IMPACT OF CORRECTIVE METALINGUISTIC
WRITTEN FEEDBACK GIVEN THROUGH
CHATGPT ON EFL PRE-SERVICE TEACHERS’
ACADEMIC WRITING SKILLS
EL IMPACTO DE LA RETROALIMENTACIÓN METALINGÜÍSTICA
ESCRITA CORRECTIVA PROPORCIONADA A TRAVÉS DE CHATGPT
EN LAS HABILIDADES DE ESCRITURA ACADÉMICA DE DOCENTES
EN FORMACIÓN DE INGLÉS COMO LENGUA EXTRANJERA.
Stewart Smith Viáfara Pérez
Universidad El Bosque
Daniela Sandoval Lasso
Universidad El Bosque
pág. 8155
DOI: https://doi.org/10.37811/cl_rcm.v10i2.23809
Impact of corrective metalinguistic written feedback given through
ChatGPT on EFL pre-service teachers’ academic writing skills
Stewart Smith Viáfara Pérez1
sviafarap@unbosque.edu.co
https://orcid.org/0009-0005-1303-4167
Universidad El Bosque
Colombia
Daniela Sandoval Lasso
dsandovalla@unbosque.edu.co
https://orcid.org/0009-0003-9118-5345
Universidad El Bosque
Colombia
ABSTRACT
This study investigates the impact of ChatGPT-generated corrective metalinguistic written feedback on
the academic writing performance of EFL undergraduate pre-service teachers. We implemented a true
experimental pretest-posttest control group design with 14 participants randomly allocated to an
experimental group (ChatGPT feedback) and a control group (teacher feedback). In addition, an analytic
rubric that rated Grammar and Accuracy, Vocabulary Range and Appropriacy, Cohesion and Coherence,
Task Achievement, Organization’ perception questionnaires and open-ended replies complemented the
quantitative data to evaluate writing performance. Descriptive data indicated enhancement in both
groups, with the experimental exhibiting superior mean gain scores (M = 7.29) compared to the control
group (M = 6.29). Paired-samples t-tests demonstrated statistically significant improvements from pre-
test to post-test in both groups (p < .001). Category-level analysis indicated stronger gains for the
experimental group in Task Achievement, Organization, Grammar, and Accuracy. Perception results
indicated more positive assessments of clarity and actionability in the experimental group (e.g., 71%
strongly concurred that the feedback was clear, and comprehensible). Findings suggest that ChatGPT-
mediated metalinguistic feedback can effectively enhance academic writing growth and may function
as an auxiliary feedback resource in EFL teacher education settings.
Keywords: Metalinguistic Written Feedback; ChatGPT; Corrective feedback; EFL Writing Skills
1 Autor principal
Correspondencia: sviafarap@unbosque.edu.co
DOI: https://doi.org/10.37811/cl_rcm.v10i2.23809
Impact of corrective metalinguistic written feedback given through
ChatGPT on EFL pre-service teachers’ academic writing skills
Stewart Smith Viáfara Pérez1
sviafarap@unbosque.edu.co
https://orcid.org/0009-0005-1303-4167
Universidad El Bosque
Colombia
Daniela Sandoval Lasso
dsandovalla@unbosque.edu.co
https://orcid.org/0009-0003-9118-5345
Universidad El Bosque
Colombia
ABSTRACT
This study investigates the impact of ChatGPT-generated corrective metalinguistic written feedback on
the academic writing performance of EFL undergraduate pre-service teachers. We implemented a true
experimental pretest-posttest control group design with 14 participants randomly allocated to an
experimental group (ChatGPT feedback) and a control group (teacher feedback). In addition, an analytic
rubric that rated Grammar and Accuracy, Vocabulary Range and Appropriacy, Cohesion and Coherence,
Task Achievement, Organization’ perception questionnaires and open-ended replies complemented the
quantitative data to evaluate writing performance. Descriptive data indicated enhancement in both
groups, with the experimental exhibiting superior mean gain scores (M = 7.29) compared to the control
group (M = 6.29). Paired-samples t-tests demonstrated statistically significant improvements from pre-
test to post-test in both groups (p < .001). Category-level analysis indicated stronger gains for the
experimental group in Task Achievement, Organization, Grammar, and Accuracy. Perception results
indicated more positive assessments of clarity and actionability in the experimental group (e.g., 71%
strongly concurred that the feedback was clear, and comprehensible). Findings suggest that ChatGPT-
mediated metalinguistic feedback can effectively enhance academic writing growth and may function
as an auxiliary feedback resource in EFL teacher education settings.
Keywords: Metalinguistic Written Feedback; ChatGPT; Corrective feedback; EFL Writing Skills
1 Autor principal
Correspondencia: sviafarap@unbosque.edu.co
pág. 8156
El impacto de la retroalimentación metalingüística escrita correctiva
proporcionada a través de ChatGPT en las habilidades de escritura
académica de docentes en formación de inglés como lengua extranjera.
RESUMEN
Este estudio investiga el impacto de la retroalimentación escrita metalingüística correctiva generada por
ChatGPT en el rendimiento de la escritura académica de los futuros profesores de inglés como lengua
extranjera (EFL) a nivel de pregrado. Implementamos un diseño de grupo control pretest-post-test
experimental verdadero con 14 participantes asignados aleatoriamente a un grupo experimental
(retroalimentación de ChatGPT) y un grupo control (retroalimentación del profesor). Asimismo, la
evaluación del desempeño escrito utilizó una rúbrica analítica que calificó la gramática y la precisión,
la amplitud y la adecuación del vocabulario, la cohesión y la coherencia, así como la consecución de la
tarea y la organización. Los cuestionarios de percepción y las respuestas abiertas complementaron los
datos cuantitativos. Los datos descriptivos indicaron una mejora en ambos grupos, con el experimental
mostrando puntuaciones medias de ganancia superiores (M = 7.29) en comparación con el grupo de
control (M = 6.29). Las pruebas t de dos muestras indicaron mejoras estadísticamente significativas
desde el pre-test al post-test en ambos grupos (p < .001). El análisis de la rúbrica por categorías revelo
ganancias más fuertes para el grupo experimental en el cumplimiento la tarea, Organización, Gramática
y Precisión. Los resultados de la percepción de cada uno de los participantes indicaron evaluaciones más
positivas de claridad y viabilidad en el grupo experimental (por ejemplo, el 71% estuvo totalmente de
acuerdo en que la retroalimentación era clara y comprensible). Los resultados sugieren que la
retroalimentación metalingüística generada a través de ChatGPT puede mejorar eficazmente el
crecimiento de la escritura académica y a su vez podría funcionar como un recurso de retroalimentación
auxiliar en entornos de formación de profesores de inglés como lengua extranjera.
Palabras clave: Retroalimentación metalingüística escrita; ChatGPT; Retroalimentación correctiva;
EFL Habilidades de escritura
Artículo recibido 02 abril 2026
Aceptado para publicación: 30 abril 2026
El impacto de la retroalimentación metalingüística escrita correctiva
proporcionada a través de ChatGPT en las habilidades de escritura
académica de docentes en formación de inglés como lengua extranjera.
