively small sample, meaningful insights could be drawn about the impact of AI assistance. In a similar vein, Wang (2024) compared the effectiveness of teacher-generated feedback with that of an AI-driven application using a sample of 25 students, necessitating close monitoring of student responses and revision behaviors.
These studies, like the present research, illustrate that small sample sizes are a common feature in preliminary investigations of AI-driven feedback in EFL writing. They allow for in-depth analysis and provide valuable insights that inform future large-scale studies on technology-assisted language.
3.2. Materials and Instruments
The materials for this research consisted of four assignments provided by the Writing 4 instructor, as well as students’ essays, including both first and revised drafts. The assignments covered two types of essays—opinion and cause-and-effect—focusing on topics found in students’ textbooks, such as animals, medicine, life span, and smartphones. Students were instructed to write essays of approximately 250 words. The first drafts were then shared via Google Drive to receive both teacher- and AI-generated feedback. The feedback provided and subsequent revisions served as the key materials for this study.
3.3. Procedure
3.3.1. Session 1—Gemini Introduction Stage—Weeks 1 and 3
As mentioned above, the course lasted for 15 weeks. During the first two weeks of the study, students were first introduced to Gemini. This introduction included explaining what Gemini is and how it can be used (e.g., specific prompts). We also discussed the ethical concerns, such as plagiarism. Students registered for a free account and received a list of specific prompts (see Appendix A) to help them edit and proofread their paragraphs. Once students completed editing and proofreading their essays using Gemini-generated feedback, they showed a screenshot of the chat to the instructor.
3.3.2. Session 2—Data Collection Stage—Weeks 4–7 and Weeks 9–13
Once the introduction stage was completed, data collection began. During a 15-week semester, students spent the nine weeks on the writing tasks, divided into two smaller periods: (1) the first period was from weeks 4 to 7 for opinion essays, and the second period was from weeks 9 to 12 for cause-and-effect essays. Week 13 was for revision, weeks 8 and 14 were for mid- and end-term exams, and week 15 was for course evaluation and marking release. For each type of essay, students had two weeks to complete a four-stage writing cycle, which included brainstorming, drafting, revising, and editing.
Table 1 describes the research procedures of the four writing tasks. For the first and third writing tasks, the students followed the TGF-AIGF order, and for the second and fourth the AIGF-TGF order. In both orders in which the two types of feedback have been executed, after writing the first drafts, students uploaded their writing to Google Drive and shared it with their instructor. Then, in the TGF-AIGF order, students received teacher-generated feedback first, followed by Gemini-generated feedback. In the AIGF-TGF order, students received Gemini-generated feedback and then received teacher-generated feedback. Participants were instructed to complete their revisions for each essay within a one-week timeframe after receiving feedback for each draft. They were also informed about the ethical implications of using external writing aids and encouraged to base their revisions solely on the feedback provided. These measures aimed to standardize the revision process while maintaining a level of ecological validity.
The last step was the same in both orders, where students revised their work using the feedback they had received. All revisions were automatically saved in Google Drive. Once revised drafts were due, they were downloaded to the researcher’s computer for subsequent analysis. The students who were taking the Writing 4 course also sent the revised work to their writing instructor for evaluation.
To ensure the effectiveness of the instructional intervention, students followed a structured two-stage feedback process. In both the TGF-AIGF and AIGF-TGF sequences, students first received feedback from the initial source (teacher or AI), revised their drafts accordingly, and then submitted a first revised version. This revised draft was then subjected to feedback from the alternate source (AI or teacher) in the second stage, resulting in a second revised version. These procedures ensured that the feedback provided in the second stage was complementary to, and built upon, the revisions made during the first stage.
For example, in the TGF-AIGF order, students initially revised their drafts based on teacher-generated feedback, which primarily addressed global concerns such as content and organization. The AI-generated feedback in the second stage reviewed these revisions and provided additional surface-level suggestions for grammar and vocabulary. Similarly, in the AIGF-TGF sequence, AI-generated feedback served as the foundation for the first revision by addressing local issues, while teacher-generated feedback in the second stage focused on higher-order revisions. To enhance transparency in the revision process, students were required to highlight the changes made during the first stage and indicate the feedback source (teacher or AI) that informed these revisions. For the second stage, additional comments from either the teacher or the AI explicitly identified further revisions needed, ensuring that the contributions of each feedback stage were clearly delineated.
3.4. Data Analysis
3.4.1. Coding Student Revisions
In line with Cho and MacArthur (2010), the “Compare” function in Microsoft Word was utilized to track all modifications students made from their original drafts. After identifying these changes, an adapted framework based on Hall’s (1990) study was applied to categorize student revisions. This framework facilitated a detailed examination of the revisions, specifically focusing on their origin and the areas they addressed. Revision origin indicates the source, distinguishing between changes prompted by AI-generated feedback, teacher-generated feedback, both (teacher-generated feedback and AI-generated feedback), or self-corrections. Revision origin identifies the source of changes made during the writing process, categorized into four types. Teacher-generated feedback refers to revisions prompted solely by feedback from the teacher, targeting specific areas for improvement. AI-generated feedback involves changes based exclusively on suggestions provided by an AI tool, which analyzes the text and offers recommendations without human intervention. Both teacher-generated feedback and AI-generated feedback occur when feedback from both sources addresses the same errors, requiring students to integrate and prioritize these inputs in their revisions. Finally, self-correction refers to instances where students independently identify and correct their own errors without relying on external feedback. Revision areas, meanwhile, classify the scope of revisions as either local or global. Local revisions address elements such as grammar, vocabulary, spelling, and mechanics, while global revisions encompass broader adjustments, including content, organization, and the development of ideas.
To ensure the reliability of the coding process, two independent raters were trained on the coding definitions and applied the framework to a subset of 15% of the student essays. Inter-rater reliability was assessed using Cohen’s kappa, resulting in a kappa coefficient of 0.87, which indicates a high level of agreement between raters. Any discrepancies were resolved through discussion, and the coding scheme was refined to maintain consistency across all revisions. This robust approach ensured that the analysis of revision origin and areas was reliable and contributed to the validity of the study’s findings.
