Technology in Language Teaching & Learning

ISSN 2652-1687
https://www.castledown.com/journals/tltl/

Technology in Language Teaching & Learning, 6(3), 1–23 (2024)

https://doi.org/10.29140/tltl.v6n3.1482

Comparing Peer, ChatGPT, and Teacher

Corrective Feedback in EFL Writing: Students’
Perceptions and Preferences
ORIT ZEEVY SOLOVEY

Tel Aviv University, Israel

[email protected]

Abstract

The effectiveness of providing Written Corrective Feedback (WCF) in English as a Foreign Language
(EFL) classrooms to enhance students’ general writing skills has been well-established. With the
advent of various feedback modes in modern classrooms, this pilot study aimed to explore EFL
students’ perceptions and preferences regarding the effectiveness of three distinct feedback modes:
peer feedback, artificial intelligence (AI), and teacher feedback. The study also aimed to determine
the distribution of feedback across specific writing components, while accounting for revisions made.
Fifteen pairs of participants completed a short writing task and received WCF from peers, followed
by an AI tool (ChatGPT), and finally from their teacher. Subsequent revisions to their writing were
made after each feedback mode. Using a qualitative approach via survey analysis, findings indicate
that participants perceived both ChatGPT and teacher feedback as effective, with peer feedback also
being generally regarded as effective by most participants. Preferences lean towards teacher WCF and
a combination of teacher and ChatGPT WCF. Additionally, the study identified feedback distribution
across writing components, with implications discussed for refining EFL classroom feedback practices.

Keywords: Written corrective feedback (WCF); English as a foreign language (EFL); peer feedback;
artificial intelligence (AI) feedback

Introduction

The provision of Written Corrective Feedback (WCF) in English as a Foreign Language (EFL)
classrooms in schools and universities has been proven effective in improving students’ general

Copyright: © 2024 Orit Zeevy Solovey. This is an open access article distributed under the terms of the ­Creative
­Commons Attribution Non-Commercial 4.0 International License, which permits unrestricted use, d­ istribution, and
reproduction in any medium, provided the original author and source are credited. Data Availability Statement:
All relevant data are within this paper.
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 2

writing skills. In the modern classroom, various modes of feedback are now available. Teacher
feedback has long been the most traditional and commonly used method of providing WCF. It
involves the teacher reviewing students’ written work, identifying errors, and providing comments
and corrections (Ferris & Roberts, 2001). Peer feedback emerged at a later stage in the field of
WCF and involves students providing feedback on their peer’s writing, promoting a collaborative
learning environment (Storch, 2005). The most recent development in WCF is artificial intelligence
(AI) feedback. One AI tool is ChatGPT (Chat Generative Pre-trained Transformer) which has been
very popular since its launch in November 2022. This tool interacts with users conversationally
by responding to questions and requests, and offers useful features related to writing assessment,
including the provision of timely feedback.

The increasing availability of feedback modes in the EFL context, along with constant technological
advancements, prompt teachers and researchers to examine students’ views, appreciation, and
preferences regarding these modes. Some students may prefer the collaborative setting of peer
feedback, others may appreciate the teacher’s expertise, while some may opt for an AI instant and
objective evaluation. Understanding students’ views and preferences is important in promoting
engagement and motivation, and in laying the groundwork for potentially incorporating student
choice in future writing tasks, through adopting a more student-centered approach that caters to
individual views and preferences.

To the best of my knowledge, there are no studies that have explored EFL students’ perceptions of
peer, ChatGPT, and teacher-WCF in combination. Hence, this study seeks to fill this gap and provide
insights into how EFL students perceive these feedback modes. This pilot study aims to gauge students’
perceptions of the effectiveness of three different feedback modes (peer, ChatGPT, and teacher) and
determine which feedback they prefer to receive. An additional aim is to identify the percentage of
feedback given to specific components of the writing while also taking into account the students’
revisions, in order to gain more insight into the feedback processes.

Literature Review

Collaborative writing, with its inherent practice of immediate feedback exchange among students,
substantially enhances writing skills. As conceptualized by Hattie and Timperley (2007), written
corrective feedback plays a major role in this process by providing awareness of various aspects of
writing performance. Storch’s (2005) study comparing texts written by pairs and individual learners
highlights the effectiveness of collaborative writing in achieving better task outcomes, grammatical
precision, and complexity. Additionally, collaborative writing fosters collective scaffolding, as termed
by Donato (1988, 1994), wherein students collectively pool their language knowledge. Empirical
evidence suggests that collaborative writing not only promotes consideration of form but also aids
in L2 vocabulary acquisition and overall writing proficiency (Garcia Mayo & Imaz Agirre, 2019;
Villarreal & Gil-Sarratea, 2019; Fernandez Dobao, 2014; Kim, 2008; Shehadeh, 2011).

In contemporary educational settings, various written corrective feedback methods are available in
EFL classrooms: peer, AI, and teacher. The most traditional and commonly used is teacher feedback,
which involves the teacher reviewing students’ written work, identifying errors, and providing comments
and corrections (Ferris & Roberts, 2001). Teacher-led feedback is critical in the revision process for
student writers (Sheen et al., 2009). According to Bitchener and Ferris (2012), teachers’ WCF, such
as edge comments, requests for clarification, and comments on grammatical issues, has been proven
to help EFL writers. Teacher feedback is valued for offering expert guidance and instructions tailored
to meet individual student needs (Lyster, 1998). It has been suggested that effective teacher feedback
3 Technology in Language Teaching & Learning, 6(3)

significantly improves students’ writing skills, increases motivation, and fosters a positive learning
experience (Ferris & Roberts, 2001).

Peer feedback, a more recent advancement in the field, involves students offering feedback on
each other’s writing, fostering a collaborative learning environment. It has arisen as an innovative
method that differs from traditional teachers’ WCF (Sheen, 2010b) since teachers traditionally
play a dominant role in providing feedback and comments during the writing process (Sheen,
2010a). Peer feedback has been increasingly examined in the literature (Hewett, 2000; Liu &
Hansen, 2002) and has demonstrated numerous benefits. Several studies (Connor & Asenavage,
1994; Cho & MacArthur, 2010; Storch, 2005; Topping, 1998) have concluded that peer feedback
supports the enhancement of learners’ writing capabilities and competencies concerning cognitive,
affective, social, and linguistic perspectives. Some researchers (for example, Sato [2013], Sato &
Lyster [2012], and Sippel & Jackson [2015]) claim that students’ engagement with peer feedback
can promote second language learning and can change their role from passive learners to active
participants. Peer interactions allow students to engage in active analysis of others’ writing, which can
lead to a deeper understanding of their own writing weaknesses (Storch, 2005). Peterson and Portier
(2014) maintain that peer feedback benefits not only the students, but also the feedback providers.
It also helps students to develop critical thinking and language analysis skills (Ferris & Roberts,
2001). As students gain experience using the given criteria to evaluate writing, their self-assessment
abilities progress. (Lundstrom & Baker, 2009). Therefore, peer feedback is considered an effective
method to improve students’ writing skills (Xiao & Lucking, 2008).

Nevertheless, some researchers (for example, Adams et al [2011], Philp et al [2010], and Zhang [1995])
doubt the pedagogic potential of peer feedback due to factors such as the learners’ distrust of their own
or their peers’ linguistic knowledge. Ruegg (2015) found that teacher feedback was more effective
in improving grammatical errors than peer feedback. Some studies that compared teacher and peer
feedback have shown that student writers prefer teachers’ comments over those of their peers, while
other studies suggest that peer feedback is beneficial for the construction of the writing process (Yang
et al., 2006). It is suggested that EFL learners can effectively take advantage of both methods to improve
their writing skills, Tai et al., (2015) compared the effects of the combination of teacher and peer
feedback versus using teacher feedback alone on the writing performance of EFL university students
within a collaborative online learning system. The results revealed that, in terms of holistic writing
skills and the subscales of content, organization, grammar, mechanics, and style, the students in the
combined feedback group demonstrated greater improvements than those who received only teacher
feedback. Yet, due to the students’ limited linguistic knowledge and writing skills, they were able to
indicate errors from a reader’s perspective, but were less able to supply consistent comprehensive and
persuasive feedback. Overall, the students viewed peer review as a demanding task, albeit helpful.
Lee (2008) suggests that it is essential to provide proper guidance to peers to ensure accurate error
identification and constructive feedback.