RESUMEN
Este estudio investiga el impacto de la retroalimentación escrita metalingüística correctiva generada por
ChatGPT en el rendimiento de la escritura académica de los futuros profesores de inglés como lengua
extranjera (EFL) a nivel de pregrado. Implementamos un diseño de grupo control pretest-post-test
experimental verdadero con 14 participantes asignados aleatoriamente a un grupo experimental
(retroalimentación de ChatGPT) y un grupo control (retroalimentación del profesor). Asimismo, la
evaluación del desempeño escrito utilizó una rúbrica analítica que calificó la gramática y la precisión,
la amplitud y la adecuación del vocabulario, la cohesión y la coherencia, así como la consecución de la
tarea y la organización. Los cuestionarios de percepción y las respuestas abiertas complementaron los
datos cuantitativos. Los datos descriptivos indicaron una mejora en ambos grupos, con el experimental
mostrando puntuaciones medias de ganancia superiores (M = 7.29) en comparación con el grupo de
control (M = 6.29). Las pruebas t de dos muestras indicaron mejoras estadísticamente significativas
desde el pre-test al post-test en ambos grupos (p < .001). El análisis de la rúbrica por categorías revelo
ganancias más fuertes para el grupo experimental en el cumplimiento la tarea, Organización, Gramática
y Precisión. Los resultados de la percepción de cada uno de los participantes indicaron evaluaciones más
positivas de claridad y viabilidad en el grupo experimental (por ejemplo, el 71% estuvo totalmente de
acuerdo en que la retroalimentación era clara y comprensible). Los resultados sugieren que la
retroalimentación metalingüística generada a través de ChatGPT puede mejorar eficazmente el
crecimiento de la escritura académica y a su vez podría funcionar como un recurso de retroalimentación
auxiliar en entornos de formación de profesores de inglés como lengua extranjera.
Palabras clave: Retroalimentación metalingüística escrita; ChatGPT; Retroalimentación correctiva;
EFL Habilidades de escritura
Artículo recibido 02 abril 2026
Aceptado para publicación: 30 abril 2026
pág. 8157
INTRODUCTION
In the past decades, the integration of artificial intelligence (AI) has come to play a significant role in
the educational context, especially in the fields of applied linguistics and language education. Writing,
as one of the most complex and demanding skills in second language acquisition, has long posed
challenges for learners and educators alike (Hyland, 2019). Corrective feedback, particularly in written
form, is seen as a feasible method to address these challenges. Bitchener and Ferris (2012) argue that
this type of feedback has been widely recognized as a pivotal pedagogical tool for enhancing learners'
accuracy and fluency. Among the various feedback mechanisms, metalinguistic corrective feedback, by
which the nature of an error is explained explicitly, has been shown to promote noticing and long-term
language development (Ellis, 2009; Lyster, Saito, & Sato, 2013). For the most part, such feedback has
been provided by teachers; nonetheless, the increasing demand for individualized and immediate
feedback has been a key factor in exploring alternative technologies.
With the emergence of AI tools such as ChatGPT, the alternatives for providing immediate and
personalized corrective feedback have increased significantly. ChatGPT has the potential to generate
metalinguistic explanations and suggest revisions in real time (Barrot, 2023), which may foster learner
agency beyond the instructional context. Corrective feedback has been studied in second language
writing; nevertheless, empirical evidence regarding the impact of ChatGPT’s corrective metalinguistic
feedback on pre-service teacher’s writing development remains limited. Existing studies have essentially
focused on teacher and peer feedback, leaving a gap in how AI-generated metalinguistic feedback
influences learners' writing accuracy.
With that in mind, this research aimed to address this gap by examining the impact of corrective
metalinguistic written feedback generated by ChatGPT in the writing skills of undergraduate students
in bilingualism (English and Spanish) in a Colombian university. We carried out this study considering
both the results of the students’ argumentative essays and their reflections on the writing process,
including whether they perceived any improvement in their writing skills. Additionally, by adopting a
mixed-method approach, the study made a distinction between learners who received ChatGPT-based
feedback and those who received traditional teacher feedback across pre and posttests essay writing
tasks. By doing so, this research aimed to offer a comprehensible understanding of the impact of using
INTRODUCTION
In the past decades, the integration of artificial intelligence (AI) has come to play a significant role in
the educational context, especially in the fields of applied linguistics and language education. Writing,
as one of the most complex and demanding skills in second language acquisition, has long posed
challenges for learners and educators alike (Hyland, 2019). Corrective feedback, particularly in written
form, is seen as a feasible method to address these challenges. Bitchener and Ferris (2012) argue that
this type of feedback has been widely recognized as a pivotal pedagogical tool for enhancing learners'
accuracy and fluency. Among the various feedback mechanisms, metalinguistic corrective feedback, by
which the nature of an error is explained explicitly, has been shown to promote noticing and long-term
language development (Ellis, 2009; Lyster, Saito, & Sato, 2013). For the most part, such feedback has
been provided by teachers; nonetheless, the increasing demand for individualized and immediate
feedback has been a key factor in exploring alternative technologies.
With the emergence of AI tools such as ChatGPT, the alternatives for providing immediate and
personalized corrective feedback have increased significantly. ChatGPT has the potential to generate
metalinguistic explanations and suggest revisions in real time (Barrot, 2023), which may foster learner
agency beyond the instructional context. Corrective feedback has been studied in second language
writing; nevertheless, empirical evidence regarding the impact of ChatGPT’s corrective metalinguistic
feedback on pre-service teacher’s writing development remains limited. Existing studies have essentially
focused on teacher and peer feedback, leaving a gap in how AI-generated metalinguistic feedback
influences learners' writing accuracy.
With that in mind, this research aimed to address this gap by examining the impact of corrective
metalinguistic written feedback generated by ChatGPT in the writing skills of undergraduate students
in bilingualism (English and Spanish) in a Colombian university. We carried out this study considering
both the results of the students’ argumentative essays and their reflections on the writing process,
including whether they perceived any improvement in their writing skills. Additionally, by adopting a
mixed-method approach, the study made a distinction between learners who received ChatGPT-based
feedback and those who received traditional teacher feedback across pre and posttests essay writing
tasks. By doing so, this research aimed to offer a comprehensible understanding of the impact of using
pág. 8158
this AI in written tasks, enabling teachers to incorporate ChatGPT into their feedback practices in a safe
and reliable manner, thereby facilitating and streamlining classroom instruction, as well as fostering
autonomy in the students' learning process, empowering them to actively participate in their own
educational journey.
LITERATURE REVIEW
Academic Writing
Academic writing is widely regarded as one of the most demanding skills in EFL learners, as it requires
both linguistic accuracy and mastery of discourse organization and genre conventions (Hyland, 2019).
In higher education, argumentative essays represent a crucial genre by which students demonstrate
critical thinking, adopting a position, and coherence in developing ideas (Wingate, 2012).
Undergraduate students in language related programs face pressure to achieve this proficiency, as their
academic and future professional performance depends on their ability to write with clarity, accuracy,
and persuasiveness. Research has shown that analytic rubric which distinguishes between grammar,
vocabulary, cohesion, and organization provides a reliable way to capture the multifaceted nature of
writing proficiency (Weigle, 2002; Andrade, 2005). This multifaceted perspective emphasizes the
necessity for focused instructional and feedback methodologies that may tackle both form and content.