3.4.2. Students’ Essay Revision Analysis
To address the two research questions, student revisions from each feedback form across the four essay tasks were quantified. Descriptive and inferential statistics, specifically paired-samples t-tests and one-way ANOVA, were then conducted to evaluate the extent and types of revisions resulting from each feedback source. Revisions were grouped based on the feedback orders: revisions from Essays 1 and 3 for the teacher-generated feedback followed by AI-generated feedback (AIGF-TGF) order, and from Essays 2 and 4 for the AI-generated feedback followed by teacher-generated feedback (AIGF-TGF) order. A paired-samples t-test was then performed to compare the revisions derived from the two orders. All statistical analyses were conducted using SPSS (version 25).
Since the essays varied in length, revisions were standardized by calculating revisions per 100 words, following the approach taken in prior research (Pham, 2022). For instance, if Student A made 13 revisions in a 240-word essay, while Student B made 9 revisions in a 290-word essay, the number of revisions was standardized by multiplying the raw revisions count by 100 and dividing it by the total word count. Thus, the revisions per 100 words for Student A’s essay equaled 5.41, whereas for Student B’s essay it was 3.10.
3.4.3. Students’ Essay Marking
Participants’ L2 writing abilities were assessed indirectly through essay scores at the end of each feedback sequence. These scores, ranging from 0 to 100, were used to analyze the correlation between the number of revisions and writing performance. The marking rubric was adopted from Shehadeh (2011), evaluating EFL learners’ writing performance in writing essays based on five subcomponents, including content, organization, vocabulary, grammar, and mechanics. The marking rubric for the essay-writing marking is presented in Table 2.
Two trained raters independently assessed each essay using the rubric. Any discrepancies greater than five points were resolved through discussion to ensure scoring reliability.
The data were analyzed using both descriptive and inferential statistics to examine the relationship between the number of revisions and essay scores after each feedback sequence across four essays. A Pearson correlation analysis was conducted to determine the strength and significance of the relationship between these two variables. The data met the normality distribution and homogeneity of variance. SPSS (Version 25) was used to compute the Pearson correlation coefficient, which measures the linear relationship between variables.
4. Results
4.1. RQ1: To What Extent Do the Two Forms (AIGF and TGF) Impact Student Revision?
Table 3 presents the results of student revisions across four essays, focusing on two main criteria: including the origin of revisions (including TGF, AIGF, both TGF and AIGF, and self-correction) and the areas of revisions (including local and global).
In terms of the origin of revisions, the data reveal that AI-generated feedback consistently resulted in higher revision frequencies across all four essays compared to teacher-generated feedback alone, with a mean of 1.74 revisions per essay (SD = 0.20) for AI-generated feedback versus 1.04 revisions per essay (SD = 0.16) for teacher-generated feedback. This indicates that students were more likely to engage in revisions when provided with AI-generated feedback rather than teacher-generated feedback alone. When both teacher-generated feedback and AI-generated feedback were combined, the frequency of revisions was the highest, reaching a mean of 1.98 revisions per essay (SD = 0.29). Self-correction, in contrast, had the lowest revision frequency across all essays, with a mean of 0.98 revisions per essay (SD = 0.22), suggesting that students were less inclined to revise independently without external feedback.
The results of one-way ANOVA tests showed that there was a significant effect of the integration between teacher- and AI-generated feedback on student revisions (F (3,55) = 68.42, p < 0.05).In addition, post hoc comparisons using Tukey’s range test indicated that the mean revision scores for both teacher- and AI-generated feedback (M = 1.98, SD = 0.29) and AI-generated feedback alone (M = 1.74, SD = 0.20) were significantly different from those for teacher-generated feedback alone (M = 1.04, SD = 0.16) and self-corrections (M = 0.98, SD = 0.22). These findings suggest that the AI-generated feedback and the combined teacher- and AI-generated feedback approach have a greater impact on student revisions than the teacher-generated feedback alone or self-initiated corrections.
In terms of areas of revisions, both local and global revisions were observed. Global revisions, focusing on content and organization, had a mean frequency of 2.78 revisions per essay (SD = 0.48). Local revisions, involving sentence-level changes, such as grammar, vocabulary, and mechanics, showed a slightly higher mean frequency of 2.97 revisions per essay (SD = 0.33). This higher frequency of local revisions may reflect students’ efforts to address surface-level aspects of their writing, possibly influenced by the detailed feedback provided by both AI and teachers. However, independent t-test results did not show any significant difference between revisions made at the local and global levels of the text (t (26) = 1.25, p > 0.05). This finding suggests that, while students most frequently made revisions at the word (e.g., vocabulary) and surface (e.g., grammar) levels, they maintained a relatively balanced approach, addressing both local and global aspects of their essays.
To explore the relationship between revision frequency and writing quality, a correlation analysis was conducted between the number of revisions and essay scores across four essays after each feedback sequence. Descriptive statistics and the correlation results are presented in Table 4.
As shown in Table 4, the number of revisions and essay scores demonstrated significant correlations in three out of four essays, with r values ranging from 0.55 to 0.77. Specifically, a moderate positive correlation was identified for essays 2 (r = 0.63, p = 0.01) and 4 (r = 0.55, p = 0.04), suggesting that a higher frequency of revisions was associated with improved essay scores. For essay 3, a strong positive correlation was observed (r = 0.77, p < 0.01), indicating a substantial relationship between revision activity and writing performance. In contrast, no significant correlation was found for essay 1 (r = 0.35, p = 0.21), suggesting that revisions in the initial essay may not have contributed as strongly to performance improvement as in later essays.
4.2. RQ2: To What Extent Do the Two Orders (TGF-AIGF and AIGF-TGF) Impact Student Revisions?
Table 5 presents a comparison of student revisions between the two feedback orders: TGF-AIGF and AIGF-TGF. The paired-samples t-test results showed no statistically significant differences between the two orders in terms of either the origin or areas of revisions (p > 0.05). However, a closer examination of the mean scores offers insights into the relative impact of each order.
In terms of the origin of revisions, students in the AIGF-TGF order made slightly more revisions across all categories compared to the TGF-AIGF order. Specifically, revisions based on teacher-generated feedback were higher in the AIGF-TGF order (M = 1.16) than in the TGF-AIGF order (M = 0.94). Similarly, revisions attributed to AI-generated feedback were slightly more frequent when students began with AI-generated feedback (M = 1.82) compared to those starting with teacher-generated feedback (M = 1.68). Additionally, combined revisions based on both feedback sources were greater in the AIGF-TGF order (M = 2.02) compared to the TPF-AIPF order (M = 1.91). However, self-correction showed an opposite trend, with slightly higher revisions in the TGF-AIGF order (M = 1.00) than in the AIGF-TGF order (M = 0.98), although this difference was minimal and statistically insignificant.
