Enhancing EFL Writing Revision Practices: The Impact of AI- and Teacher-Generated Feedback and Their Sequences
Open accessThe DOI matches its Crossref record; 66 of the 66 references it cites were found in a registry or an index; 1 of the 45 sources its claims cite were checked for support.
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.
Overview
CoChat Summary
TL;DR
This preliminary study examined how AI-generated (Gemini) and teacher-generated feedback, and their two sequences, influenced revision practices among 14 Vietnamese EFL undergraduates over a 15-week writing course. AI-generated feedback and combined AI+teacher feedback produced more revisions than teacher-generated feedback alone, while order differences were not statistically significant though AIGF-TGF showed slightly more revisions.
Key findings
- AI-generated feedback yielded a mean of 1.74 revisions per 100 words (SD = 0.20) versus 1.04 (SD = 0.16) for teacher-generated feedback alone; combined teacher- and AI-generated feedback was highest at 1.98 (SD = 0.29), and self-correction was lowest at 0.98 (SD = 0.22). One-way ANOVA showed a significant effect, F(3,55) = 68.42, p < 0.05, with post hoc tests indicating combined and AI-alone feedback differed significantly from teacher-alone and self-correction.
- Local revisions (M = 2.97, SD = 0.33) slightly exceeded global revisions (M = 2.78, SD = 0.48), but the difference was not significant, t(26) = 1.25, p > 0.05.
- No statistically significant differences were found between the TGF-AIGF and AIGF-TGF orders (p > 0.05), though AIGF-TGF had slightly higher means: teacher-generated revisions 1.16 vs. 0.94, AI-generated 1.82 vs. 1.68, combined 2.02 vs. 1.91, local 3.06 vs. 2.89, and global 2.92 vs. 2.64.
- Revision frequency correlated significantly with essay scores in three of four essays: Essay 2 r = 0.63, p = 0.01; Essay 3 r = 0.77, p < 0.01; Essay 4 r = 0.55, p = 0.04; Essay 1 was not significant, r = 0.35, p = 0.21. Stage patterns differed by order: in TGF-AIGF, global revisions decreased from 1.72 to 0.92 and local increased from 1.26 to 1.63; in AIGF-TGF, local decreased from 1.75 to 1.31 and global increased from 1.21 to 1.71.
Methods
- Fourteen second-year Vietnamese English majors with low-to-intermediate English proficiency completed four essays (two opinion, two cause-and-effect, approximately 250 words each) in a 15-week essay writing course; Gemini provided AI-generated feedback, and feedback order was TGF-AIGF for Essays 1 and 3 and AIGF-TGF for Essays 2 and 4.
- Revisions were tracked using Microsoft Word’s Compare function and coded by origin (teacher-generated, AI-generated, both, self-correction) and area (local vs. global) using an adapted Hall (1990) framework; two raters coded 15% of essays with Cohen’s kappa = 0.87, and revisions were standardized per 100 words.
- Descriptive and inferential statistics in SPSS v25 included paired-samples t-tests, one-way ANOVA, and Pearson correlations; essay scores from 0 to 100 used Shehadeh’s (2011) rubric with two trained raters.
Limitations
- The paper states that the small sample size limits generalizability; the analysis focused on revision frequency and provides only indirect evidence linking revision frequency to improved essay scores, leaving revision quality unexplored; and the quantitative approach should be supplemented by mixed-method research.
- Evident from the text: the study is preliminary/exploratory with 14 participants and no control group, feedback order is confounded with essay type and time (Essays 1 and 3 vs. 2 and 4), and AI feedback is limited to Gemini, so findings may not generalize to other tools or contexts.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
1 Valid · 6 Not valid · 29 Not checked · 85 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“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.”
- Also stated in: § 4.2. RQ2: To What Extent Do the Two Orders (TGF-AIGF and AIGF-TGF) Impact Student Revisions? · plain
“students in the AIGF-TGF order made slightly more revisions across all categories compared to the TGF-AIGF order.”
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on 3 assumptions not shown here.
- Also stated in: § 4.2. RQ2: To What Extent Do the Two Orders (TGF-AIGF and AIGF-TGF) Impact Student Revisions? · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“the AIGF-TGF order showed a slightly greater quantity of revisions made by students”
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“as it provided specific, actionable, and comprehensive suggestions.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“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.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“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”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
56 more claims
Create a free account to see all 121 claims and to verify the 44 references still open.