Corrective and metalinguistic written feedback
Corrective written feedback (WCF) has been recognized as a central pedagogical tool to support such
development. Ferris (2010) and Bitchener and Ferris (2012) highlight its role in drawing learners’
attention to linguistic errors while enhancing long-term accuracy and fluency. Within this framework,
metalinguistic corrective feedback, feedback that not only identifies an error but also explains the
underlying rule or principle, has been found to enhance learners’ noticing and promote greater self-
regulation in revision (Ellis, 2009; Lyster, Saito, & Sato, 2013). Previous studies suggest that
metalinguistic feedback is particularly effective in academic contexts, where learners are expected to
internalize conventions and apply them autonomously across genres (Shintani & Ellis, 2015). Therefore,
grounding corrective practices in explicit explanation rather than simple error indication aligns
appropriately with the needs of undergraduate EFL students, especially those preparing to become
language professionals.
this AI in written tasks, enabling teachers to incorporate ChatGPT into their feedback practices in a safe
and reliable manner, thereby facilitating and streamlining classroom instruction, as well as fostering
autonomy in the students' learning process, empowering them to actively participate in their own
educational journey.
LITERATURE REVIEW
Academic Writing
Academic writing is widely regarded as one of the most demanding skills in EFL learners, as it requires
both linguistic accuracy and mastery of discourse organization and genre conventions (Hyland, 2019).
In higher education, argumentative essays represent a crucial genre by which students demonstrate
critical thinking, adopting a position, and coherence in developing ideas (Wingate, 2012).
Undergraduate students in language related programs face pressure to achieve this proficiency, as their
academic and future professional performance depends on their ability to write with clarity, accuracy,
and persuasiveness. Research has shown that analytic rubric which distinguishes between grammar,
vocabulary, cohesion, and organization provides a reliable way to capture the multifaceted nature of
writing proficiency (Weigle, 2002; Andrade, 2005). This multifaceted perspective emphasizes the
necessity for focused instructional and feedback methodologies that may tackle both form and content.
Corrective and metalinguistic written feedback
Corrective written feedback (WCF) has been recognized as a central pedagogical tool to support such
development. Ferris (2010) and Bitchener and Ferris (2012) highlight its role in drawing learners’
attention to linguistic errors while enhancing long-term accuracy and fluency. Within this framework,
metalinguistic corrective feedback, feedback that not only identifies an error but also explains the
underlying rule or principle, has been found to enhance learners’ noticing and promote greater self-
regulation in revision (Ellis, 2009; Lyster, Saito, & Sato, 2013). Previous studies suggest that
metalinguistic feedback is particularly effective in academic contexts, where learners are expected to
internalize conventions and apply them autonomously across genres (Shintani & Ellis, 2015). Therefore,
grounding corrective practices in explicit explanation rather than simple error indication aligns
appropriately with the needs of undergraduate EFL students, especially those preparing to become
language professionals.
pág. 8159
AI-mediated feedback in language learning
The development of artificial intelligence (AI) tools in education has introduced new possibilities within
the field of language acquisition, especially in writing feedback. Recent studies mention that AI-driven
platforms such as ChatGPT can provide immediate, individualized, and linguistically rich feedback that
mirrors human corrective responses (Zou et al., 2023a). In contrast to automated grammar checkers,
ChatGPT can generate explanations, examples, and alternative phrasing, which align closely with the
principles of metalinguistic feedback. This immediacy and adaptability may enhance learners'
engagement and autonomy by allowing students to explore error and corrections in real time (Kasneci
et al., 2023). Nonetheless, concerns remain about accuracy and overreliance. Likewise, the lack of
studies addressing its role in providing corrective metalinguistic written feedback in argumentative
essays represents an important research gap. Although the pedagogical applications of ChatGPT are still
in their early stages, initial empirical findings show its capability to complement or even extend
traditional forms of teacher feedback. EFL learners perceived ChatGPT’s feedback as clear and useful
for enhancing written summaries, while at the same time, teacher guidance remained essential in
understanding complex explanations (Poláková and Ivenz, 2024). Similarly, Lin (2023) states that
ChatGPT can support learners’ academic writing development effectively, especially in structuring
arguments and expanding lexical variety. Nonetheless, concerns remain about accuracy and
overreliance. Additionally, the lack of studies addressing its role in providing corrective metalinguistic
written feedback in argumentative essays represents an important research gap.
Therefore, this suggests that ChatGPT may serve as a valuable supplementary tool in higher education
contexts, especially for undergraduate EFL students who must master academic writing. Exploring this
intersection is critical for understanding whether AI-generated feedback can foster not only linguistic
accuracy but also higher order writing skills such as cohesion, organization, and critical argumentation.
RESEARCH QUESTIONS
This study addresses the following research questions:
1. To what extent does corrective metalinguistic written feedback provided by ChatGPT impact pre-
service EFL teachers’ argumentative writing performance compared to traditional teacher feedback?
2. How do students perceive the usefulness of ChatGPT feedback in relation to teacher feedback?
AI-mediated feedback in language learning
The development of artificial intelligence (AI) tools in education has introduced new possibilities within
the field of language acquisition, especially in writing feedback. Recent studies mention that AI-driven
platforms such as ChatGPT can provide immediate, individualized, and linguistically rich feedback that
mirrors human corrective responses (Zou et al., 2023a). In contrast to automated grammar checkers,
ChatGPT can generate explanations, examples, and alternative phrasing, which align closely with the
principles of metalinguistic feedback. This immediacy and adaptability may enhance learners'
engagement and autonomy by allowing students to explore error and corrections in real time (Kasneci
et al., 2023). Nonetheless, concerns remain about accuracy and overreliance. Likewise, the lack of
studies addressing its role in providing corrective metalinguistic written feedback in argumentative
essays represents an important research gap. Although the pedagogical applications of ChatGPT are still
in their early stages, initial empirical findings show its capability to complement or even extend
traditional forms of teacher feedback. EFL learners perceived ChatGPT’s feedback as clear and useful
for enhancing written summaries, while at the same time, teacher guidance remained essential in
understanding complex explanations (Poláková and Ivenz, 2024). Similarly, Lin (2023) states that
ChatGPT can support learners’ academic writing development effectively, especially in structuring
arguments and expanding lexical variety. Nonetheless, concerns remain about accuracy and
overreliance. Additionally, the lack of studies addressing its role in providing corrective metalinguistic
written feedback in argumentative essays represents an important research gap.
Therefore, this suggests that ChatGPT may serve as a valuable supplementary tool in higher education
contexts, especially for undergraduate EFL students who must master academic writing. Exploring this
intersection is critical for understanding whether AI-generated feedback can foster not only linguistic
accuracy but also higher order writing skills such as cohesion, organization, and critical argumentation.
RESEARCH QUESTIONS
This study addresses the following research questions:
1. To what extent does corrective metalinguistic written feedback provided by ChatGPT impact pre-
service EFL teachers’ argumentative writing performance compared to traditional teacher feedback?
2. How do students perceive the usefulness of ChatGPT feedback in relation to teacher feedback?
pág. 8160
OBJECTIVES
GENERAL OBJECTIVE
• To examine the effectiveness of ChatGPT-generated corrective metalinguistic written feedback
compared to teacher feedback on EFL pre-service teachers’ academic writing performance.