Regarding areas of revisions, the AIGF-TGF order also showed marginally higher mean scores. Local revisions, which focus on sentence-level issues such as grammar, vocabulary, and mechanics, were slightly more frequent in the AIGF-TGF order (M = 3.06) compared to the TGF-AIGF order (M = 2.89). Global revisions, addressing higher-order concerns like content and organization, were similarly more common in the AIGF-TGF order (M = 2.92) than in the TGF-AIGF order (M = 2.64). While these differences approached significance for local revisions (t = 1.77, p = 0.10), neither category demonstrated statistically significant results.
In addition to student revisions in two orders, Table 6 summarizes the mean (M) and standard deviation (SD) of global and local revisions across two feedback stages for TFG-AIGF and AIGF-TFG orders. Regarding the sequence of TGF-AIGF, the descriptive statistics reveal that there was a significant decrease in global revisions from stage 1 (M = 1.72, SD = 0.33) to stage 2 (M = 0.92, SD = 0.13), as indicated by a paired t-test result of t (13) = 7.89, p < 0.01). Conversely, local revisions significantly increased during this order, with mean scores rising from stage 1 (M = 1.26, SD = 0.23) to stage 2 (M = 1.63, SD = 0.30), supported by t (13) = −3.64, p = 0.01).
In contrast, the AIGF-TGF order demonstrated a reverse trend. Local revisions significantly decreased from stage 1 (M = 1.75, SD = 0.30) to stage 2 (M = 1.31, SD = 0.21), with a corresponding t-value of t (13) = −5.92, p < 0.01. At the same time, global revisions increased substantially, with mean scores improving from stage 1 (M = 1.21, SD = 0.33) to stage 2 (M = 1.71, SD = 0.45), reflected in t (13) = −5.20, p = 0.01).
5. Discussion and Pedagogical Implications
This study investigated the effects of teacher-generated feedback, AI-generated feedback, and their integration (TGF-AIGF and AIGF-TPF orders) on student revisions and writing quality in an EFL context. The findings offer valuable insights into how different feedback forms and orders influence student revisions and provide pedagogical implications for improving writing instruction.
5.1. Feedback Forms: AI-Generated Feedback vs. Teacher-Generated Feedback
The results indicate that students engaged more frequently with revisions when provided with AI-generated feedback compared to teacher-generated feedback alone. This finding can be attributed to the specific and actionable nature of AI-generated feedback, which often focused on surface-level issues such as grammar, spelling, and vocabulary. Such feedback is typically easier and faster to implement, requiring fewer cognitive resources compared to teacher-generated feedback, which primarily targeted higher-order concerns like logic, coherence, and argumentation. The complexity of teacher-generated feedback may have contributed to students perceiving it as more challenging to address, potentially leading to lower engagement in revisions. This distinction between the ease of implementing surface-level versus higher-order feedback aligns with prior studies, such as those by Montgomery and Baker (2007) and Mao and Crosthwaite (2019), which highlight the cognitive demands of addressing global writing issues.
The findings also suggested that the combination of teacher-generated feedback and AI-generated feedback had a greater impact on student revisions than either teacher-generated feedback alone or self-correction. This result highlights the complementary role of teacher- and AI-generated feedback. This integration is necessary and potentially effective as it leverages the strengths of both feedback sources. According to the literature, AI-generated feedback provides a greater volume of direct, concise, and formative suggestions, often focused on surface-level issues like grammar and vocabulary. In contrast, teacher-generated feedback offers more contextualized, selective insights tailored to both content-specific and language-specific aspects of students’ work (Guo & Wang, 2024; Yoon et al., 2023). By combining these approaches, the integration minimizes the lack of context-specific insights inherent in AI-generated feedback while reducing the overburden on teachers in providing feedback.
As advanced yet accessible technology, AI tools are increasingly being adopted by students outside the classroom. Therefore, EFL teachers must proactively introduce AI to learners, providing clear explanations of its strengths and limitations for delivering feedback. For instance, while AI excels at identifying linguistic and structural issues, it lacks the emotional depth and lived experience necessary to recognize and nurture the unique voice and identity present in student writing. Acknowledging these limitations is essential to fostering a balanced understanding of AI’s role in writing development. Additionally, integrating AI-generated feedback offers an opportunity to cultivate students’ self-regulated learning skills. By guiding students to critically evaluate the feedback generated by AI, teachers can empower them to discern the utility of suggestions and make informed revisions. Such metacognitive engagement not only enhances the quantity of revisions made by students but also promotes autonomy in learning (Barrot, 2023).
The findings of the first research question initially explored the potential role of AI-generated feedback as a supplementary tool alongside teacher-generated feedback. However, the question of how the sequence of feedback—whether teacher-generated feedback followed by AI-generated feedback (TGF-AIGF) or vice versa (AIGF-TGF)—impacts student revisions will be further discussed in the following sections.
5.2. Areas of Revision: Local vs. Global
Both local and global revisions were observed, with local revisions (M = 2.97, SD = 0.33) slightly outpacing global revisions (M = 2.78, SD = 0.48). This indicates that students focused more on sentence-level changes, such as grammar, vocabulary, and mechanics, than on higher-order issues like content and organization. The higher frequency of local revisions likely reflects the influence of AI-generated feedback, which excels at identifying specific surface-level errors and providing actionable suggestions for improvement. However, the balanced approach observed between local and global revisions suggests that the integration of AI- and teacher-generated feedback effectively encouraged students to address both surface-level and content-related issues in their writing. This dual focus highlights the potential for AI-generated feedback to serve as a complementary tool in guiding students through comprehensive text revisions.