- Probably not a substantial claim. 35
- It may state two separate claims as one. 37
- The quote only partly states this claim. 1
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“while Tran [22] demonstrated that AI-generated feedback, when sequenced with teacher input, effectively supported revisions in argumentative writing.”
From Evaluating the impact of using ChatGPT on EFL students’ argumentative skills (2026) · cites this paper as Tran [22]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Few studies have looked at the long-term effects of ChatGPT and similar tools on the development of writing skills, particularly in actual classroom settings.”
From Evaluating the impact of using ChatGPT on EFL students’ argumentative skills (2026) · cites this paper as Tran [22]
Experimental — accuracy not yet measured
Statistics recomputed
Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.
- Consistentt(13) = -3.64reported p = .01 · recomputed p = .003
“t (13) = -3.64, p = .01”
The paper prints p at a floor (such as p < .001)
- Consistentt(13) = -5.2reported p = .01 · recomputed p < .001
“t (13) = -5.20, p = .01”
The paper prints p at a floor (such as p < .001)
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3.4.2. Students' Essay Revision Analysis
“revisions from Essays 1 and 3 for the teacher-generated feedback followed by AI-generated feedback (AIGF-TGF) order”
§ 3.3.2. Session 2-Data Collection Stage-Weeks 4-7 and Weeks 9-13
“For the first and third writing tasks, the students followed the TGF-AIGF order, and for the second and fourth the AIGF-TGF order.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.2. RQ2: To What Extent Do the Two Orders (TGF-AIGF and AIGF-TGF) Impact Student Revisions?
“| | Local revisions | 1.75 | 0.30 | 1.31 | 0.21 | 5.92 | 0.01 |”
§ 4.2. RQ2: To What Extent Do the Two Orders (TGF-AIGF and AIGF-TGF) Impact Student Revisions?
“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.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.1. RQ1: To What Extent Do the Two Forms (AIGF and TGF) Impact Student Revision?
“| Essay 4 | 6.21 | 0.91 | 86.71 | 4.16 | 0.55 | 0.04 ** |”
§ 4.1. RQ1: To What Extent Do the Two Forms (AIGF and TGF) Impact Student Revision?
“Notes: * 0.01 < p < 0.05, ** p < 0.01, moderate correlation ( r = 0.40-0.69); strong correlation ( r = 0.70-0.89).”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.2. RQ2: To What Extent Do the Two Orders (TGF-AIGF and AIGF-TGF) Impact Student Revisions?
“| AIGF-TGF | Global revisions | 1.21 | 0.33 | 1.71 | 0.45 | - 5.20 | 0.00 |”
§ 4.2. RQ2: To What Extent Do the Two Orders (TGF-AIGF and AIGF-TGF) Impact Student Revisions?
“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).”
Citations
Cites
Hattie, J., & Timperley, H. (2007) · Review of Educational Research · doi:10.3102/003465430298487 · Cited by 12,905 · cited 1×
“Feedback is broadly defined as information offered by an agent about specific aspects of an individual’s performance or understanding (Hattie & Timperley, 2007).”
ChatGPT for Good? On Opportunities and Challenges of Large Language Models for Education
Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., et al. (2023) · doi:10.35542/osf.io/5er8f · Cited by 7,552
ChatGPT for good? On opportunities and challenges of large language models for education
Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., et al. (2023) · Learning and Individual Differences · doi:10.1016/j.lindif.2023.102274 · Cited by 7,326
Kluger, A. N., & DeNisi, A. (1996) · Psychological Bulletin · doi:10.1037/0033-2909.119.2.254 · Cited by 5,882 · cited 1×
“Similarly, feedback provision refers to actions undertaken by external agents to supply information concerning certain facets of task performance (Kluger et al., 1996).”
ChatGPT: Bullshit spewer or the end of traditional assessments in higher education?
(2023) · Journal of Applied Learning & Teaching · doi:10.37074/jalt.2023.6.1.9 · Cited by 1,806 · cited 2×
“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).”
Practical and ethical challenges of large language models in education: A systematic scoping review
Yan, L., Sha, L., Zhao, L., Li, Y., Martinez‐Maldonado, R., Chen, G., Li, X., Jin, Y., & Gašević, D. (2024) · British Journal of Educational Technology · doi:10.1111/bjet.13370 · Cited by 1,435 · cited 2×
“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).”