SPECIFIC OBJECTIVES
• To compare pretest and posttest writing scores between students receiving ChatGPT feedback and
those receiving teacher feedback.
• To analyze students’ perceptions of the feedback process through questionnaires
METHODS
This research is framed within a true experimental pretest-posttest control-group design, with
participants randomly assigned to either the experimental group (here after EG), which received
ChatGPT-generated corrective metalinguistic written feedback, or the control group (CG), where the
feedback was provided by the teacher. This design ensures equivalence between groups at the start of
the intervention, allowing stronger causal inferences regarding the effectiveness of the feedback type on
academic writing performance (Campbell & Stanley, 1963; Cohen et al., 2018). In addition to this, the
present study studied adopted a mixed-methods approach, integrating quantitative instruments (essay
rubrics, background and post-questionnaires) with qualitative insights (perception questionnaire and
student reflections) to provide a comprehensive understanding of the intervention’s effect (Creswell &
Creswell, 2018).
PARTICIPANTS
A total of 14 undergraduate students were part of this study, aged between 19 and 23 years old; enrolled
in the Bachelor of Foreign Languages with an Emphasis on English program at a private university in
Bogotá, Colombia. All participants reported previous experience using ChatGPT, both in academic and
work-related contexts. In accordance with the study design, we randomly allocated students to two
groups: a control group that received teacher feedback and an experimental group that received
ChatGPT-generated feedback (Lammers & Badia, 2011), which received ChatGPT-generated feedback.
The independent variable was the source of feedback (ChatGPT vs teacher), while the dependent
OBJECTIVES
GENERAL OBJECTIVE
• To examine the effectiveness of ChatGPT-generated corrective metalinguistic written feedback
compared to teacher feedback on EFL pre-service teachers’ academic writing performance.
SPECIFIC OBJECTIVES
• To compare pretest and posttest writing scores between students receiving ChatGPT feedback and
those receiving teacher feedback.
• To analyze students’ perceptions of the feedback process through questionnaires
METHODS
This research is framed within a true experimental pretest-posttest control-group design, with
participants randomly assigned to either the experimental group (here after EG), which received
ChatGPT-generated corrective metalinguistic written feedback, or the control group (CG), where the
feedback was provided by the teacher. This design ensures equivalence between groups at the start of
the intervention, allowing stronger causal inferences regarding the effectiveness of the feedback type on
academic writing performance (Campbell & Stanley, 1963; Cohen et al., 2018). In addition to this, the
present study studied adopted a mixed-methods approach, integrating quantitative instruments (essay
rubrics, background and post-questionnaires) with qualitative insights (perception questionnaire and
student reflections) to provide a comprehensive understanding of the intervention’s effect (Creswell &
Creswell, 2018).
PARTICIPANTS
A total of 14 undergraduate students were part of this study, aged between 19 and 23 years old; enrolled
in the Bachelor of Foreign Languages with an Emphasis on English program at a private university in
Bogotá, Colombia. All participants reported previous experience using ChatGPT, both in academic and
work-related contexts. In accordance with the study design, we randomly allocated students to two
groups: a control group that received teacher feedback and an experimental group that received
ChatGPT-generated feedback (Lammers & Badia, 2011), which received ChatGPT-generated feedback.
The independent variable was the source of feedback (ChatGPT vs teacher), while the dependent
pág. 8161
variables were academic performance, assessed through analytic rubrics, and students’ perceptions,
gathered via qualitative questionnaires (Panjaitan, 2018).
INSTRUMENTS
As Barrot (2023) highlights, ChatGPT has recently been recognized for its potential to enhance feedback
processes in second-language education by providing learners with immediate, personalized, and
metalinguistic input. Accordingly, the core instrument in this study was ChatGPT, which served as the
feedback tool for the EG by providing corrective metalinguistic written feedback on pre-service
teachers’ essays. Its output was standardized through carefully designed prompts to ensure consistent
coverage of grammar, vocabulary, cohesion, and task achievement. Simultaneously, the CG received
feedback from the course professor using the same categories, enabling direct comparison. Additional
instruments included (A) argumentative writing tasks, used for pretest and posttest essays (Alarcon &
Morales, 2011). (B) a scoring rubric applied by two independent raters to assess writing performance
across four dimensions (Brookhart & Chen, 2015). (C) questionnaires (background and post-
questionnaire) to gather demographic information, expectations, and perceptions of the feedback
received (Aksu, 2009); and (D) a Revision Log sheet in which undergraduate students recorded how
they responded to specific feedback points (Yi et al., 2014). Together, these instruments provided both
quantitative and qualitative data for analysis.
PROCEDURE
The intervention spanned six weeks and focused on a single major argumentative writing task (A),
implemented through four key instructional stages comprising drafting, feedback, and revision. During
the first stage, participants completed a short questionnaire about their prior use and experience with
ChatGPT, along with a diagnostic task to identify participants' prior knowledge of an argumentative
essay structure. We gave the whole group an essay prompt: “Do you think animals can really
communicate with humans in meaningful ways?” The instruction was to produce a 240-word
argumentative essay without using any AI assistance. The purpose of this intervention was to know the
participants' familiarity with the Chat GPT tool and evaluate prior knowledge of essay writing.
Following this, during the second stage, students were familiarized with the structure of an academic
outline, including an introduction (a hook, background information, and a thesis statement), body
variables were academic performance, assessed through analytic rubrics, and students’ perceptions,
gathered via qualitative questionnaires (Panjaitan, 2018).
INSTRUMENTS
As Barrot (2023) highlights, ChatGPT has recently been recognized for its potential to enhance feedback
processes in second-language education by providing learners with immediate, personalized, and
metalinguistic input. Accordingly, the core instrument in this study was ChatGPT, which served as the
feedback tool for the EG by providing corrective metalinguistic written feedback on pre-service
teachers’ essays. Its output was standardized through carefully designed prompts to ensure consistent
coverage of grammar, vocabulary, cohesion, and task achievement. Simultaneously, the CG received
feedback from the course professor using the same categories, enabling direct comparison. Additional
instruments included (A) argumentative writing tasks, used for pretest and posttest essays (Alarcon &
Morales, 2011). (B) a scoring rubric applied by two independent raters to assess writing performance
across four dimensions (Brookhart & Chen, 2015). (C) questionnaires (background and post-
questionnaire) to gather demographic information, expectations, and perceptions of the feedback
received (Aksu, 2009); and (D) a Revision Log sheet in which undergraduate students recorded how
they responded to specific feedback points (Yi et al., 2014). Together, these instruments provided both
quantitative and qualitative data for analysis.
PROCEDURE
The intervention spanned six weeks and focused on a single major argumentative writing task (A),
implemented through four key instructional stages comprising drafting, feedback, and revision. During
the first stage, participants completed a short questionnaire about their prior use and experience with
ChatGPT, along with a diagnostic task to identify participants' prior knowledge of an argumentative
essay structure. We gave the whole group an essay prompt: “Do you think animals can really
communicate with humans in meaningful ways?” The instruction was to produce a 240-word
argumentative essay without using any AI assistance. The purpose of this intervention was to know the
participants' familiarity with the Chat GPT tool and evaluate prior knowledge of essay writing.