The findings also raise the question of whether AI-generated feedback can replace or supplement peer feedback in writing instruction. Previous studies on online peer feedback have highlighted an imbalance between local and global revisions, with peer feedback often focusing more on surface-level issues, such as grammar and mechanics, while overlooking content-related aspects (Pham, 2022; Tran & Ma, 2024; Zhang, 2020). This uneven distribution is attributed to peers’ limited expertise and confidence in providing critical, higher-order feedback (Topping, 2009; Zong et al., 2022). In contrast, AI-generated feedback demonstrates a more consistent distribution across local and global issues (Guo & Wang, 2024) as well as a high agreement with instructors (Dai et al., 2023), suggesting that it may address some limitations of peer feedback.
5.3. Feedback Orders: AIGF-TGF vs. TGF-AIGF
Although no statistically significant differences were found between the two feedback orders (Table 4), the mean scores suggest that the AIGF-TGF order slightly outperformed the TGF-AIGF order in both origin and area categories. The slight advantage of the AIGF-TGF order can be attributed to the nature of AI-generated feedback, which is particularly effective at addressing surface-level issues, such as grammar, vocabulary, and basic organization (Guo & Wang, 2024; Lin & Crosthwaite, 2024; Yoon et al., 2023). This initial focus on lower-order concerns allows students to refine their drafts to a level where teachers can more effectively address higher-order concerns, such as idea development, content coherence, and argument structure. By working on a more polished draft in the second stage, teachers can provide targeted, cognitively demanding feedback that goes beyond the capabilities of AI tools, ultimately enhancing the quality of student revisions. This sequencing could maximize the strengths of both feedback sources, enhancing revision quality and student engagement. Moreover, this order aligns with cognitive load theory (Evans et al., 2024), as students first address less cognitively demanding issues (surface-level concerns) before progressing to more complex tasks (higher-order revisions).
Secondly, the findings regarding student revisions across feedback stages (Table 6) further support the above argument by highlighting the distinct revision patterns that emerged for the two feedback orders. For TGF-AIGF, global revisions significantly decreased from stage 1 to stage 2, while local revisions increased. In contrast, the AIGF-TGF sequence demonstrated a reverse pattern, with local revisions decreasing and global revisions increasing. These patterns underline the complementary roles of AI and teacher-generated feedback, as AI focuses on surface-level issues while teacher-generated feedback fosters higher-order thinking and structural improvements (Bitchener & Storch, 2016). This indicates that the order of feedback delivery can influence the focus of student revisions, with potential implications for designing effective feedback strategies.
In addition to a larger number of revisions in the AIGF-TGF order, the findings of this study also highlight a moderate to strong positive correlation between the number of revisions and essay scores in essays 2, 3, and 4, indicating that effective revision practices facilitated by the two feedback orders positively impact students’ writing performance. Specifically, moderate correlations were observed in essays 2 and 4, which were part of the AIGF-TGF order. Meanwhile, a strong positive correlation was noted in essay 3, belonging to the TGF-AIGF order. These results align with prior research emphasizing the significant impact of targeted revisions on writing quality when feedback is effectively utilized (Ferris, 2014; Bitchener & Storch, 2016).
The absence of a significant correlation in essay 1 can be attributed to participants’ initial unfamiliarity with the iterative feedback and revision process. Conversely, the stronger correlations observed in later essays suggest that students developed greater proficiency in utilizing feedback as they gained more experience with the process. While the findings do not conclusively identify which feedback order was more effective, they further confirm the complementary roles of AI-generated feedback and teacher-generated feedback in supporting effective revisions. This synergy between feedback forms likely contributed to students’ improved writing outcomes over time. These findings resonate with recent studies highlighting the potential of integrating AI-driven tools with traditional feedback methods to enhance writing performance (Aladini, 2023; Marzuki et al., 2023).
To implement the AIGF-TGF order effectively, educators should provide clear guidance to students on how to interpret and use AI-generated feedback. For example, teachers could introduce activities that help students distinguish between surface-level and global issues and use AI writing tools to address the former before seeking teacher-generated feedback for the latter. Additionally, EFL teachers should emphasize the complementary nature of AI- and human-generated feedback, encouraging students to view AI-generated suggestions as a foundation rather than a final solution. Teachers should also train students to take advantage of AI-generated feedback as a tool to enhance their language knowledge and to practice effective prompting or navigating AI tools using frameworks such as the Role-Task-Format template to obtain formative and valuable feedback. Furthermore, training teachers to critically evaluate and supplement AI-generated feedback is crucial. By understanding the strengths and limitations of AI tools, teachers can provide informed, higher-order guidance that builds on AI-generated revisions. It is believed that each feedback order—AIGF-TGF and TGF-AIGF—has its own potential and impact, so EFL teachers should determine which order to use based on learning objectives, the students’ proficiency levels, and the focus of feedback practices. This comprehensive approach not only enhances the feedback process but also equips students with self-regulation and critical thinking skills, enabling them to engage more effectively with both AI tools and teacher-generated feedback in their writing journey.
5.4. Limited Self-Corrections
Self-correction demonstrated the lowest mean revision frequency, underscoring students’ heavy reliance on external feedback for making revisions. This finding diverges from Pham’s (2022) study, which reported that self-initiated revisions constituted nearly half of all revisions. The discrepancy may stem from differences in student proficiency levels, critical thinking capacity, and self-regulated learning abilities. These results highlight a critical issue: many students lack the autonomy and cognitive skills required to independently identify and address areas for improvement in their writing. The low frequency of self-correction raises concerns about students’ independence and learner agency in writing tasks, emphasizing the need for interventions aimed at developing these skills. Relying solely on external feedback, whether from teachers or AI, risks creating a dependency that can hinder students’ long-term writing development. Encouraging self-correction as part of the writing process is essential to fostering autonomous, self-regulated learners who can confidently navigate writing challenges without constant external input.
To address this issue, educators should integrate strategies that explicitly focus on enhancing students’ critical thinking and self-regulation skills. This can include providing guided practice in self-assessment, teaching students to recognize and evaluate their own errors, and offering metacognitive strategies to support independent problem-solving. For example, scaffolding techniques such as peer review activities, reflective journals, or structured self-checklists can help students develop the capacity to analyze their work critically and initiate revisions without external prompts. By integrating these strategies, educators can empower students to take greater ownership of their writing process, thereby promoting a balance between external feedback and self-correction. This shift not only enhances students’ independence but also equips them with the lifelong skills necessary for effective writing and learning.