The most recent development in the field of WCF is AI feedback. Artificial Intelligence has witnessed
significant progress in recent years, and the use of AI-based technology is becoming increasingly
popular in the field of education. There are multiple ways of applying AI-based tools in student
learning environments. Steiss et al., (2024) maintain that AI may function as an automated writing
evaluation tool, increasing the amount of feedback students receive and diminishing the burden on
teachers to provide frequent feedback to large classes. According to Wu (2024), AI technology has
made significant strides in vocabulary acquisition, grammar correction, and discourse generation. AI
tools can offer assistance both during and after the writing process. AI technology not only examines
grammar and spelling but also offers extensive support in identifying writing problems and suggesting
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 4

improvements (Alharbi, 2023). One AI tool is ChatGPT (Chat Generative Pre-trained Transformer),
a chatbot that has been very popular since its launch in November 2022 by the United States-based
startup OpenAI. ChatGPT interacts with users conversationally, responding to questions and requests
in a dialogic manner (OpenAI, 2023). This tool, powered by a large language model, can understand
and produce human-like responses across a wide range of topics by computing techniques and a large
amount of information to connect ideas and understand prompts in context (Barrot, 2023). Wenzlaff
and Spaeth (2022) affirm that ChatGPT is essentially equivalent to humans in composing explanatory
answers.

ChatGPT can be a helpful writing tool in the second language classroom by addressing some of the
writers’ needs and offering useful features related to writing assessment, including the provision
of timely feedback. By prompting it to give feedback, ChatGPT identifies the strengths and
weaknesses of a text and offers recommendations for improvement. It can also evaluate the clarity,
focus, and structure of a text (Barrot, 2023). Several studies have examined the effectiveness of
employing ChatGPT as a feedback tool, resulting in diverse findings (Steiss et al., 2024; Wu, 2024).
Specifically, a study conducted by Dai et al. (2023) investigated the feasibility of utilizing ChatGPT
to offer students constructive feedback aimed at enhancing their learning experience. The findings
demonstrated that ChatGPT consistently produces more readable feedback, aiding students in better
comprehension and subsequent improvement efforts. However, it was noted that ChatGPT lacks the
reliability seen in instructor assessments of student performance. Furthermore, ChatGPT exhibited
a notable capability to generate process-focused feedback, which was deemed more effective in
shaping students’ task strategies than task-focused feedback. The overall conclusion underscores
ChatGPT’s significant potential in guiding students to enhance their skills and foster the develop of
learning skills.

Additionally, Su et al. (2023) explored the potential applications of ChatGPT in supporting students
across various tasks, including outline preparation, content revision, and proofreading. By providing
ChatGPT with an argumentative writing outline and an evaluation rubric, the researchers observed
its ability to assess the coherence between claims and sub-claims, evaluate the quality of supporting
evidence and rebuttals, and suggest potential counterarguments and claims. This approach has the
potential to enhance the structural aspect of the argumentation. In the editing phase, ChatGPT received
evaluation rubrics for argumentative writing content. The feedback generated accurately pinpointed the
strengths and weaknesses of the writing, offering valuable suggestions for improvement. The researchers
acknowledged, however, that due to ChatGPT’s limited generative capacity, some suggestions were
somewhat vague, with abstract comments, necessitating further elaboration. During the proofreading
stage, an evaluation checklist addressing the use of lexical devices, syntactic structures, and textual
devices was provided to ChatGPT. While the feedback primarily took an evaluative stance, it lacked
sufficient elaboration, limiting its potential for facilitating learning. Furthermore, the consistency of
ChatGPT’s feedback was noted to be an issue, as it yielded different responses upon each submission
of student writing, diminishing the feedback’s overall usefulness. Despite these limitations, the
researchers highlighted ChatGPT’s efficacy as a proofreading aid. When employed to assess individual
sentences for grammar, ChatGPT demonstrated the ability to evaluate grammatical accuracy, elucidate
the intended meaning of the sentence, and suggest alternative expressions. However, the researchers
state that the responsibility rests with students to critically analyze and determine whether to accept the
revisions proposed by ChatGPT.

Furthermore, Yoon et al., (2023) evaluated the quality of the feedback produced by ChatGPT
as to the coherence and cohesion of essays written by English language learners. They used 50
argumentative essays to be assessed utilizing a rubric. A two-step approach was employed in the
5 Technology in Language Teaching & Learning, 6(3)

feedback evaluation. First, each sentence in the feedback was categorized into subtypes based on its
function (for example, positive reinforcement and problem statement). Then, each sentence in the
feedback was evaluated for accuracy and usability in the feedback according to these types. Both the
analysis of feedback types and the evaluation of accuracy and usability revealed that most feedback
sentences were very abstract, generic, and did not provide concrete suggestions for improvement. The
accuracy in detecting repetitive ideas and erroneous use of cohesive devices depended on superficial
linguistic features and was often incorrect. They concluded that ChatGPT, without specific training
for the feedback generation task, did not offer effective feedback on coherence and cohesion in these
students’ essays.

Upon receiving feedback of any nature, students stand to benefit from revising their writing. As part
of a process-orientated pedagogy, teachers have encouraged their students to write multiple drafts
of their papers and have explored various feedback methods to help them improve at each stage of
the writing process. (Ferris, 1997). Beason (1993) states that “feedback and revision are valuable
pedagogical tools…the research typically indicates that high school and college students improve their
drafts upon receiving feedback” (p. 396). WCF aims to teach skills that help students improve their
writing proficiency and produce written texts with fewer errors and more clarity (Williams, 2003). The
integration of collaborative writing with multiple feedback modes (peer, ChatGPT, and teacher) may
offer a rich context for understanding students’ perceptions and preferences regarding the different
feedback sources.

Methods

The purpose of this pilot study was to gauge students’ perceptions regarding the effectiveness of the
different feedback modes and to find out which feedback they prefer to receive. An additional aim was
to identify the percentage of feedback given to specific components of the writing while considering
the revisions performed by students, to gain more insight into the feedback processes. This study was
guided by the following three research questions:

RQ 1: Do students perceive peer, ChatGPT, and teacher WCF as effective?

RQ 2: What WCF (peer, ChatGPT, teacher) do students prefer to receive?
RQ 3: What percentage of the WCF is given to specific components of the writing, by
peers, ChatGPT, and the teacher?

A university class of 30 freshmen EFL students participated in the study. The students were enrolled
in a B2 (upper-intermediate) level course. This is the university’s most advanced compulsory English
course geared toward improving English reading, writing, and other academic skills to enable students
to function competently in the academic and professional world. There were 8 males and 22 females
with an age range of 19–26.

The strategy of collaborative writing (Elola & Oskoz, 2010), process approaches, and multiple
revisions (Hyland, 2003) were adopted for this study. Students were randomly divided into 15 pairs
and were assigned a writing task during which they had to use a checklist to verify that all the writing
requirements had been met. Although the assignment was not graded, “The Composition Grading
Feedback Sheet” developed by Brown and Bailey (1984) was used in this study. It was adapted to
serve as a checklist, as illustrated in Table 8, which consisted of five criteria: organization, content,
grammar, mechanics, and quality of expression (see Appendix A). The checklist items align with the
writing instruction provided throughout the course, and participants used this same checklist to confirm
adherence to writing requirements on two prior occasions preceding the study.
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 6

The study’s sequencing of feedback modes (peer-ChatGPT-teacher) was strategically devised to

optimize feedback effectiveness, considering practical factors such as time constraints, varying
familiarity, and the promptness of each mode. The rationale behind this order rests on several key
considerations. Firstly, recognizing the inherent time limitations of teacher feedback compared to
peer and ChatGPT feedback, the decision was made to schedule teacher feedback last. This ensured
students received timely feedback from their peers and ChatGPT before the more time-consuming
teacher feedback process. Secondly, the sequence considered students’ familiarity with receiving peer
feedback versus feedback from an AI tool. By initiating with peer feedback, followed by ChatGPT
feedback, the aim was to gradually introduce students to the less familiar mode of automated feedback
while leveraging their existing comfort with peer evaluation. Thirdly, prioritizing peer and ChatGPT
feedback in the initial rounds allowed for faster turnaround times, enabling students to promptly receive
feedback on their written work. This highlights the importance of timely feedback in supporting student
learning and improvement. Lastly, while acknowledging the potential for order effects introduced
by the sequencing of feedback modes, consistency in evaluation criteria across all feedback modes
was ensured. Participants were also briefed on the rationale behind the chosen feedback sequence to
maintain transparency and understanding.