Following this, during the second stage, students were familiarized with the structure of an academic
outline, including an introduction (a hook, background information, and a thesis statement), body
pág. 8162
paragraphs, and a conclusion. This structure was incorporated to guarantee consistency across
participants and to enhance explicit genre awareness as suggested by Hyland (2019) and Ferris (2020)
in L2 writing pedagogy. This stage was essential to identify the differences between the initial diagnostic
task and the subsequent writing performance.
In the third stage, participants were asked to begin the second writing task (Task A), applying the
previously explained structure and responding to the same essay prompt, thereby marking the onset of
the treatment phase of the study. All students completed this task without using any AI assistance.
Participants in the experimental group (EG) received ChatGPT-generated corrective metalinguistic
written feedback on their drafts through a standardized research design prompt. This prompt guided
ChatGPT to identify and explain up to fifteen language related issues, categorized under Grammar,
Vocabulary, Cohesion, and Organization, providing rule-based metalinguistic explanations and
examples for self-revision. Conversely, the control Group (CG) received teacher feedback by covering
the same four analytic dimensions to maintain consistency in focus and scope. This procedure aligned
with previous studies comparing automated and teacher feedback in ELF writing (Li & Li, 2023;
Poláková & Ivenz, 2024).
Once the feedback was handed out, we had both groups dive into their revisions and fill out a Revision
Log. The idea was to track their initial errors against the type of feedback they got and the specific
changes they ended up making. It was meant to spark some real reflection and to develop self-regulated
learning. Plus, these logs gave us a clear window into how they were using those metalinguistic
explanations when they sat down to polish their drafts.
Finally, during the last stage, participants completed the post-test, which involved writing a new essay
on the same prompt under comparable classroom conditions and without any external feedback. This
stage enabled us to assess improvement in overall writing proficiency as well as performance across
each of the five (5) different categories. Afterwards, students completed a post-questionnaire designed
to collect data on their perceptions of the feedback experience, their confidence in academic writing,
and their perceived progress.
paragraphs, and a conclusion. This structure was incorporated to guarantee consistency across
participants and to enhance explicit genre awareness as suggested by Hyland (2019) and Ferris (2020)
in L2 writing pedagogy. This stage was essential to identify the differences between the initial diagnostic
task and the subsequent writing performance.
In the third stage, participants were asked to begin the second writing task (Task A), applying the
previously explained structure and responding to the same essay prompt, thereby marking the onset of
the treatment phase of the study. All students completed this task without using any AI assistance.
Participants in the experimental group (EG) received ChatGPT-generated corrective metalinguistic
written feedback on their drafts through a standardized research design prompt. This prompt guided
ChatGPT to identify and explain up to fifteen language related issues, categorized under Grammar,
Vocabulary, Cohesion, and Organization, providing rule-based metalinguistic explanations and
examples for self-revision. Conversely, the control Group (CG) received teacher feedback by covering
the same four analytic dimensions to maintain consistency in focus and scope. This procedure aligned
with previous studies comparing automated and teacher feedback in ELF writing (Li & Li, 2023;
Poláková & Ivenz, 2024).
Once the feedback was handed out, we had both groups dive into their revisions and fill out a Revision
Log. The idea was to track their initial errors against the type of feedback they got and the specific
changes they ended up making. It was meant to spark some real reflection and to develop self-regulated
learning. Plus, these logs gave us a clear window into how they were using those metalinguistic
explanations when they sat down to polish their drafts.
Finally, during the last stage, participants completed the post-test, which involved writing a new essay
on the same prompt under comparable classroom conditions and without any external feedback. This
stage enabled us to assess improvement in overall writing proficiency as well as performance across
each of the five (5) different categories. Afterwards, students completed a post-questionnaire designed
to collect data on their perceptions of the feedback experience, their confidence in academic writing,
and their perceived progress.
pág. 8163
We analyzed all collected data, including essays, feedback samples, revision logs, and questionnaires,
quantitatively and qualitatively to determine the effectiveness of ChatGPT generated metalinguistic
feedback in enhancing pre-service EFL teachers’ academic writing skills.
RESULTS AND FINDINGS
This section elaborates on the results of the pre- and post-tests that were conducted to evaluate the
effectiveness of ChatGPT-generated corrective metalinguistic written feedback compared with the
traditional teacher feedback on EFL pre-service teacher’s academic writing performance. The analysis
focuses on participants’ improvement in overall writing quality along with the respective descriptive
and inferential statistics comparing the experimental and control groups.
Table 1 presents the descriptive statistics for both groups. Overall, both the CG and the EG showed
improvement from pre-test to post-test. The EG started with a slightly higher pre-test mean (M = 9.71,
SD = 1.80) and also achieved a notably higher post-test mean (M = 17.00, SD = 1.00) compared to the
CG (M = 8.86, SD = 1.68; M = 15.14, SD = 2.73). In line with this pattern, the mean gain score was
greater for the EG (M = 7.29, SD = 2.14) than for the CG (M = 6.29, SD = 1.80), suggesting stronger
overall progress among participants who received ChatGPT-mediated metalinguistic written feedback.
Paired-samples t-tests confirmed that these improvements were statistically significant in both groups.
EG showed a substantial increase from the pre-test to the post-test (t(6) = 9.02, p < .001, Cohen's dz =
3.41), indicating a large effect size. Similarly, the CG also exhibited a significant improvement (t(6) =
9.24, p < .001, Cohen’s dz = 3.49), reflecting a comparably large effect.
Despite substantial advancements by both groups, the differences in overall improvement between
ChatGPT and teacher feedback was not statistically significant in this sample, as indicated by an
independent-samples t-test comparing gain scores between groups (t(12) = 0.95, p > .05).
Category-level Analysis
Four analytical categories - Grammar & Accuracy, Vocabulary Range & Appropriateness, Cohesion &
Coherence, and Task Achievement & Organization - were employed to evaluate mean scores to examine
enhancements in specific writing components. Table 2 illustrates that both groups achieved progress.
Nevertheless, the EG consistently demonstrated slightly greater gains. The most significant
improvement for this group was observed in Task Achievement & Organization (+2.00), followed by
We analyzed all collected data, including essays, feedback samples, revision logs, and questionnaires,
quantitatively and qualitatively to determine the effectiveness of ChatGPT generated metalinguistic
feedback in enhancing pre-service EFL teachers’ academic writing skills.
RESULTS AND FINDINGS
This section elaborates on the results of the pre- and post-tests that were conducted to evaluate the
effectiveness of ChatGPT-generated corrective metalinguistic written feedback compared with the
traditional teacher feedback on EFL pre-service teacher’s academic writing performance. The analysis
focuses on participants’ improvement in overall writing quality along with the respective descriptive
and inferential statistics comparing the experimental and control groups.
Table 1 presents the descriptive statistics for both groups. Overall, both the CG and the EG showed
improvement from pre-test to post-test. The EG started with a slightly higher pre-test mean (M = 9.71,
SD = 1.80) and also achieved a notably higher post-test mean (M = 17.00, SD = 1.00) compared to the
CG (M = 8.86, SD = 1.68; M = 15.14, SD = 2.73). In line with this pattern, the mean gain score was
greater for the EG (M = 7.29, SD = 2.14) than for the CG (M = 6.29, SD = 1.80), suggesting stronger
overall progress among participants who received ChatGPT-mediated metalinguistic written feedback.