In conclusion, the integration of AI- and teacher-generated feedback, particularly in a structured order, holds significant potential to improve student writing quality and revision practices. However, fostering learner independence and critical thinking remains essential to developing well-rounded, autonomous writers in EFL contexts.
6. Conclusions
This study underscores the potential of integrating AI-generated feedback and teacher-generated feedback to enhance student revisions in EFL academic writing. The findings reveal that combining the strengths of both feedback types—AI-generated feedback’s focus on specific, detailed corrections at the surface level (e.g., grammar, vocabulary) and TGF’s emphasis on global, higher-order issues (e.g., organization, coherence)—can result in more comprehensive revisions. While students most frequently addressed surface-level issues, the complementary nature of AI-generated feedback and TGF enabled a balanced approach to revisions, enhancing both local and global aspects of their essays.
In terms of feedback sequencing, students in the AIGF-TGF order made slightly more revisions than those in the TGF-AIGF order. This suggests that beginning with AI-generated feedback may effectively scaffold surface-level revisions, allowing students to focus more deeply on global concerns during subsequent teacher-generated feedback. Additionally, distinct revision patterns were observed, including a decrease in global revisions and an increase in local revisions from stage 1 to stage 2 in the TGF-AIGF order, while the reverse trend was noted in the AIGF-TGF order. These findings further highlight the complementary focus of each feedback type and the importance of sequencing in facilitating effective revisions.
Moreover, the study found moderate to strong positive correlations between revision frequency and essay scores in three out of four essays, indicating that revisions significantly contribute to improved writing performance. This highlights the iterative nature of the writing process and the role of feedback in driving meaningful revisions that enhance the quality of student writing.
These insights contribute to the growing body of research on AI-assisted writing instruction, providing evidence for the efficacy of integrating AI tools into pedagogical practices. By leveraging the complementary strengths of AI-generated feedback and teacher-generated feedback, educators can design feedback strategies that not only promote revision frequency but also address both surface-level and higher-order writing concerns. Future research could further explore the long-term effects of such integrated feedback approaches and investigate their applicability across diverse educational contexts.
However, this study has several limitations that should be acknowledged. First, the small sample size limits the generalizability of the findings, as the results may not be representative of broader student populations. Nonetheless, as a preliminary investigation, this study provides valuable insights into the integration of teacher- and AI-generated feedback, forming a basis for future research with larger and more diverse participant groups. Second, while the study primarily focused on the frequency of revisions, the analysis provides only indirect evidence linking revision frequency to improved essay scores, leaving the quality of individual revisions unexplored. Future studies should address this limitation by employing rubric-based assessments or qualitative analyses to evaluate the impact of revisions on specific aspects of writing, such as coherence, clarity, and argumentation. Finally, this study relied on a quantitative approach to examine the relationship between feedback and revisions. Future research should adopt mixed-method approaches to capture both quantitative data and qualitative insights into students’ experiences with feedback and its role in their writing development. By addressing these limitations, future investigations can offer a more comprehensive understanding of how different feedback types influence writing outcomes.
The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Review Board of University of Foreign Languages and International Studies, Hue University (DHH2024-07-15 approved on 8 April 2024).
Informed consent was obtained from all subjects involved in the study.
The data that support the findings of this study are available from the corresponding author on reasonable request.
The author declares no conflict of interest.
Footnotes
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Research procedure.
| Weeks | TGF-AIGF Order | AIGF-TGF Order |
|---|---|---|
| Weeks 1–3 | Gemini Introduction Stage | |
| Weeks 4–7 (Opinion Essays) | Essay 1 | Essay 2 |
|
| |
| Week 8 | Mid-term exam | |
| Weeks 9–12 (Cause-and-effect essays) | Essay 3 | Essay 4 |
| Week 13 | Revision | |
| Week 14 | End-of-term exam | |
| Week 15 | Course evaluation and marking release | |
The marking rubric for pre-tests and post-tests.
| Criteria | Points | Description ( |
|---|---|---|
| 1. Content | 20 points | “Knowledge of the subject; development of thesis statement; development of topic; relevance of supporting ideas; quality of main ideas and details.” ( |
| 2. Organization | 20 points | “Fluency of expression; clarity in the information organization; effective sequencing and development of ideas; effective use of transitions.” ( |
| 3. Vocabulary | 25 points | “Range; accuracy of word/idiom use; mastery of word forms; effectiveness in the transmission of meaning.” ( |
| 4. Grammar | 25 points | “Use of sentence structures and constructions; accuracy and correctness in agreement, number, tense, word order, articles, pronouns, prepositions, and negation.” ( |
| 5. Mechanics | 10 points | “Conventions of spelling; punctuation; capitalization; indentation” ( |
Student revisions in the two forms: TGF and AIGF.
| Student Revisions | Essay 1 | Essay 2 | Essay 3 | Essay 4 | Mean | SD |
|---|---|---|---|---|---|---|
| Origin of revisions | ||||||
| TGF | 1.01 | 1.19 | 0.86 | 1.11 | 1.04 | 0.16 |
| AIGF * | 1.58 | 1.65 | 1.79 | 1.95 | 1.74 | 0.20 |
| TGF and AIGF * | 1.85 | 1.80 | 1.96 | 2.33 | 1.98 | 0.29 |
| Self-correction | 1.39 | 1.12 | 0.62 | 0.82 | 0.98 | 0.22 |
| Areas of revisions | ||||||
| Local | 2.92 | 2.96 | 2.86 | 3.17 | 2.97 | 0.33 |
| Global | 2.91 | 2.80 | 2.37 | 3.04 | 2.78 | 0.48 |
Note: Revisions were measured per 100 words. * Statistically significant.
Correlations between number of revisions and essay scores.
| Essays | Total Revisions | Essay Scores | Pearson Correlation | p | ||
|---|---|---|---|---|---|---|
| M | SD | M | SD | |||
| Essay 1 | 5.83 | 1.03 | 77.57 | 4.07 | 0.35 | 0.21 |
| Essay 2 | 5.76 | 1.41 | 82.93 | 4.99 | 0.63 | 0.01 * |
| Essay 3 | 5.23 | 0.80 | 83.07 | 3.29 | 0.77 | 0.00 ** |
| Essay 4 | 6.21 | 0.91 | 86.71 | 4.16 | 0.55 | 0.04 ** |
Notes: * 0.01 < p < 0.05, ** p < 0.01, moderate correlation (r = 0.40–0.69); strong correlation (r = 0.70–0.89).
