Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The 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.

Enhancing EFL Writing Revision Practices: The Impact of AI- and Teacher-Generated Feedback and Their Sequences

Open access
Tran, Thi Thanh Thao · 2025 · Education Sciences

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

Crossref · 10.3390/educsci15020232

Cites

66 of 66 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

1 Valid · 6 Not valid · 29 Not checked · 85 Own results and interpretations

56 more claims

Create a free account to see all 121 claims and to verify the 44 references still open.

See all claims — free account
  • Probably not a substantial claim. 35
  • It may state two separate claims as one. 37
  • The quote only partly states this claim. 1

Create a free account to see these claims.

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?

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)

2 of 2

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.

4 of 4

Citations

Cites

Show all 63 references

Research works printed but not linked: 7

  1. 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
  2. Faigley, 1981, Analyzing review, College Composition and Communication, 32, 400, 10.58680/ccc198115887 · doi:10.58680/ccc198115887 · Paper page
  3. 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
  4. Reid, J. M. (1993). Teaching ESL writing, Prentice Hall Regents.
  5. 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
  6. Yoon, S. Y., Miszoglad, E., and Pierce, L. R. (2023). Evaluation of ChatGPT feedback on ELL writers’ coherence and cohesion. arXiv.
  7. Zainurrahman, 2024, Examining Bing AI as a solution to EFL writing feedback challenges, Professional Journal of English Education, 7, 459

Cited by

See all 68

Metadata

Authors
Tran, T. T. T.
Affiliations
Hue University (VN)
Corresponding author
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
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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Enhancing EFL Writing Revision Practices: The Impact of AI- and Teacher-Generated Feedback and Their Sequences. https://app.cochat.ai/papers/10.3390/educsci15020232
The paper
Tran, T. T. T. (2025). Enhancing EFL Writing Revision Practices: The Impact of AI- and Teacher-Generated Feedback and Their Sequences. Education Sciences. https://doi.org/10.3390/educsci15020232

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.