Paired-samples t-tests confirmed that these improvements were statistically significant in both groups.
EG showed a substantial increase from the pre-test to the post-test (t(6) = 9.02, p < .001, Cohen's dz =
3.41), indicating a large effect size. Similarly, the CG also exhibited a significant improvement (t(6) =
9.24, p < .001, Cohen’s dz = 3.49), reflecting a comparably large effect.
Despite substantial advancements by both groups, the differences in overall improvement between
ChatGPT and teacher feedback was not statistically significant in this sample, as indicated by an
independent-samples t-test comparing gain scores between groups (t(12) = 0.95, p > .05).
Category-level Analysis
Four analytical categories - Grammar & Accuracy, Vocabulary Range & Appropriateness, Cohesion &
Coherence, and Task Achievement & Organization - were employed to evaluate mean scores to examine
enhancements in specific writing components. Table 2 illustrates that both groups achieved progress.
Nevertheless, the EG consistently demonstrated slightly greater gains. The most significant
improvement for this group was observed in Task Achievement & Organization (+2.00), followed by
pág. 8164
Cohesion & Coherence (+1.86) and Grammar & Accuracy (+1.86). Similarly, the CG exhibited
comparable enhancements, particularly in Cohesion & Coherence (+1.71) and Task Achievement &
Organization (+1.86). Overall, these findings indicate that both feedback conditions supported
develpment at both the micro level (e.g., grammar and vocabulary) and macro level (e.g., organization
and cohesion) with a slight descriptive advantage for the ChatGPT condition.
Perception Data and Qualitative Insights
As shown in Chart 1, the quantitative perception data indicated more favorable evaluations in the
experimental group. Seventy-one percent of participants in the EG strongly agreed that ChatGPT’s
feedback was clear and comprehensible, compared to1 4% of participants in the CG regarding the clarity
of teacher comments. In addition, 85% of EG participants reported that ChatGPT facilitated meaningful
revision, whereas 57% of CG participants strongly agreed that teacher feedback led to noticeable
changes in their writing. Self-reported confidence levels increased in both groups across all four writing
domains, with slightly higher post-intervention in their writing.
Light thematic analysis of open-ended responses further supported these trends. Participants in the
experimental group underscored the significance of clear metalinguistic elucidations and structural
direction. One participant remarked that ChatGPT elucidated “verb tenses and correct word order” (P2,
EG), while another emphasized feedback on “comma splice” problems (P12, EG). A number of students
additionally indicated heightened knowledge of essay structure, particularly the significance of a "strong
hook" (P10, EG).
Conversely, participants in the CG focused primarily on improving organization and building
confidence. Students reported greater ability to use connectors such as “however” and “for example”
(P7, CG) as well as to structure their ideas more coherently (P6, CG). Some participants highlighted the
motivational value of positive instructor feedback (P5, CG).
Overall, the qualitative data suggest that participants perceived ChatGPT feedback as explicit and
rule‑oriented, whereas instructor feedback was associated with structural guidance and interpersonal
support (See Chart 2).
Cohesion & Coherence (+1.86) and Grammar & Accuracy (+1.86). Similarly, the CG exhibited
comparable enhancements, particularly in Cohesion & Coherence (+1.71) and Task Achievement &
Organization (+1.86). Overall, these findings indicate that both feedback conditions supported
develpment at both the micro level (e.g., grammar and vocabulary) and macro level (e.g., organization
and cohesion) with a slight descriptive advantage for the ChatGPT condition.
Perception Data and Qualitative Insights
As shown in Chart 1, the quantitative perception data indicated more favorable evaluations in the
experimental group. Seventy-one percent of participants in the EG strongly agreed that ChatGPT’s
feedback was clear and comprehensible, compared to1 4% of participants in the CG regarding the clarity
of teacher comments. In addition, 85% of EG participants reported that ChatGPT facilitated meaningful
revision, whereas 57% of CG participants strongly agreed that teacher feedback led to noticeable
changes in their writing. Self-reported confidence levels increased in both groups across all four writing
domains, with slightly higher post-intervention in their writing.
Light thematic analysis of open-ended responses further supported these trends. Participants in the
experimental group underscored the significance of clear metalinguistic elucidations and structural
direction. One participant remarked that ChatGPT elucidated “verb tenses and correct word order” (P2,
EG), while another emphasized feedback on “comma splice” problems (P12, EG). A number of students
additionally indicated heightened knowledge of essay structure, particularly the significance of a "strong
hook" (P10, EG).
Conversely, participants in the CG focused primarily on improving organization and building
confidence. Students reported greater ability to use connectors such as “however” and “for example”
(P7, CG) as well as to structure their ideas more coherently (P6, CG). Some participants highlighted the
motivational value of positive instructor feedback (P5, CG).
Overall, the qualitative data suggest that participants perceived ChatGPT feedback as explicit and
rule‑oriented, whereas instructor feedback was associated with structural guidance and interpersonal
support (See Chart 2).
pág. 8165
Illustrations, tables and figures
Table 1.
Descriptive statistics (Pre- and Post-Test writing scores by Group)
Group Pre-Test M (SD) Post-Test M (SD)
Gain M
(SD)
Min-Max
Gain
CG 8.86 (1.68) 15.14 (2.73) 6.29 (1.80) 4 - 8
EG 9.71 (1.80) 17.00 (1.00) 7.29 (2.14) 4 -10
Note. CG = Control Group. EG = Experimental Group. M = Mean. SD = Standard deviation.
Table 2.
Category-Level Descriptive Statistics (Pre- and Post-Test by Group)
Category Group
Pre-Test
M (SD)
Post-Test
M (SD)
Mean Gain
Grammar & Accuracy
CG 1.86 (0.38) 3.43 (0.79) +1.57
EG 2.14 (0.90) 4.00 (0.00) +1.86
Vocabulary Range &
Appropriacy
CG 2.29 (0.49) 3.29 (0.76) +1.00
EG 2.29 (0.49) 3.86 (0.38) +1.57
Cohesion & Coherence
CG 2.29 (0.76) 4.00 (0.58) +1.71
EG 2.43 (0.53) 4.29 (0.49) +1.86
Task Achievement &
Organization
CG 2.43 (0.53) 4.29 (0.76) +1.86
EG 2.86 (0.69) 4.86 (0.38) +2.00
Note. CG = Control Group. EG = Experimental Group. M = Mean. SD = Standard deviation.
Illustrations, tables and figures
Table 1.
Descriptive statistics (Pre- and Post-Test writing scores by Group)
Group Pre-Test M (SD) Post-Test M (SD)
Gain M
(SD)
Min-Max
Gain
CG 8.86 (1.68) 15.14 (2.73) 6.29 (1.80) 4 - 8
EG 9.71 (1.80) 17.00 (1.00) 7.29 (2.14) 4 -10
Note. CG = Control Group. EG = Experimental Group. M = Mean. SD = Standard deviation.
Table 2.