During the initial round of feedback, each pair was asked to use the checklist to review another pair’s
writing and provide feedback on areas that required correction. On the checklist, they indicated
whether each writing criterion was fulfilled, provided comments on areas requiring correction, and
proposed alternative options. Subsequently, each pair had to decide whether they agreed with their
peers’ comments and make necessary revisions to their work. Next, each pair had to input their revised
work into ChatGPT, which was prompted by the checklist, and was instructed to provide feedback as
to what needs to be altered. Students received feedback from ChatGPT on the fulfillment of the writing
criteria and copied ChatGPT’s comments and suggestions into the checklist. Students then assessed
this feedback, decided whether to accept it, and proceeded to modify their writing accordingly. Finally,
each pair submitted their modified work to the teacher who used the same checklist to identify areas for
correction and provide suggestions for improvement. Students had to determine whether they agreed
with the teacher’s feedback and revise their work. Following these feedback cycles, participants
completed a feedback survey (see Appendix B).

The study employed a qualitative approach to explore students’ perceptions and preferences regarding
different feedback modes and to identify the distribution of feedback across specific components of
the writing. To address these objectives, data was gathered using two methods. First, a checklist was
employed after each feedback round, necessitating an assessment of writing criteria fulfillment as
indicated by peers, ChatGPT, and the teacher. Participants were also directed to copy ChatGPT’s
comments and suggestions into the checklist. Second, a feedback survey was administered upon the
completion of the entire feedback process. The survey addressed relevant aspects of the research
objectives, incorporating feedback from multiple sources (peers, ChatGPT, and the teacher). This
diversity enhances credibility by providing different perspectives on the effectiveness of various
feedback modes, thereby strengthening the findings. Participants evaluated the effectiveness of each
mode, confirming the accuracy of their responses and further enhancing credibility. Reflexivity
is demonstrated in the rationale behind the feedback modes sequencing, acknowledging potential
biases, and study decisions. Additionally, the uniform application of the same checklist by peers,
ChatGPT, and the teacher throughout the study reinforces dependability by ensuring transparency
and consistency. The detailed explanation of the feedback sequence and the participants’ briefing
promotes transparency, enabling readers to assess the reliability of the findings. The checklists along
with the survey files were uploaded to the course Moodle site by the participants. The researcher later
retrieved these files from Moodle and employed content analysis to organize and analyze the data.
7 Technology in Language Teaching & Learning, 6(3)

Results

The qualitative approach used in this study involved a systematic content analysis of survey data to
explore students’ perceptions and preferences regarding different modes of written corrective feedback
(WCF) – peer, ChatGPT, and teacher feedback. The survey data were divided into three sets, each
corresponding to one of the research questions. Five primary codes (A–E) were assigned to organize
and analyze the data: A. Types of corrections required, B. Understanding of comments, C. Agreement
with comments, D. Perceived helpfulness of comments and reasons for this perception, E. Feedback
preference and reasons for this preference.

Categories B–D (Understanding, Agreement, Perceived Helpfulness) addressed the first research
question concerning the effectiveness of peer, ChatGPT, and teacher feedback. Understanding the
feedback was considered crucial for its effective application. Agreement with feedback reflected
students’ judgments regarding its validity and relevance, while perceived helpfulness indicated the
utility of the provided corrections. Category E (Feedback Preference) was related to the second
research question, investigating students’ preferences regarding the source of feedback (peer,
ChatGPT, or teacher). Category A (Types of Corrections Required) pertained to the third research
question, focusing on the distribution of feedback across specific components of writing. It aimed to
identify the percentage of WCF directed towards different writing components by peers, ChatGPT,
and teachers.

Categories B–D were further subdivided to include responses specific to each feedback mode (peer,
ChatGPT, teacher), enabling a detailed analysis of feedback perceptions across different sources.
Within each category, all responses to specific questions were gathered to identify different aspects
related to feedback. This approach facilitated the identification of prevalent themes or opinions within
the data. The frequency of specific types of responses within each category was counted to quantify the
prevalence of particular themes or opinions.

Peer Feedback

In the study, a total of 15 pairs provided feedback on various aspects of their peers’ work. Respondents
were asked about the specific components they believed required correction, and the responses
were analyzed. As displayed in Table 1, the most commonly identified areas for improvement were
organization and mechanics, with 33.33% of respondents (five pairs) mentioning it, followed by
organization alone, which was mentioned by 26.67% of the participants (four pairs). Content and
mechanics, as well as content alone, were each mentioned by 13.33% of the respondents (two pairs
each). A smaller proportion of participants, 6.67% each, identified organization and grammar, and
organization and quality of expression as areas requiring correction (one pair each). Regarding the
understanding of comments, all 15 pairs (100%) reported understanding them entirely. When asked
whether they agreed with or accepted the comments, the majority, 66.67%, (ten pairs) responded
affirmatively. However, 20% of the respondents (three pairs) expressed disagreement with the
comments, indicating, for instance, that “It seemed incorrect to us.” Two pairs, 13.33% expressed
partial agreement, noting, for example, that “We agreed with comments about punctuation, but did not
agree that there are not enough explanations/exemplifications of the ideas. We think there are enough.”
Concerning the perceived helpfulness of the feedback, 73.33% of the respondents (eleven pairs) found
it helpful, specifying that “It helped us change the focus of the writing”and “It helped to improve the
organization”. However, 26.67% (four pairs) did not find the feedback helpful, since, for instance, their
peers “…made some comments, but then we read our text several more times and found mistakes that
they didn’t find”.
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 8

Table 1 Summary of Students’ Perceptions of Peer Feedback

Questions: Responses:
What did your peers think 33.33% – Organization and mechanics
required correction? 26.67% – Organization
13.33% – Content and mechanics
13.33% – Content
6.67% – Organization and grammar
6.67% – Organization and quality of expression
Did you understand all the 100% – Yes
comments?
Did you agree with/accept the 66.67% – Yes
comments? If not, why? 20% – No (“It seemed incorrect to us.”)
13.33% – Partially (“We agreed with comments about punctuation,
but did not agree that there are not enough explanations/
exemplifications of the ideas. We think there are enough.”)
Did you find the feedback helpful? 73.33% – Yes (“It helped us change the focus of the writing”; “It
Why? helped to improve the organization.”)
26.67% – No (“They made some comments, but then we read our
text several more times and found mistakes that they didn’t find.”)
Note. The study aimed to explore EFL students’ perceptions of the effectiveness of peer, ChatGPT,
and teacher feedback, and uncover students’ preferences. Data analysis results demonstrate that most
participants regarded peer feedback as effective. An additional aim was to identify the percentage of the
feedback given to specific components of the writing. It was found that peers mostly commented on the
organization and mechanics of the writing.

ChatGPT Feedback

As shown in Table 2, ChatGPT provided feedback to 15 pairs suggesting various aspects that required
correction. Specifically, 20% of the participants (three pairs) received suggestions related to all
categories (organization, content, grammar, mechanics, and quality of expression.) Also, 20% of the
participants received feedback related to the content, quality of expression, and grammar. Another
20% of participants were given feedback regarding mechanics and grammar. Additionally, 13.33% of
participants (two pairs) received suggestions for organization, mechanics, and grammar, while the same
percentage received feedback on organization, content, and grammar. Lastly, 13.33% of participants
were provided feedback on content, mechanics, and grammar.

The following instances exemplify the participants’ use of ChatGPT and the ensuing feedback. In terms
of the writing’s organization, students inquired whether a thesis statement was present. Responses
varied from affirmations like “Yes, there is a thesis sentence in the provided passage”, to more detailed
suggestions when a thesis statement was perceived to be absent. For instance:

Your paragraph does not explicitly state a thesis statement, but the central idea of the
paragraph is that …. To turn the central idea of the paragraph into a thesis statement,
you could write something like ….