Student revisions in the two orders: TPF-AIPF and AIPF-TPF.
| Student Revisions | TGF-AIGF | AIGF-TGF | t | p |
|---|---|---|---|---|
| Origin of revisions | ||||
| TGF | 0.94 | 1.16 | 1.92 | 0.07 |
| AIGF | 1.68 | 1.82 | 1.02 | 0.32 |
| TGF and AIGF | 1.91 | 2.02 | 1.49 | 0.15 |
| Self-correction | 1.00 | 0.98 | 0.41 | 0.69 |
| Areas of revisions | ||||
| Local | 2.89 | 3.06 | 1.77 | 0.10 |
| Global | 2.64 | 2.92 | 1.25 | 0.23 |
Note: Revisions were measured per 100 words.
Student revisions across feedback stages in TGF-AIGF and AIGF-TGF orders.
| Feedback Orders | Stages | Stage 1 | Stage 2 | t | p | ||
|---|---|---|---|---|---|---|---|
| M | SD | M | SD | ||||
| TGF-AIGF | Global revisions | 1.72 | 0.33 | 0.92 | 0.13 | 7.89 | 0.00 |
| Local revisions | 1.26 | 0.23 | 1.63 | 0.30 | −3.64 | 0.01 | |
| AIGF-TGF | Global revisions | 1.21 | 0.33 | 1.71 | 0.45 | −5.20 | 0.00 |
| Local revisions | 1.75 | 0.30 | 1.31 | 0.21 | 5.92 | 0.01 | |
Appendix A
Sample prompts for revising and proofreading.
| Stages | Sample Prompts |
| Revising | 1. Act as an experienced English language writing coach [Role]. Evaluate my essay on [topic] [Task]. Provide feedback on task achievement, organization, vocabulary, and grammar in a structured list format [Format]. |
| Proofreading | 1. Act as a grammar and vocabulary expert [Role]. Identify and correct all grammatical errors, punctuation mistakes, and vocabulary inaccuracies in my essay [Task]. Provide an edited version of the essay along with a summary of changes [Format]. |
References
Abdullah, N.; Aziz, A. Between preaching and practicing—The incongruities of teachers’ written feedback beliefs and practices: A case study. Journal of Educational Research and Indigenous Studies; 2020; 1,
Aladini, A. AI applications impact on improving EFL university academic writing skills and their logical thinking. Al-’ulūm Al-Tarbawiyyaẗ; 2023; 31,
Allen, L. K.; McNamara, D. S. León, J. A.; Escudero, I. Five building blocks for comprehension strategy instruction. Reading comprehension in educational settings; John Bejamins: 2017; pp. 125-144. [DOI: https://dx.doi.org/10.1075/swll.16.05all]
Allen, T. J.; Mizumoto, A. ChatGPT over my friends: Japanese English-as-a-Foreign-Language learners’ preferences for editing and proofreading strategies. RELC Journal; 2024; [DOI: https://dx.doi.org/10.1177/00336882241262533]
Barrot, J. S. Using ChatGPT for second language writing: Pitfalls and potentials. Assessing Writing; 2023; 57, 100745. [DOI: https://dx.doi.org/10.1016/j.asw.2023.100745]
Bitchener, J.; Storch, N. Written corrective feedback for L2 development; Multilingual Matters: 2016.
Bonilla López, M. An updated typology of written corrective feedback: Resolving terminology issues. Educación (Universidad de Costa Rica); 2021; [DOI: https://dx.doi.org/10.15517/revedu.v45i1.43289]
Bozorgian, H.; Yazdani, A. Direct written corrective feedback with metalinguistic explanation: Investigating language analytic ability. Iranian Journal of Language Teaching Research; 2021; 9,
Camarata, T.; Slieman, T. A. Improving student feedback quality: A simple model using peer review and feedback rubrics. Journal of Medical Education and Curricular Development; 2020; 7, 2382120520936604. [DOI: https://dx.doi.org/10.1177/2382120520936604]
Chen, M.; Cui, Y. The effects of AWE and peer feedback on cohesion and coherence in continuation writing. Journal of Second Language Writing; 2022; 57, 100915. [DOI: https://dx.doi.org/10.1016/j.jslw.2022.100915]
Cho, K.; MacArthur, C. Student revision with peer and expert reviewing. Learning and Instruction; 2010; 20,
Choudhuri, R.; Liu, D.; Steinmacher, I.; Gerosa, M.; Sarma, A. How far are we? the triumphs and trials of generative AI in learning software engineering. IEEE/ACM 46th International Conference on Software Engineering; Lisbon, Portugal, April 14–20; 2023; pp. 1-13. [DOI: https://dx.doi.org/10.1145/3597503.3639201]
Crosthwaite, P.; Ningrum, S.; Lee, I. Research trends in L2 written corrective feedback: A bibliometric analysis of three decades of Scopus-indexed research on L2 WCF. Journal of Second Language Writing; 2022; 58, 100934. [DOI: https://dx.doi.org/10.1016/j.jslw.2022.100934]
Dai, W.; Lin, J.; Jin, H.; Li, T.; Tsai, Y.-S.; Gašević, D.; Chen, G. Can large language models provide feedback to students? A case study on ChatGPT. 2023 IEEE International Conference on Advanced Learning Technologies (ICALT); Orem, UT, USA, July 10–13; 2023; [DOI: https://dx.doi.org/10.1109/ICALT58122.2023.00100]
Er, E.; Dimitriadis, Y.; Gašević, D. Collaborative peer feedback and learning analytics: Theory-oriented design for supporting class-wide interventions. Assessment & Evaluation in Higher Education; 2021; 46,
Evans, P.; Vansteenkiste, M.; Parker, P.; Kingsford-Smith, A.; Zhou, S. Cognitive load theory and its relationships with motivation: A self-determination theory perspective. Educational Psychology Review; 2024; 36,
Faigley, L.; Witte, S. Analyzing review. College Composition and Communication; 1981; 32,
Ferris, D. R. Responding to student writing: Teachers’ philosophies and practices. Assessing Writing; 2014; 19, pp. 6-23. [DOI: https://dx.doi.org/10.1016/j.asw.2013.09.004]
Ferris, D. R.; Roberts, B. Error feedback in L2 writing classes: How explicit does it need to be?. Journal of Second Language Writing; 2001; 10,
Ghafouri, M.; Hassaskhah, J.; Mahdavi-Zafarghandi, A. From virtual assistant to writing mentor: Exploring the impact of a ChatGPT-based writing instruction protocol on EFL teachers’ self-efficacy and learners’ writing skill. Language Teaching Research; 2024; [DOI: https://dx.doi.org/10.1177/13621688241239764]
Goldstein, L. M. Questions and answers about teacher written commentary and student revision: Teachers and students working together. Journal of Second Language Writing; 2004; 13,
Google DeepMind. Introducing gemini: A multimodal AI model; 2023; Available online: https://blog.google/technology/ai/google-gemini-ai/ (accessed on 2 October 2024).