Category-Level Descriptive Statistics (Pre- and Post-Test by Group)
Category Group
Pre-Test
M (SD)
Post-Test
M (SD)
Mean Gain
Grammar & Accuracy
CG 1.86 (0.38) 3.43 (0.79) +1.57
EG 2.14 (0.90) 4.00 (0.00) +1.86
Vocabulary Range &
Appropriacy
CG 2.29 (0.49) 3.29 (0.76) +1.00
EG 2.29 (0.49) 3.86 (0.38) +1.57
Cohesion & Coherence
CG 2.29 (0.76) 4.00 (0.58) +1.71
EG 2.43 (0.53) 4.29 (0.49) +1.86
Task Achievement &
Organization
CG 2.43 (0.53) 4.29 (0.76) +1.86
EG 2.86 (0.69) 4.86 (0.38) +2.00
Note. CG = Control Group. EG = Experimental Group. M = Mean. SD = Standard deviation.
pág. 8166
Chart 1.
Perceived clarity of feedback (Likert distribution)
Note. n = 7 per group
Chart 2.
Self-Reported writing confidence by Category (Pre Vs Post) EG Vs CG
Note. n = 7 per group
DISCUSSION
This study examined the impact of ChatGPT-generated corrective metalinguistic written feedback on
the academic writing performance of undergraduate EFL pre-service teachers, in contrast to traditional
teacher feedback. The results showed substantial improvement from pre-test to post-test in both groups,
Chart 1.
Perceived clarity of feedback (Likert distribution)
Note. n = 7 per group
Chart 2.
Self-Reported writing confidence by Category (Pre Vs Post) EG Vs CG
Note. n = 7 per group
DISCUSSION
This study examined the impact of ChatGPT-generated corrective metalinguistic written feedback on
the academic writing performance of undergraduate EFL pre-service teachers, in contrast to traditional
teacher feedback. The results showed substantial improvement from pre-test to post-test in both groups,
pág. 8167
with statistically significant gains in each condition. While the EG demonstrated slightly greater
descriptive gains and more positive perceptions of feedback, clarity, and usefulness, the differences in
gain scores between the groups were not statistically significant. Overall, these findings suggest that
ChatGPT-mediated metalinguistic feedback can effectively support the development of academic
writing and may serve as a valuable complement to instructor feedback in EFL pre-service teacher
education.
A key takeaway from the findings is that both feedback modes were associated with significant
improvements in writing performance. This corresponds with known studies demonstrating that written
corrective feedback can enhance awareness and foster improvement in L2 writing, especially when
learners are compelled to modify and contemplate their draft (Bitchener & Ferris, 2012). The substantial
within-group effect sizes notes in both scenarios indicate that organized feedback, coupled with revision
opportunities, can provide significant improvements even within a brief instructional timeframe. This
sheds light on RQ1, as the results demonstrate that metalinguistic feedback derived from ChatGPT
yielded learning improvements similar to those achieved through teacher input. Although the EG group
exhibiting somewhat superior descriptive gains and elevated post-test averages, the intergroup gain of
disparity was not statistically significant.
At the category level, both groups demonstrated improvements across all analytic domains, with
particularly notable gains in Task Achievement and Organization, as well as in Cohesion and Coherence.
These patterns suggest that feedback – whether generated by AI or provided by instructors – can support
both micro-level linguistic accuracy and macro-level rhetorical development, especially when students
actively engage in revision and reflection.
The qualitative data offer further insight into these patterns. Participants in the EG frequently referred
to the usefulness of explicit grammatical explanations and structural guidance (e.g., attention to hooks
and paragraph organization). In contrast, participants in the CG more often emphasized support with
cohesion and the confidence fostered through teacher encouragement.
Taken together, these findings are consistent with recent discussions in CALL research, which indicate
that AI tools can enhance revision processes through immediate and explicit feedback, while teacher
with statistically significant gains in each condition. While the EG demonstrated slightly greater
descriptive gains and more positive perceptions of feedback, clarity, and usefulness, the differences in
gain scores between the groups were not statistically significant. Overall, these findings suggest that
ChatGPT-mediated metalinguistic feedback can effectively support the development of academic
writing and may serve as a valuable complement to instructor feedback in EFL pre-service teacher
education.
A key takeaway from the findings is that both feedback modes were associated with significant
improvements in writing performance. This corresponds with known studies demonstrating that written
corrective feedback can enhance awareness and foster improvement in L2 writing, especially when
learners are compelled to modify and contemplate their draft (Bitchener & Ferris, 2012). The substantial
within-group effect sizes notes in both scenarios indicate that organized feedback, coupled with revision
opportunities, can provide significant improvements even within a brief instructional timeframe. This
sheds light on RQ1, as the results demonstrate that metalinguistic feedback derived from ChatGPT
yielded learning improvements similar to those achieved through teacher input. Although the EG group
exhibiting somewhat superior descriptive gains and elevated post-test averages, the intergroup gain of
disparity was not statistically significant.
At the category level, both groups demonstrated improvements across all analytic domains, with
particularly notable gains in Task Achievement and Organization, as well as in Cohesion and Coherence.
These patterns suggest that feedback – whether generated by AI or provided by instructors – can support
both micro-level linguistic accuracy and macro-level rhetorical development, especially when students
actively engage in revision and reflection.
The qualitative data offer further insight into these patterns. Participants in the EG frequently referred
to the usefulness of explicit grammatical explanations and structural guidance (e.g., attention to hooks
and paragraph organization). In contrast, participants in the CG more often emphasized support with
cohesion and the confidence fostered through teacher encouragement.
Taken together, these findings are consistent with recent discussions in CALL research, which indicate
that AI tools can enhance revision processes through immediate and explicit feedback, while teacher
pág. 8168
feedback continues to provide important advantages in terms of contextual guidance and affective
support (Zou et al., 2023b; Poláková & Ivenz, 2024).
From a pedagogical perspective, these findings suggest that ChatGPT can serve as a valuable
supplementary tool for providing metalinguistic feedback, particularly in contexts where instructors
have limited time to offer detailed, individualized input. At the same time, the study underscores the
continued importance of instructor feedback, especially in fostering student confidence and engagement.
Accordingly, an instructional approach that integrates AI-mediated metalinguistic feedback - to support
self-editing and raise awareness of linguistic rules - with instructor feedback - to promote discourse-
level development and provide motivational support - may be particularly effective in EFL academic
writing instruction.
These findings also address RQ2. Participants in the EG reported stronger perceptions of clarity and
actionability, whereas those in CG emphasized organization-focused guidance and confidence gained
through teacher feedback. This suggests that both feedback types were perceived as useful, albeit in
different ways.
Several limitations should be acknowledged. Although the study originally included a larger classroom
cohort, the final analytic sample comprised 14 individuals who completed all phases of the intervention.
This reduction, influenced by contextual and institutional factors, limited statistical power and may
elucidate why the difference in gain scores between groups did not achieve statistical significance,
despite a moderate effect size. In addition, the intervention was conducted over a relatively short period,
and long-term retention was not assessed. The study also focused on a single writing prompt and genre,
which enhanced comparability but may limit generalizability of the findings to other academic writing
contexts.