In cases where a thesis statement could not be identified, the response indicated:

Based on the information provided, it is unclear what the thesis sentence of your essay
is. A thesis sentence is a statement that presents the main idea or argument of the essay
and is typically included in the introduction paragraph.
9 Technology in Language Teaching & Learning, 6(3)

Table 2 Summary of Students’ Perceptions of ChatGPT Feedback

Questions: Responses:
What did ChatGPT suggest 20% – All categories
correcting? 20% – Content, quality of expression, and grammar
20% – Mechanics and grammar
13.33% – Organization, mechanics, and grammar
13.33% – Organization, content, and grammar
13.33% – Content, mechanics, and grammar
Did you understand all the 93.33% – Yes
comments? 6.67% – Not entirely (“We didn’t understand why the grammar is
not correct”)
Did you agree with/ accept the 100% – Yes
comments? If not, why?
Did you find the feedback helpful? 100% – Yes (“We saw big differences between our essay
Why? compared to the ChatGPT essay. We could see where we were
wrong and why”; “ChatGPT corrected the essay very fast and
phrased everything better”.)
Note. Data analysis results demonstrate that all participants perceived ChatGPT feedback as effective. It was
also revealed that an equal percentage of ChatGPT’s comments related to all aspects of writing; content,
quality of expression, and grammar; and mechanics and grammar.

Concerning the understanding of the comments, 93.33% (fourteen pairs) indicated that they
comprehended the suggestions made by ChatGPT, yet, 6.67% (one pair) reported not entirely
understanding the comments, “We didn’t understand why the grammar is not correct”. All participants
(100%) reported that they agreed with and accepted the suggestions provided by ChatGPT, and that
they thought the feedback was helpful. Participants claimed, “We saw big differences between our
essay compared to the ChatGPT essay. We could see where we were wrong and why” and “ChatGPT
corrected the essay very fast and phrased everything better”.

Teacher Feedback

As demonstrated in Table 3, analysis of the responses revealed that the most common area for
improvement was mechanics and grammar, as reported by 26.67% of the respondents (four pairs).
Additionally, 20% of the participants (three pairs) were given feedback related to content, mechanics,
and grammar, while another 13.33% (two pairs) reported that the teacher identified organization,
content, and grammar as areas in need of corrections. An equal percentage of participants (13.33%)
were provided improvement suggestions about organization, content, and quality of expression.
Moreover, 13.33% of the participants received feedback on the quality of expression. A smaller
proportion of participants, 6.67% each (one pair), was given feedback related to the quality of
expression, mechanics, and grammar, and some participants (6.67%) were provided feedback across
all categories. All participants (100%) reported that they comprehended and accepted the suggestions
provided by the teachers, and they all perceived the feedback as helpful. They thought, for example,
that “The teacher corrected the mistakes and gave explanations about every mistake, so we know why
it’s a mistake”.

Feedback Preferences

As presented in Table 4, the results indicated that 46.67% of the participants (seven pairs) preferred
receiving feedback solely from the teacher. They noted that “The teacher is the expert”, and that
they “trust teacher feedback the most and can learn from it”. Additionally, 33.33% of the participants
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 10

Table 3 Summary of Students’ Perceptions of Teacher Feedback

Questions: Responses:
What did the teacher think required 26.67% – Mechanics and grammar
correction? 20% – Content, mechanics, and grammar
13.33% – Organization, content, and grammar
13.33% – Organization, content, and quality of expression
13.33% – Quality of expression
6.67% – Quality of expression, mechanics, and grammar
6.67% – All categories
Did you understand all the comments? 100% – Yes
Did you agree with/ accept the comments? 100% – Yes
If not, why?
Did you find the feedback helpful? Why? 100% – Yes (“The teacher corrected the mistakes and gave
explanations about every mistake, so we know why it’s a
mistake”.)
Note. Data analysis results demonstrate that all participants perceived teacher feedback as effective.
Additionally, it was observed that the primary emphasis of teacher feedback was on mechanics and grammar,
with a subsequent focus on content, mechanics, and grammar in that order.

Table 4 Summary of Students’ Feedback Preferences

Questions: Responses:
What feedback or 46.67% – Teacher feedback (“The teacher is the expert.”; “We trust teacher
feedback combination feedback the most and can learn from it.”)
do you prefer to receive? 33.33% – Teacher and ChatGPT feedback (“Because ChatGPT helped us
Why? correct many things in the essay, but then we submitted it to the teacher who
noticed more things we need to correct.”; “The teacher is the most reliable
source we can get feedback from, but if we want a quick answer when the
teacher is not immediately available we can use GPT.”)
6.67% – ChatGPT feedback (“Chat GPT is quick and helpful.”; “GPT
instructions and corrections were the clearest.”; “It also gives positive
feedback.”)
6.67% – Peer and teacher feedback (“We don’t believe in technology, and
prefer to get feedback from peers and teachers. We can ask them questions
about things we don’t understand.”)
6.67% – Peer feedback (“The opinion of peers gives an indication if students
like us understand what we wrote, and they know how to give us feedback”.)
Note. The results indicate a predominant preference among participants for teacher feedback, with a subsequent
inclination towards a blend of teacher and ChatGPT feedback. Conversely, a lesser preference was observed for
peer feedback, a combination of peer and teacher feedback, and feedback solely from ChatGPT.

(five pairs) expressed a preference for receiving feedback from both their teacher and ChatGPT, for
the following reasons, “Because ChatGPT helped us correct many things in the essay, but then we
submitted it to the teacher who noticed more things we need to correct”, and;

The teacher is the most reliable source we can get feedback from, but if we want a quick
answer when the teacher is not immediately available, we can use GPT.

One pair, 6.67%, indicated a preference for receiving feedback only from ChatGPT because it is “…
quick and helpful”, its “…instructions and corrections were the clearest”, and “It also gives positive
11 Technology in Language Teaching & Learning, 6(3)

feedback”. Another pair, 6.67%, preferred a combination of feedback from peers and the teacher. They
indicated, “We don’t believe in technology, and prefer to get feedback from peers and teachers. We can
ask them questions about things we don’t understand”. Finally, one pair, 6.67%, preferred receiving
feedback exclusively from their peers, stating that “The opinion of peers gives an indication if students
like us understand what we wrote, and they know how to give us feedback”.

Discussion

This pilot study sought to gauge students’ perceptions regarding the effectiveness of three different
feedback modes (peer, ChatGPT, and teacher), and find out which feedback they prefer to receive.
It also attempted to identify the percentage of feedback given to specific components of the writing,
while considering the revisions performed, to gain more insight into the feedback processes. The first
research question addressed whether students perceive peer, ChatGPT, and teacher WCF as effective.
According to the findings, all the participants (15 pairs) acknowledged comprehension of their peers’
comments, with the majority (10 pairs), agreeing with and accepting the comments. Yet, 3 pairs
expressed disagreement with the comments, while 2 pairs indicated partial agreement. Most respondents
(11 pairs) found their peer’s feedback to be helpful, though 4 pairs did not. Regarding the feedback
provided by ChatGPT, 14 pairs indicated full comprehension of the comments and suggestions, while
only 1 pair reported partial understanding. All participants noted that they agreed with and accepted the
suggestions provided by ChatGPT, deeming the feedback helpful. In the context of teacher feedback,
all participants affirmed both their understanding of the teacher’s suggestions and their acceptance of
them. They all perceived the teacher’s feedback as helpful.

The second research question explored students’ preference for WCF from peers, ChatGPT, and the
teacher. The findings reveal that the majority (46.67%) favor teacher-provided WCF, and 33.33%
prefer receiving a combination of teacher and ChatGPT WCF. An equal percentage of participants
(6.67% each) indicated a preference for peer feedback; a combination of peer and teacher feedback;
and ChatGPT feedback, respectively. Several observations were noted upon inspecting the results of
the first and second research questions. Firstly, it was observed that while the majority of respondents
(11 pairs) perceived their peers’ feedback as beneficial, only 1 pair expressed a preference for getting
peer feedback alone, and merely 1 pair favored a combination of peer and teacher feedback. This aligns
with Colpitts’ (2016) study that examined student perceptions of receiving teacher corrective feedback,
peer corrective feedback, and giving peer corrective feedback. The results showed a considerable
preference for receiving teacher feedback. Peer feedback has also been challenged by Lin and Yang
(2011), who maintain that the quality of student comments may not be sufficient, and interpersonal
relationships may affect learner intentions and the provision of honest comments. Nevertheless,
the findings diverge from the conclusions drawn in second language writing research (for example,
Hedgcock and Lefkowitz [1992] and Paulus [1999]), which suggest that peer feedback comments have
the potential to lead to substantial revisions. Additionally, these studies propose that, in comparison
to teacher feedback, revisions informed by peer comments may exhibit improvements in vocabulary,
organization, and content.