Guo, K.; Wang, D. To resist it or to embrace it? Examining ChatGPT’s potential to support teacher feedback in EFL writing. Education and Information Technologies; 2024; 29,
Hall, C. Managing the Complexity of Revising Across Languages. TESOL Quarterly; 1990; 24,
Hartshorn, K. J.; Evans, N. W.; Merrill, P. F.; Sudweeks, R. R.; Strong-Krause, D.; Anderson, N. J. Effects of dynamic corrective feedback on ESL writing accuracy. TESOL Quarterly; 2010; 44,
Hattie, J.; Timperley, H. The power of feedback. Review of Educational Research; 2007; 77,
Hyland, K.; Hyland, F. Feedback on second language students’ writing. Language Teaching; 2006; 39,
Kasneci, E.; Sessler, K.; Küchemann, S.; Bannert, M.; Dementieva, D.; Fischer, F.; Gasser, U.; Groh, G.; Günnemann, S.; Hüllermeier, E.; Krusche, S.; Kutyniok, G.; Michaeli, T.; Nerdel, C.; Pfeffer, J.; Poquet, O.; Sailer, M.; Schmidt, A.; Seidel, T.; Kasneci, G. ChatGPT for good? On opportunities and challenges of large language models for education. Learning and Individual Differences; 2023; 103, 102274. [DOI: https://dx.doi.org/10.1016/j.lindif.2023.102274]
Kim, J.; Ham, Y. H.; Lee, S. S. Differences in student-AI collaboration process on a drawing task: Focusing on students’ attitude toward AI and the level of drawing skills. Australasian Journal of Educational Technology; 2024; 40,
Kluger, A. N.; DeNisi, A.; Steinberg, R. J. The effects of feedback interventions on performance: A historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin; 1996; 119,
Kohnke, L.; Moorhouse, B. L.; Zou, D. ChatGPT for language teaching and learning. RELC Journal; 2023; 54,
Koltovskaia, S. Postsecondary L2 writing teachers’ use and perceptions of Grammarly as a complement to their feedback. ReCALL (Cambridge, England); 2023; 35,
Lee, I. Understanding teachers’ written feedback practices in Hong Kong secondary classrooms. Journal of Second Language Writing; 2008; 17,
Lee, I. Ten mismatches between teachers’ beliefs and written feedback practice. ELT Journal; 2009; 63,
Leki, I. The preferences of ESL students for error correction in college-level writing classes. Foreign Language Annals; 1991; 24,
Li, Y.; Sha, L.; Yan, L.; Lin, J.; Raković, M.; Galbraith, K.; Lyons, K.; Gašević, D.; Chen, G. Can large language models write reflectively. Computers and Education: Artificial Intelligence; 2023; 4, 100140. [DOI: https://dx.doi.org/10.1016/j.caeai.2023.100140]
Lin, S.; Crosthwaite, P. The grass is not always greener: Teacher vs. GPT-assisted written corrective feedback. System; 2024; 127, 103529. [DOI: https://dx.doi.org/10.1016/j.system.2024.103529]
Link, S.; Mehrzad, M.; Rahimi, M. Impact of automated writing evaluation on teacher feedback, student revision, and writing improvement. Computer Assisted Language Learning; 2022; 35,
Liu, M.; Zhang, L. J.; Biebricher, C. Investigating students’ cognitive processes in generative AI-assisted digital multimodal composing and traditional writing. Computers and Education; 2024; 211, 104977. [DOI: https://dx.doi.org/10.1016/j.compedu.2023.104977]
Mao, S. S.; Crosthwaite, P. Investigating written corrective feedback: (Mis)alignment of teachers’ beliefs and practice. Journal of Second Language Writing; 2019; 45, pp. 46-60. [DOI: https://dx.doi.org/10.1016/j.jslw.2019.05.004]
Marzuki,; Widiati, U.; Rusdin, D.; Darwin, D.; Indrawati, I. The impact of AI writing tools on the content and organization of students’ writing: EFL teachers’ perspective. Cogent Education; 2023; 10,
Mizumoto, A.; Eguchi, M. Exploring the potential of using an AI language model for automated essay scoring. Research Methods in Applied Linguistics; 2023; 2,
Mizumoto, A.; Shintani, N.; Sasaki, M.; Teng, M. F. Testing the viability of ChatGPT as a companion in L2 writing accuracy assessment. Research Methods in Applied Linguistics; 2024; 3,
Montgomery, J. L.; Baker, W. Teacher-written feedback: Student perceptions, teacher self-assessment, and actual teacher performance. Journal of Second Language Writing; 2007; 16,
Pham, H. T. P. Computer-mediated and face-to-face peer feedback: Student feedback and revision in EFL writing. Computer Assisted Language Learning; 2022; 35,
Piolat, A. Clapham, C.; Corson, D. Writers’ assessment and evaluation of their texts. Encyclopaedia of language and education. Flight. 7: Language testing and assessment; Kluwer Academic: 1997; pp. 189-198.
Reid, J. M. Teaching ESL writing; Prentice Hall Regents: 1993.