ChatGPT for Language Teaching and Learning
Kohnke, L., Moorhouse, B. L., & Zou, D. (2023) · RELC Journal · doi:10.1177/00336882231162868 · Cited by 1,371 · cited 1×
“Similarly, Kohnke et al. (2023) explored the challenges and opportunities of ethically integrating ChatGPT into language learning.”
Feedback on second language students' writing
Hyland, K., & Hyland, F. (2006) · Language Teaching · doi:10.1017/s0261444806003399 · Cited by 1,016 · cited 3×
“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 (Hyland & Hyland, 2006).”
InfluentialError feedback in L2 writing classes
Ferris, D., & Roberts, B. (2001) · Journal of Second Language Writing · doi:10.1016/s1060-3743(01)00039-x · Cited by 981 · cited 1×
“…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).”
Topping, K. J. (2009) · Theory Into Practice · doi:10.1080/00405840802577569 · Cited by 785 · cited 1×
“This uneven distribution is attributed to peers’ limited expertise and confidence in providing critical, higher-order feedback (Topping, 2009; Zong et al., 2022).”
Research works printed but not linked: 7
- Abdullah, 2020, Between preaching and practicing—The incongruities of teachers’ written feedback beliefs and practices: A case study, Journal of Educational Research and Indigenous Studies, 1, 1
- Faigley, 1981, Analyzing review, College Composition and Communication, 32, 400, 10.58680/ccc198115887 · doi:10.58680/ccc198115887 · Paper page
- Kim, 2024, 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, 40, 19
- Reid, J. M. (1993). Teaching ESL writing, Prentice Hall Regents.
- Rowland, 2023, 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, 17, T31
- Yoon, S. Y., Miszoglad, E., and Pierce, L. R. (2023). Evaluation of ChatGPT feedback on ELL writers’ coherence and cohesion. arXiv.
- Zainurrahman, 2024, Examining Bing AI as a solution to EFL writing feedback challenges, Professional Journal of English Education, 7, 459
Cited by
Wiboolyasarin, W., Wiboolyasarin, K., Tiranant, P., Jinowat, N., & Boonyakitanont, P. (2025) · Ampersand · doi:10.1016/j.amper.2025.100224 · Cited by 206
Generative AI and L2 Written Feedback Studies: A Scoping Review
Crosthwaite, P., & Sun, S. (2026) · RELC Journal · doi:10.1177/00336882251386530 · Cited by 50
“Selective uptake was noted when feedback was excessive or unclear, and uptake varied by pro fi ciency or technological competence (e.g., Tran, 2025; Yan and Zhang, 2024).”
InfluentialZou, B., Wang, C., He, H., Li, C., Purwanto, E., & Wang, P. (2025) · Learning and Motivation · doi:10.1016/j.lmot.2025.102158 · Cited by 46
Moradi, H. (2025) · International Journal of Educational Technology in Higher Education · doi:10.1186/s41239-025-00530-4 · Cited by 82
“For beginners and lower-intermediate learners, educators should implement scaffolded approaches where ChatGPT interactions are initially structured with predefined prompts and guided tasks (Tran, 2025).”
Cui, Y. (2025) · Computers and Education: Artificial Intelligence · doi:10.1016/j.caeai.2025.100391 · Cited by 26
Metadata
- Authors
- Tran, T. T. T.
- Affiliations
- Hue University (VN)
- Corresponding author
- Tran, Thi Thanh Thao
- ORCID
- Tran, Thi Thanh Thao
- Venue
- Education Sciences
- Publisher
- MDPI AG
- Published
- 2025-02-13
- Volume
- 15
- Issue
- 2
- Pages
- 232
- ISSN
- 2227-7102 (ISSN-L)
- OpenAlex
- W4407527232
- Cited by
- 68
- References
- 63
- Citations by year
- 2026: 51 · 2025: 17
- FWCI
- 240.95
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Education Sciences (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- Đại học Huế
- Topics
- Social Sciences › Arts and Humanities › Language and Linguistics › EFL/ESL Teaching and Learning
- Social Sciences › Education › Student Assessment and Feedback
- Social Sciences › Education › English Language Learning and Teaching
- Keywords
- EFL writing, writing revision, AI-generated feedback, Gemini, ChatGPT, academic writing
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.