In the present investigation, students’ reduced inclination toward peer feedback might be attributed to
inadequate training in feedback provision. Students in this study were equipped only with a checklist
for providing feedback. According to Hansen and Liu (2005), who devised adaptable guidelines for
writing instructors in the EFL context, the key to fostering a positive reception of peer feedback lies
in meticulous teacher planning and comprehensive student training. They argue that planning and
training should be continuous processes, spanning the periods before, during, and after peer response,
although the majority should be carried out before peer response commences. Modeling the feedback
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 12

process and gradually developing feedback skills throughout the course are essential steps, ensuring
that peer feedback motivates students to actively engage in their learning (Ferris & Hedgcock, 2005;
Kamimura, 2006).

Secondly, it was noted that while all respondents deemed ChatGPT’s feedback helpful, only 1 pair
indicated a preference for receiving ChatGPT feedback alone, and 5 pairs expressed a preference for
a combination of teacher and ChatGPT feedback. This corresponds with Dai et al.’s (2023) findings
that ChatGPT consistently produces readable feedback, aiding students in better comprehension and
subsequent improvement efforts, but lacks the reliability seen in instructor assessments of student
performance. It could be that students appreciate ChatGPT’s immediate feedback, yet sense its
incapability of “understanding” the text it generates or the context of the information. Scholars have
pointed out that despite their powerful synthesis ability, “chatbots merely learn statistical associations
between words...rather than understand their meanings” (van Dis et al., 2023, p. 615). Nevertheless,
the results of the current study contrast with some of the findings of Su, Lin, and Lai (2023) that
revealed ChatGPT’s ability to assess the coherence between claims and sub-claims, evaluate the
quality of supporting evidence and rebuttals, and suggest potential counterarguments and claims in an
argumentative writing outline.

In the present study, a notable illustration of ChatGPT serving as an effective feedback tool is its
provision of explanations, specific corrections, or alternatives to students’ work in numerous instances.
For example, when a concluding sentence was absent in a piece of writing, ChatGPT provided the
following response, as can be seen in Table 5:

Overall, it seems that your paragraph is missing a clear concluding sentence.

A concluding sentence is important because it helps to summarize the main points of
the paragraph and bring closure to the argument you have made. Here’s an example
of a concluding sentence that you could use for your paragraph: … This sentence
reiterates the main idea of your paragraph and emphasizes…

Additionally, in cases where vocabulary errors were detected, ChatGPT offered targeted corrections,
such as:

For the most part, the vocabulary is correct. However, there are a few minor errors
that could be corrected: In the first sentence, “people get confused” could be
changed to “people become confused” for consistency with the tense of the rest of

Table 5 ChatGPT’s Comment on the Concluding Sentence in One Pair’s Writing

Concluding sentence Overall, it seems that your paragraph is missing a clear concluding
sentence. A concluding sentence is important because it helps to
summarize the main points of the paragraph and bring closure to the
argument you have made. Here’s an example of a concluding sentence that
you could use for your paragraph: “In conclusion, having almost unlimited
options in life can provide us with the freedom to make choices that align
with our personal values and goals, enabling us to live happier and more
fulfilling lives.” This sentence reiterates the main idea of your paragraph and
emphasizes the benefits of having almost unlimited options in life.
Note. During the second round of feedback, each pair input their work into ChatGPT which was prompted by a
checklist, and was instructed to provide feedback as to what needs to be altered. Students received feedback
from ChatGPT on the fulfillment of writing criteria and copied ChatGPT’s comments and suggestions into the
checklist.
13 Technology in Language Teaching & Learning, 6(3)

the text…” or: “In the first sentence, it should be “a” instead of “some” …In the
second sentence, it should be “real” instead of “the real” …, In the fourth sentence,
it should be “their” instead of “there”…and “avoid choosing” instead of “avoid
from the choose.

However, this study identified a less effective use of ChatGPT for feedback, demonstrated
by the observed inconsistency of responses, as depicted in Table 6 and Table 7. For instance,
when different students sought clarification on punctuation correctness, some received responses
such as:

As an AI language model, I cannot say for sure that there are no punctuation errors
in the essay, as punctuation can be subjective and there are many different styles and
conventions that may be used. However, based on my analysis, the essay appears to
have good punctuation overall.

Conversely, others were given responses like:

Yes, the punctuation in the essay is generally correct. There are a few minor punctu-
ation errors, such as missing commas and a misspelled word, but they do not signifi-
cantly affect the clarity of the text.

Moreover, some students received very specific answers like:

“In the first sentence, there should be a comma after “these days”…In the second main
idea sentence, there should be a comma after “study”…In the third main idea sentence,
there should be a comma after “anywhere”…”

Table 6 ChatGPT’s Comment on the Clarity of Sentences in One Pair’s Writing

All the sentences are clear As an AI language model, I cannot definitively say that all of the sentences
are clear and do not need any improvement, as language is a complex and
nuanced subject. However, based on my analysis, the sentences in the
essay appear to be well-written and communicate their intended meaning
effectively.
Note. Students received feedback from ChatGPT on the fulfillment of writing criteria and copied ChatGPT’s
comments and suggestions into the checklist.

Table 7 ChatGPT’s Comment on the Clarity of Sentences in Another Pair’s Writing

All the sentences are clear The sentences are generally clear, but there are a few areas where the
wording could be improved for greater clarity. For example, the second
sentence in the paragraph could be revised to say, “Having too many
options can be overwhelming and lead to mental exhaustion.” This would
clarify the cause-and-effect relationship between having multiple choices
and feeling overwhelmed.
Note. Students received feedback from ChatGPT on the fulfillment of writing criteria and copied ChatGPT’s
comments and suggestions into the checklist.
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 14

Similarly, when students inquired about sentence clarity, responses varied. Some received answers
such as:

As an AI language model, I cannot definitively say that all of the sentences are clear
and do not need any improvement, as language is a complex and nuanced subject.
However, based on my analysis, the sentences in the essay appear to be well-written
and communicate their intended meaning effectively.

On the other hand, others were provided with responses like:

The sentences are generally clear, but there are a few areas where the wording could be
improved for greater clarity. For example, the second sentence in the paragraph could be
revised to say… This would clarify the cause-and-effect relationship between…and…”

Su, Lin, and Lai (2023) also noted a lack of consistency in ChatGPT’s feedback, with varied responses
emerging upon each submission of student writing. This implies a necessity for multiple inputs into
ChatGPT, with subsequent examination of the diverse responses to better understand and address
potential issues.

Additionally, it was observed that most students favor teacher-provided WCF. This inclination towards
teacher WCF might stem from the traditional educational framework students are accustomed to,
wherein instructors traditionally hold the authority to assess and offer feedback on students’ work.
It may be inherently linked to students’ trust in and reliance on feedback from teachers, who are
perceived to possess expertise. Such a preference may also be reinforced by the established norms and
expectations within the educational environment.

The tendency to favor teacher WCF also aligns with socio-constructivist theories, which emphasize
the collaborative nature of teaching and learning and the role of social interactions in knowledge
construction (Vygotsky & Cole, 1978; Bruner, 1985; Rogoff, 1990; Lave & Wenger, 1990). According
to Vygotsky and Cole’s (1978) Zone of Proximal Development, learning occurs within the “zone”
where learners engage in collaborative activities with more knowledgeable instructors. With the help
of an instructor, students can understand and master knowledge and skills that they would not be able
to on their own (Schreiber & Valle, 2013). Once the students master a certain skill they can complete it
independently. In this theory, the instructor plays an integral role in students’ acquisition of knowledge
(Chen, 2015; Schreiber & Valle, 2013).

The third research question addressed the distribution of WCF across specific aspects of writing
provided by peers, ChatGPT, and the teacher. The results indicate that within peer feedback, the
areas for improvement were primarily organization and mechanics, with organization alone being
next. Subsequently, attention was directed toward content and mechanics alongside content alone.
Following these, feedback addressed organization and grammar, as well as organization and quality of
expression. Hence, it was reasonable to anticipate that participants’ revisions would primarily address
the organization of their writing, with some attention to content and mechanics, while placing relatively
less emphasis on grammar and quality of expression. Lack of sufficient linguistic knowledge among
reviewers may contribute to the inaccurate identification of linguistic errors. This concern is evident
in criticisms challenging the validity of feedback associated with peer review in EFL classrooms,
especially when reviewers have lower language proficiency (Covill, 2010; Guenette, 2007).