Rollinson, P. Using peer feedback in the ESL writing class. ELT Journal; 2005; 59,
Rowland, D. R. Two frameworks to guide discussions around levels of acceptable use of generative AI in student academic research and writing. Journal of Academic Language and Learning; 2023; 17,
Rudolph, J.; Tan, S.; Tan, S. ChatGPT: Bullshit spewer or the end of traditional assessments in higher education?. Journal of Applied Learning and Teaching; 2023; 6,
Shehadeh, A. Effects and student perceptions of collaborative writing in second language. Journal of Second Language Writing; 2011; 20,
Sommers, N. Leeds, B. Revision strategies of student writers and experienced adult writers. Writing in a second language: Insights from first and second language teaching and research; Longman: 1996; pp. 128-136.
Stewart, J. Noam Chomsky says ChatGPT is a form of “high-tech plagiarism”; My Modern Met: 2023; Available online: https://mymodernmet.com/noam-chomsky-chat-gpt/ (accessed on 10 October 2024).
Su, Y.; Lin, Y.; Lai, C. Collaborating with ChatGPT in argumentative writing classrooms. Assessing Writing; 2023; 57, 100752. [DOI: https://dx.doi.org/10.1016/j.asw.2023.100752]
Teng, M. F. A systematic review of ChatGPT for English as a foreign language writing: Opportunities, challenges, and recommendations. International Journal of TESOL Studies; 2024; 6,
Teng, M. F.; Ma, M. Assessing metacognition-based student feedback literacy for academic writing. Assessing Writing; 2024; 59, 100811. [DOI: https://dx.doi.org/10.1016/j.asw.2024.100811]
Topping, K. J. Peer assessment. Theory into Practice; 2009; 48,
Tran, T. T. T.; Ma, Q. Online peer feedback training based on self-regulated learning in english as a foreign language writing: Perceived usefulness and students’ engagement. Studies in Educational Evaluation; 2024; 83, 101418. [DOI: https://dx.doi.org/10.1016/j.stueduc.2024.101418]
Utami, S. P. T.; Andayani, A.; Winarni, R.; Sumarwati, S. Utilization of artificial intelligence technology in an academic writing class: How do Indonesian students perceive?. Contemporary Educational Technology; 2023; 15,
Wale, B. D.; Bogale, Y. N. Using inquiry-based writing instruction to develop students’ academic writing skills. Asian-Pacific Journal of Second and Foreign Language Education; 2021; 6,
Wang, D. Teacher- versus AI-generated (Poe application) corrective feedback and language learners’ writing anxiety, complexity, fluency, and accuracy. International Review of Research in Open and Distance Learning; 2024; 25,
Weissman, J. ChatGPT is a plague upon education (opinion); Inside Higher Ed: 2023; Available online: https://www.insidehighered.com/views/2023/02/09/chatgpt-plague-upon-education-opinion (accessed on 10 October 2024).
Woo, D. J.; Susanto, H.; Yeung, C. H.; Guo, K.; Fung, A. K. Y. Exploring AI-Generated text in student writing: How does AI help?. Language Learning & Technology; 2024; 28,
Woodworth, J.; Barkaoui, K. Perspectives on using automated writing evaluation systems to provide written corrective feedback in the ESL classroom. TESL Canada Journal; 2020; 37,
Yan, L.; Sha, L.; Zhao, L.; Li, Y.; Martinez-Maldonado, R.; Chen, G.; Li, X.; Jin, Y.; Gašević, D. Practical and ethical challenges of large language models in education: A systematic scoping review. British Journal of Educational Technology; 2024; 55,
Yang, H.; Gao, C.; Shen, H. Learner interaction with, and response to, AI-programmed automated writing evaluation feedback in EFL writing: An exploratory study. Education and Information Technologies; 2024; 29,
Yoon, S. Y.; Miszoglad, E.; Pierce, L. R. Evaluation of ChatGPT feedback on ELL writers’ coherence and cohesion. arXiv; 2023; arXiv: 2310.06505
Yu, S.; Zheng, Y.; Jiang, L.; Liu, C.; Xu, Y. “I even feel annoyed and angry”: Teacher emotional experiences in giving feedback on student writing. Assessing Writing; 2021; 48, 100528. [DOI: https://dx.doi.org/10.1016/j.asw.2021.100528]
Zainurrahman, Z.; Rodliyah, R. S. Examining Bing AI as a solution to EFL writing feedback challenges. Professional Journal of English Education; 2024; 7,
Zhang, Z. V. Engaging with automated writing evaluation (AWE) feedback on L2 writing: Student perceptions and revisions. Assessing Writing; 2020; 43, 100439. [DOI: https://dx.doi.org/10.1016/j.asw.2019.100439]
Zong, Z.; Schunn, C. D.; Wang, Y. Do experiences of interactional inequality predict lower depth of future student participation in peer review?. Computers in Human Behavior; 2022; 127, 107056. [DOI: https://dx.doi.org/10.1016/j.chb.2021.107056]
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
© 2025 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Abstract
The use of different generative AIs such as OpenAI’s ChatGPT, Microsoft Copilot, or Google’s Gemini has been implemented and studied in different aspects of language education. However, exploring how the combination of teacher-generated feedback and AI-generated feedback influences student revision practices in EFL academic writing remains largely unexplored. To fill in the gap, this preliminary study investigates the impact of two forms of feedback, including teacher-generated feedback and AI-generated feedback, as well as the orders in which the two types of feedback have been executed, that is, teacher-generated feedback before AI-generated feedback (TGF-AIGF) or AI-generated feedback before teacher-generated feedback (AIGF-TGF), on EFL students’ writing revision practices in a 15-week course with fourteen Vietnamese undergraduates. Using Gemini as an AI-generated feedback tool, the study analyzed student revisions in four essays, focusing on local (grammar and vocabulary) and global (content and organization) aspects. Findings revealed that AI-generated feedback consistently resulted in higher revision frequencies compared to teacher-generated feedback alone, as it provided specific, actionable, and comprehensive suggestions. The integration of teacher- and AI-generated feedback yielded the highest revision frequencies, demonstrating complementary strengths, including AI-generated feedback that addressed surface-level issues, while teacher-generated feedback focused on higher-order concerns. Although no statistically significant differences were found between the two orders in which the two types of feedback have been executed, the AIGF-TGF order showed a slightly greater quantity of revisions made by students, allowing AI-generated feedback to scaffold surface-level revisions before teacher-generated feedback addressed global issues. These results highlight the potential of combining AI- and teacher-generated feedback to enhance writing revisions and provide pedagogical insights for integrating AI tools into academic writing courses.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
Details