After the initial round of revisions and inputting the text into ChatGPT for feedback, a smaller number
of pairs received suggestions to work on organization. This corresponds with the prior revisions
15 Technology in Language Teaching & Learning, 6(3)

made by the participants. The areas marked for improvement in ChatGPT feedback included various
recommendations. Some pairs were advised to pay attention to all aspects of their writing, while others
were directed to focus on content, quality of expression, and grammar. Certain pairs were guided to
address issues related to mechanics and grammar. Fewer pairs received guidance to address either
organization, mechanics, and grammar; organization, content, and grammar; or content, mechanics,
and grammar. Notably, ChatGPT placed less emphasis on the necessity for enhancing the quality of
expression, while it precisely identified mechanics, content, and notably, grammar as areas warranting
attention. Given the inherently abstract nature of expression quality, this observation suggests a
potential requirement for refining the prompts employed in assessing expression quality. As noted by
Atlas (2023), language models like ChatGPT may deviate from the intended direction if prompts or
questions are not carefully worded. Thus, it is crucial to formulate clear and detailed prompts, enabling
the model to precisely comprehend the desired task. Another important principle is to encourage
ChatGPT to elaborate on its responses by providing feedback and requesting further information.

Following the second round of revisions and the submission of the revised work for evaluation by the
teacher, the feedback provided by the teacher highlighted specific areas requiring improvement. The
primary focus centered on mechanics and grammar, followed by content, mechanics, and grammar.
Additionally, the teacher addressed the following as requiring corrections: organization, content, and
grammar; organization, content, and the quality of expression; and quality of expression alone. In fewer
instances, pairs received feedback concerning the quality of expression, mechanics, and grammar;
as well as feedback spanning all writing categories. Much like ChatGPT’s assessment, the teacher’s
evaluation emphasized the need to address mechanics, content, and, notably, grammar. Furthermore, it
also identified the quality of expression as an area that warranted attention and improvement.

Despite technological advances and the potential of ChatGPT to function as a written corrective
feedback tool, it is worth noting that even after students received feedback from the tool and made
subsequent revisions, the teacher still identified errors in their writing. As some participants noted,
“GPT helped us a lot with making the writing good but the teacher paid attention to things that we
and GPT didn’t see.” Teachers have the expertise, experience, and real-world context that contribute
to the identification of writing errors and the provision of appropriate feedback. According to
Atlas (2023), although ChatGPT and other language models prove beneficial in proofreading and
editing student writing, they should not be viewed as replacements for human editing and feedback.
While these tools can help to identify errors and issues, they may not always be able to understand
the context and intent of the writing. Consequently, it is important to employ ChatGPT and other
language models in conjunction with human editing and feedback to guarantee the most accurate
and effective results.

Implications for Practice and Future Research

The study’s findings regarding the perceived value of ChatGPT’s feedback among participants have
implications for the EFL classroom. Drawing from autonomous learning theory, as explained by
Little (1991), which emphasizes the importance of learners taking ownership of their learning process
and actively seeking feedback to enhance their knowledge, it becomes evident that integrating peer
and AI-based feedback can foster learner autonomy. These forms of feedback offer immediate and
accessible support for self-directed learning. However, effective autonomous learning necessitates
learners’ ability to critically evaluate and integrate feedback into their learning strategies, a process
that may benefit from instructors’ guidance and scaffolding, as discussed by Benson (2013). Therefore,
recognizing the potential of ChatGPT’s feedback, teachers could consider its incorporation into the
language learning process to further empower students in their autonomous learning efforts. This
integration would necessitate training students to formulate clear and detailed prompts, ensuring the
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 16

model precisely comprehends the task at hand. Moreover, encouraging students to prompt ChatGPT
to elaborate on its responses can enhance the depth of feedback. Additionally, as proposed by Atlas
(2023), teachers can train a ChatGPT model using a dataset of student essays previously graded and
corrected by the teacher. This trained model can then analyze new essays, flagging errors for the
teacher’s review and subsequent feedback to the student. Yet, it is important to consider the potential
inconsistency in responses from ChatGPT. This underscores the necessity for multiple inputs into
ChatGPT and subsequent examination of diverse responses to better understand and address potential
issues.

Rather than banning the use of AI tools in the classroom, educators can harness these tools to facilitate
student learning, provided appropriate guidance is given (Otsuki, 2020; Hellmich & Vinall, 2021;
Carvalho et al, 2022). Teachers often face the challenges of large class sizes and time constraints
which may limit the level of individual attention that students receive (Lee, 2008). In this context,
AI feedback can complement teacher feedback by providing immediate support, enabling students to
work on their writing independently and making timely corrections. As suggested by Barrot (2023),
ChatGPT can be useful as an L2 writing assistant and should be adopted as a supplementary tool for
essay composition. According to Barrot (2023), students can be encouraged to write their original
outputs first and then refine them using ChatGPT. Teachers can emphasize the value of the writing
process, and leverage ChatGPT’s editing capabilities to teach students correct language forms and
style. Students can instruct ChatGPT to edit their work, review the changes made by ChatGPT, and
decide whether to accept or reject them based on their assessment.

While integrating ChatGPT into EFL classrooms holds promise, it also presents practical challenges
and requires careful consideration of educational institutions’ readiness to implement such
technology effectively. Educational institutions need to ensure they possess the essential technical
infrastructure to incorporate AI-based tools such as ChatGPT into their classrooms. This involves
securing reliable internet connectivity, providing access to computers, and ensuring compatibility
with the software platform. Teachers need training to effectively integrate ChatGPT into their
instructional practices. This training should involve not only the technical aspects of using the tool
but also strategies for integrating AI feedback with traditional teaching methods and addressing
potential challenges. Integration of ChatGPT should be aligned with the pedagogical goals and
objectives of the language learning curriculum. Teachers should consider how AI-based feedback
complements instructional practices and supports students’ language learning goals.

Students also require instruction in the proper use of AI tools. Warschauer et al. (2023) introduce
an AI literacy framework aimed at seamlessly integrating AI tools into L2 writing instruction. The
framework suggests several key steps. Firstly, students must grasp the fundamental functionalities,
strengths, weaknesses, and biases of AI writing tools. Secondly, they should develop proficiency in
accessing and navigating these tools to address specific communication tasks, such as drafting papers
or gathering background information. Thirdly, students need to expertly prompt AI to generate content
aligned with their writing objectives. Fourthly, they should verify the accuracy of AI-generated content
throughout the writing and revision stages. Lastly, students must learn to ethically and effectively
incorporate AI-generated texts into their writing, acknowledging and citing their use of AI in the
authoring process. They claim that these elements are essential to building second language students’
understanding of how to use AI to support their learning.

Several limitations of this pilot study must be noted. Firstly, the sequence of feedback provision—
peer-ChatGPT-teacher—employed in this study could have influenced the results. It is plausible that
altering the feedback order might have produced different outcomes. Secondly, the small number of
participants means that generalization of the findings is difficult to make. Thirdly, the quality and
17 Technology in Language Teaching & Learning, 6(3)

effectiveness of feedback may differ across teachers, peers, and AI systems, introducing potential
variability in students’ perceptions and preferences. Fourthly, students’ perceptions and attitudes
toward feedback can be influenced by various factors, including their prior experiences, language
proficiency, and individual preferences. These factors were not addressed in the current study. Lastly,
this study only captured students’ perceptions at a particular point in time. It did not account for
potential shifts in attitudes or preferences over time, especially in light of the novel use of ChatGPT
as a feedback tool.

Given that much is still unknown about the use of ChatGPT as a WCF tool in the EFL context,
future research should investigate its long-term effects. Furthermore, there is a need for longitudinal
studies focusing on feedback combination strategies. Such studies can provide insights into catering
feedback approaches to diverse learner needs in EFL contexts. Another avenue of future study could
be an exploration of the factors influencing students’ preferences for specific feedback methods.
Uncovering these factors could help refine feedback practices and identify areas warranting
improvement in feedback processes.

Conclusion

This qualitative pilot study attempted to explore EFL students’ perceptions of the effectiveness of
peer, artificial intelligence (AI), and teacher feedback, and to uncover students’ preferences. The
participants completed a short writing task which was given written corrective feedback by peers, an
AI tool (ChatGPT), and the teacher. Each feedback mode was followed by students’ revision of their
writing. The data analysis results demonstrated that participants prefer to receive teacher WCF, and a
combination of teacher and ChatGPT WCF. An additional aim was to identify the percentage of the
feedback given to specific components of the writing. The distribution of WCF across specific aspects
of the writing has been discussed. The implication for the EFL classroom is that teachers can integrate
ChatGPT into the language learning process, provided appropriate guidance is given. Future research
should investigate the long-term effects of using ChatGPT as a WCF tool in the EFL context, study
feedback combination strategies, and explore the factors influencing students’ preferences for specific
feedback methods.

This research received no specific grant from any funding agency in the public, commercial, or not-for-
profit sectors. Funding for this research was covered by the author(s) of the article.

References

Adams, R., Nuevo, A., & Egi, T. (2011). Explicit and implicit feedback, modified output, and SLA:
Does explicit and implicit feedback promote learning and learner-learner interactions?
Modern Language Journal, 95(1), 42–63. https://doi.org/10.1111/j.1540-4781.2011.01242.x
Alharbi, W. (2023). AI in the Foreign Language Classroom: A Pedagogical Overview of
Automated Writing Assistance Tools. Education Research International. https://doi.
org/10.1155/2023/4253331
Atlas, S. (2023). ChatGPT for higher education and professional development: A guide to
conversational AI. College of Business Faculty Publications. https://digitalcommons.uri.
edu/cba_facpubs/548
Barrot, J.S. (2023). Using ChatGPT for second language writing: Pitfalls and potentials. Assessing
Writing, 57, 100745. https://doi.org/10.1016/j.asw.2023.100745
Beason, L. (1993). Feedback and revision in writing across the curriculum classes. Research in The
Teaching of English, 27, 395–421.
Benson, P. (2013). Teaching and researching: Autonomy in language learning. Routledge.
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 18

Bitchener, J., & Ferris, D. (2012). Written corrective feedback in second language acquisition and
writing. Routledge.
Brown, J. D., & Bailey, K. M. (1984). A categorical instrument for scoring second language
writing skills. Language Learning, 34(4), 21–38. https://doi.org/10.1111/j.1467-1770.1984.
tb00350.x
Bruner, J. (1985). Vygotsky: A historical and conceptual perspective. Culture, Communication, and
Cognition: Vygotskian perspectives, 21, 34.
Carvalho, L., Martinez-Maldonado, R., Tsai, Y.S., Markauskaite, L., & De Laat, M. (2022). How can
we design for learning in an AI world? Computers and Education: Artificial Intelligence, 3,
Article ID 100053. https://doi.org/10.1016/j.caeai.2022.100053
Chen, B. (2015). Social constructivism of language and meaning. Croatian Journal of Philosophy,
15(43), 87–113.
Cho, K., & MacArthur, C. (2010). Student revision with peer and expert reviewing. Learning and
Instruction, 20(4), 328–338.
Colpitts, B. D. (2016). Japanese students’ perceptions of peer corrective feedback in an EFL
classroom. Acta Humanistica et Scientifica Humanities Series, 49, 345–358.
Connor, U., & Asenavage, K. (1994). Peer response groups in ESL writing classes: How much
impact on revision? Journal of Second Language Writing, 3(3), 257–276. https://doi.
org/10.1016/1060-3743(94)90019-1
Covill, A. E. (2010). Comparing peer review and self-review as ways to improve college
students’ writing. Journal of Literacy Research, 42(2), 199–226. https://doi.org/10.1080/
10862961003796207.
Dai, W., Lin, J., Jin, H., Li, T., Tsai, Y. S., Gašević, D., & Chen, G. (2023). Can large language Models
provide feedback to students? A case study on ChatGPT. In 2023 IEEE International Conference
on Advanced Learning Technologies (ICALT) (pp. 323–325), Orem, UT, USA. doi: 10.1109/
ICALT58122.2023.00100.
Donato, R. (1988). Beyond group: A psycholinguistic rationale for collective activity in second-
language learning. University of Delaware.
Donato, R. (1994). Collective scaffolding in second language learning. In J. P. Lantolf, & G. Appel
(Eds.), Vygotskian approaches to second language research (pp. 33–56). Ablex.
Elola, I., & Oskoz, A. (2010). Collaborative writing: Fostering foreign language and writing
conventions development. Language Learning & Technology, 14(3), 51–71.
Fernandez Dobao, A. (2014). Attention to form in collaborative writing tasks: Comparing pair and
small group interaction. Canadian Modern Language Review, 70(2), 158–187. https://doi.
org/10.3138/cmlr.1768
Ferris, D.R. (1997). The Influence of Teacher Commentary on Student Revision. TESOL Quarterly,
31(2), 315–339, https://doi.org/10.2307/3588049.
Ferris, D., & Hedgcock, J. (2005). Teaching ESL composition: Purpose, process, and practice.
Lawrence Erlbaum.
Ferris, D. R., & Roberts, B. (2001). Error feedback in L2 writing classes: How explicit does it
need to be? Journal of Second Language Writing, 10(3), 161–184, https://doi.org/10.1016/
S1060- 3743(01)00039-X.
Garcia Mayo, M.P., & Imaz Agirre, A. (2019). Task modality and pair formation method: Their impact
on patterns of interaction and LREs among EFL primary school children. System, 80, 165–175.
https://doi.org/10.1016/j.system.2018.11.011
Guenette, D. (2007). Is feedback pedagogically correct?: Research design issues in studies of
feedback on writing. Journal of Second Language Writing, 16(1), 40–53. https://doi.
org/10.1016/j.jslw.2007.01.001
Hansen, J. G., & Liu, J. (2005). Guiding principles for effective peer response. ELT Journal, 59(1),
31–38. https://doi.org/10.1093/elt/cci004
19 Technology in Language Teaching & Learning, 6(3)

Hattie, J., & Timperley, H. (2007). The Power of Feedback. Review of Educational Research 77(1),
81–112, https://doi.org/10.3102/003465430298487.
Hedgcock, J., & Lefkowitz, N. (1992). Collaborative oral/aural revision in foreign language
writing instruction. Journal of Second Language Writing, 1(3), 255–276. https://doi.
org/10.1016/1060-3743(92)90006-B
Hellmich, E., & Vinall, K. (2021). FL instructor beliefs about machine translation: Ecological insights
to guide research and practice. International Journal of Computer-Assisted Language Learning
and Teaching (IJCALLT), 11(4), 1–18.
Hewett, B. L. (2000). Characteristics of interactive oral and computer-mediated peer group talk and its
influence on revision. Computers and Composition, 17(3), 265–288. https://doi.org/10.1016/
S8755-4615(00)00035-9
Hyland, K. (2003). Second language writing. Cambridge University Press. https://doi.org/10.1017/
CBO9780511667251.
Kamimura, T. (2006). Effects of peer feedback on EFL student writers at different levels of English
proficiency: A Japanese context. TESL Canada Journal, 23(2), 12–39.
Kim, Y. (2008). The contribution of collaborative and individual tasks to the acquisition
of L2 vocabulary. The Modern Language Journal, 92(1), 114–130. https://doi.org/10.1111/
j.1540-4781.2008.00690.x
Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge
University Press.
Lee, I. (2008). Understanding teachers’ written feedback practices in Hong Kong secondary
classrooms. Journal of Second Language Writing, 17(2), 69– 85. https://doi.org/10.1016/j.
jslw.2007.10.001
Lin, W. C., & Yang, S. C. (2011). Exploring students’ perceptions of integrating Wiki technology and
peer feedback into English writing courses. English Teaching: Practice and Critique, 10(2),
88–103.
Little, D. (1991). Learner Autonomy 1: Definitions, issues and problems. Dublin: Authentik.
Liu, J., & Hansen, J. G. (2002). Peer response in second language classroom. University of Michigan
Press.
Lundstrom, K., & Baker, W. (2009). To give is better than to receive: The benefits of peer review to the
reviewer’s own writing. Journal of Second Language Writing, 18(1), 30–43.
Lyster, R. (1998). Recasts, repetition, and ambiguity in L2 classroom discourse. Studies in Second
Language Acquisition, 20(1), 51–81. https://doi.org/10.1017/S027226319800103X
OpenAI. (2023). ChatGPT: Optimizing language models for dialogue. https://openai.com/blog/
chatgpt/
Otsuki, G.J. (2020). OK computer: To prevent students cheating with AI text-generators, we should
bring them into the classroom. The Conversation (blog). https://theconversation.com/
ok-computer-to-prevent-students-cheating-with-ai-text-generators-we-should-bring-them-
into-the-classroom-129905
Paulus, T. M. (1999). The effect of peer and teacher feedback on student writing. Journal of Second
Language Writing, 8(3), 265–289. https://doi.org/10.1016/S1060-3743(99)80117-9
Peterson, S. S., & Portier, C. (2014). Grade one peer and teacher feedback on student writing. Education
3–13, 42(3), 237–257, DOI: 10.1080/03004279.2012.670256.
Philp, J., Walter, S., & Basturkmen, H. (2010). Peer interaction in the foreign language classroom:
What factors foster a focus on form? Language Awareness, 19(4), 261–279. https://doi.org/10.
1080/09658416.2010.516831
Rogoff, B. (1990). Apprenticeship in thinking: Cognitive development in social context. Oxford
University Press.
Ruegg, R. (2015). The relative effects of peer and teacher feedback on improvement in EFL students’
writing ability. Linguistics and Education, 29, 73–82.
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 20

Sato, M. (2013). Beliefs about peer interaction and peer corrective feedback: Efficacy of classroom
intervention. Modern Language Journal, 97(3), 611–633. https://doi.org/10.1111/j.1540-
4781.2013.12035.x
Sato, M., & Lyster, R. (2012). Peer interaction and corrective feedback for accuracy and fluency
development: Monitoring, practice, and proceduralization. Studies in Second Language
Acquisition, 34(4), 591–626. https://doi.org/10.1017/S0272263112000356
Schreiber, L. M., & Valle, B. E. (2013). Social constructivist teaching strategies in the small group
classroom. Small Group Research, 44(4), 395–411.
Sheen, Y. (2010a). Differential effects of oral and written corrective feedback in the ESL classroom.
Studies in Second Language Acquisition, 32(2), 203–234. https://doi.org/10.1017/
S0272263109990507
Sheen, Y. (2010b). Introduction: The role of oral and written corrective feedback in SLA. Studies in
Second Language Acquisition, 32(2), 169–179. doi:10.1017/S0272263109990489
Sheen, Y., Wright, D., & Moldawa, A. (2009). Differential effects of focused and unfocused written
correction on the accurate use of grammatical forms by adult ESL learners. System, 37(4),
556–569. https://doi.org/10.1016/j.system.2009.09.002
Shehadeh, A. (2011). Effects and student perceptions of collaborative writing in L2. Journal of Second
Language Writing, 20(4), 286–305. https://doi.org/10.1016/j.jslw.2011.05.010
Sippel, L., & Jackson, C. N. (2015). Teacher vs. peer oral corrective feedback in the German
language classroom. Foreign Language Annals, 48(4), 688–705. https://doi.org/10.1111/
flan.12164
Steiss, J., Tate, T., Graham, S., Cruz, J., Hebert, M., Wang, J., Moon, Y., Tseng, W., Warschauer,
M., & Olson, C. B. (2024). Comparing the quality of human and ChatGPT feedback of
students’ writing. Learning and Instruction, 91, 101894.
Storch, N. (2005). Collaborative writing: Product, process, and students’ reflections. Journal of Second
Language Writing, 14(3), 153–173. https://doi.org/10.1016/j.jslw.2005.05.002
Su, Y., Lin, Y., & Lai, C. (2023). Collaborating with ChatGPT in argumentative writing class-
rooms. Assessing Writing, 57, 100752. https://doi.org/10.1016/j.asw.2023.100752
Tai, H.C., Lin, W.C., & Yang, S.C. (2015). Exploring the effects of peer review and teachers’ corrective
feedback on EFL students’ online writing performance. Journal of Educational Computing
Research, 53(2), 284–309. https://doi.org/10.1177/0735633115597490
Topping, K. (1998). Peer assessment between students in colleges and universities. Review of
Educational Research, 68(3), 249–276.
van Dis, E. A., Bollen, J., Zuidema, W., van Rooij, R., & Bockting, C. L. (2023). ChatGPT: Five priorities
for research. Nature, 614(7947), 224–226. https://doi.org/10.1038/d41586-023-00288-7
Villarreal, I., & Gil-Sarratea, N. (2019). The effect of collaborative writing in an EFL secondary
setting. Language Teaching Research. Epub ahead of print 15 February 2019. DOI:
10.1177/1362168819829017.
Vygotsky, L. S., & Cole, M. (1978). Mind in society: Development of higher psychological processes.
Harvard University Press.
Warschauer, M., Tseng,W., Soobin, Y., Webster, T., Jacob, S., Du, Q., & Tate, T. (2023). The affordances
and contradictions of AI-generated text for writers of English as a second or foreign language.
Journal of Second Language Writing, 62, 101071. https://doi.org/10.1016/j.jslw.2023.101071
Wenzlaff, K., & Spaeth, S. (2022). Smarter than Humans? Validating how OpenAI’s ChatGPT
model explains Crowdfunding, Alternative Finance and Community Finance. Validating
how OpenAI’s ChatGPT model explains Crowdfunding, Alternative Finance and Community
Finance. (December 22, 2022). https://ssrn.com/abstract=4302443
Williams, J. G. (2003). Providing feedback on ESL students’ written assignments. The Internet TESL
Journal, 9, 1–5.
21 Technology in Language Teaching & Learning, 6(3)

Wu, Y. (2024). Study on the impact of utilizing ChatGPT and other AI tools for feedback in
EAP writing classrooms on the discursive writing performance of English major
students. Transactions on Social Science, Education and Humanities Research, 4, 143–150.
Xiao, Y., & Lucking, R. (2008). The impact of two types of peer assessment on students’ performance
and satisfaction within a Wiki environment. Internet and Higher Education, 11(3–4), 186–193.
https://doi.org/10.1016/j.iheduc.2008.06.005
Yang, M., Badger, R., & Yu, Z. (2006). A comparative study of peer and teacher feedback in a
Chinese EFL writing class. Journal of Second Language Writing, 15(3), 179–200. https://doi.
org/10.1016/j.jslw.2006.09.004
Yoon, S. Y., Miszoglad, E., & Pierce, L. R. (2023). Evaluation of ChatGPT Feedback on ELL
Writers’ Coherence and Cohesion. arXiv preprint arXiv:2310.06505. https://doi.org/10.48550/
arXiv.2310.06505
Zhang, S. (1995). Reexamining the affective advantage of peer feedback in the ESL writing class.
Journal of second language writing, 4(3), 209–222. https://doi.org/10.1016/1060-
3743(95)90010-1
Comparing Peer, ChatGPT, and Teacher Corrective Feedback in EFL Writing 22

Appendix A

Table 8 Feedback Checklist.

Criteria √ if criterion Comments: What needs to be corrected/

fulfilled improved?
Organization
Thesis statement
Three ideas
Each idea is followed by two
additional sentences
Concluding sentence
Content
The thesis indicates the
topic, and includes a claim
about the topic
The three ideas support the
thesis statement
Each idea is followed by two
additional sentences that
explain/exemplify the idea
The concluding sentence
restates the thesis and
sums up the ideas/gives
a suggestion/a personal
statement about the topic
Grammar
The sentences are
grammatically correct
Mechanics
The punctuation is correct
The spelling is correct
The choice of words is
accurate
Quality of expression
All the sentences are clear
All the sentences are
logically connected
Note. A feedback checklist is used in each review round to ensure the fulfillment of specific writing criteria and
to incorporate comments from the evaluators. The checklist is adapted from the criteria developed by Brown
and Bailey (1984), “The Composition Grading Feedback Sheet”.
23 Technology in Language Teaching & Learning, 6(3)

Appendix B

Feedback Survey

1. a) What did your peers think required correction? Circle your answers.
Organization  Content  Grammar  Mechanics  Quality of expression
b) Did you understand all the comments?
c) Did you agree with/accept the comments?
If not, why?
d) Did you find the feedback helpful?
Why?

2. a) What did ChatGPT suggest correcting? Circle your answers.

Organization  Content  Grammar  Mechanics  Quality of expression
b) Did you understand all the comments?
c) Did you agree with/accept the comments?
If not, why?
d) Did you find the feedback helpful?
Why?

3. a) What did the teacher think required correction? Circle your answers.
Organization  Content  Grammar  Mechanics  Quality of expression
b) Did you understand all the comments?
c) Did you agree with/accept the comments?
If not, why?
d) Did you find the feedback helpful?
Why?

4. a) What feedback or feedback combination do you prefer to receive? Check (√) your preferred
feedback mode, or specify your preferred combination.

__ Peer feedback
__ ChatGPT feedback
__ Teacher feedback
__ A combination of: ______________________________________________________