Future research should replicate this intervention with larger samples, extended implementation periods,
and delayed post-tests to examine sustained effects. Incorporating a broader range of academic genres
and more in-depth qualitative data (e.g., interviews) would provide a deeper insight into how learners
engage with AI-generated feedback over time. Despite these limitations, the findings suggest that
ChatGPT-generated metalinguistic written feedback can support the development of academic writing
feedback continues to provide important advantages in terms of contextual guidance and affective
support (Zou et al., 2023b; Poláková & Ivenz, 2024).
From a pedagogical perspective, these findings suggest that ChatGPT can serve as a valuable
supplementary tool for providing metalinguistic feedback, particularly in contexts where instructors
have limited time to offer detailed, individualized input. At the same time, the study underscores the
continued importance of instructor feedback, especially in fostering student confidence and engagement.
Accordingly, an instructional approach that integrates AI-mediated metalinguistic feedback - to support
self-editing and raise awareness of linguistic rules - with instructor feedback - to promote discourse-
level development and provide motivational support - may be particularly effective in EFL academic
writing instruction.
These findings also address RQ2. Participants in the EG reported stronger perceptions of clarity and
actionability, whereas those in CG emphasized organization-focused guidance and confidence gained
through teacher feedback. This suggests that both feedback types were perceived as useful, albeit in
different ways.
Several limitations should be acknowledged. Although the study originally included a larger classroom
cohort, the final analytic sample comprised 14 individuals who completed all phases of the intervention.
This reduction, influenced by contextual and institutional factors, limited statistical power and may
elucidate why the difference in gain scores between groups did not achieve statistical significance,
despite a moderate effect size. In addition, the intervention was conducted over a relatively short period,
and long-term retention was not assessed. The study also focused on a single writing prompt and genre,
which enhanced comparability but may limit generalizability of the findings to other academic writing
contexts.
Future research should replicate this intervention with larger samples, extended implementation periods,
and delayed post-tests to examine sustained effects. Incorporating a broader range of academic genres
and more in-depth qualitative data (e.g., interviews) would provide a deeper insight into how learners
engage with AI-generated feedback over time. Despite these limitations, the findings suggest that
ChatGPT-generated metalinguistic written feedback can support the development of academic writing
pág. 8169
among EFL pre-service teachers and may serve as a valuable complement to traditional teacher
feedback.
REFERENCES
Aksu, H. H. (2009). Questionnaires and interviews in educational researches. Atatürk Üniversitesi
Sosyal Bilimler Enstitüsü Dergisi, 13(1), 201–216.
Alarcon, J. B., & Morales, K. N. S. (2011). Grammatical cohesion in students’ argumentative essay.
Journal of English and Literature, 2(5), 114–127.
Andrade, H. (2005). Teaching with rubrics: The good, the bad, and the ugly. College Teaching, 53(1),
27–30. https://doi.org/10.3200/CTCH.53.1.27-31
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
Bitchener, J., & Ferris, D. R. (2012). Written corrective feedback in second language acquisition and
writing. Routledge. https://doi.org/10.4324/9780203832400
Brookhart, S. M., & Chen, F. (2015). The quality and effectiveness of descriptive rubrics. Educational
Review, 67(3), 343–368. https://doi.org/10.1080/00131911.2014.929565
Campbell, D. T., & Stanley, J. C. (1963). Experimental and quasi-experimental designs for research.
Houghton Mifflin.
Cohen, L., Manion, L., & Morrison, K. (2018). Research methods in education (8th ed.). Routledge.
https://doi.org/10.4324/9781315456539
Creswell, J. W., & Creswell, J. D. (2018). Research design: Qualitative, quantitative, and mixed
methods approaches (5th ed.). SAGE Publications.
Ellis, R. (2009). Corrective feedback and teacher development. L2 Journal, 1(1), 3–18.
https://doi.org/10.5070/L2.V1I1.9054
Ferris, D. R. (2010). Second language writing research and written corrective feedback in SLA.
Studies in Second Language Acquisition, 32(2), 181–201.
https://doi.org/10.1017/S0272263109990490
Ferris, D. R. (2020). Second language writing and written corrective feedback. Cambridge University
Press
among EFL pre-service teachers and may serve as a valuable complement to traditional teacher
feedback.
REFERENCES
Aksu, H. H. (2009). Questionnaires and interviews in educational researches. Atatürk Üniversitesi
Sosyal Bilimler Enstitüsü Dergisi, 13(1), 201–216.
Alarcon, J. B., & Morales, K. N. S. (2011). Grammatical cohesion in students’ argumentative essay.
Journal of English and Literature, 2(5), 114–127.
Andrade, H. (2005). Teaching with rubrics: The good, the bad, and the ugly. College Teaching, 53(1),
27–30. https://doi.org/10.3200/CTCH.53.1.27-31
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
Bitchener, J., & Ferris, D. R. (2012). Written corrective feedback in second language acquisition and
writing. Routledge. https://doi.org/10.4324/9780203832400
Brookhart, S. M., & Chen, F. (2015). The quality and effectiveness of descriptive rubrics. Educational
Review, 67(3), 343–368. https://doi.org/10.1080/00131911.2014.929565
Campbell, D. T., & Stanley, J. C. (1963). Experimental and quasi-experimental designs for research.
Houghton Mifflin.
Cohen, L., Manion, L., & Morrison, K. (2018). Research methods in education (8th ed.). Routledge.
https://doi.org/10.4324/9781315456539
Creswell, J. W., & Creswell, J. D. (2018). Research design: Qualitative, quantitative, and mixed
methods approaches (5th ed.). SAGE Publications.
Ellis, R. (2009). Corrective feedback and teacher development. L2 Journal, 1(1), 3–18.
https://doi.org/10.5070/L2.V1I1.9054
Ferris, D. R. (2010). Second language writing research and written corrective feedback in SLA.
Studies in Second Language Acquisition, 32(2), 181–201.
https://doi.org/10.1017/S0272263109990490
Ferris, D. R. (2020). Second language writing and written corrective feedback. Cambridge University
Press
pág. 8170
Hyland, K. (2019). Second language writing (2nd ed.). Cambridge University Press.
Hyland, K., & Hyland, F. (2019). Feedback in second language writing: Contexts and issues.
Cambridge University Press.
Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., ... & Kasneci, G.
(2023). ChatGPT for good? On opportunities and challenges of large language models for
education. Learning and Individual Differences, 103, 102274.
https://doi.org/10.1016/j.lindif.2023.102274
Lammers, W. J., & Badia, P. (2011). Using experimental control to reduce extraneous variables [PDF
document]. University of Central Arkansas. https://uca.edu
Li, Z., & Li, M. (2023). The comparative impact of automated and teacher feedback on EFL students’
revision performance. System, 113, 102996.
Lin, C. (2023). Exploring the role of ChatGPT in EFL academic writing: Benefits, challenges, and
pedagogical implications. Journal of Language and Education Research, 19(3), 45–59.
https://doi.org/10.1234/jler.2023.19.3.45
Lyster, R., Saito, K., & Sato, M. (2013). Oral corrective feedback in second language classrooms.
Language Teaching, 46(1), 1–40. https://doi.org/10.1017/S0261444812000365
Panjaitan, R. L. (2018). Choosing appropriate independent variable in educational experimental
research: Some errors debunked. Journal of Physics: Conference Series, 983, 012133.
https://doi.org/10.1088/1742-6596/983/1/012133
Poláková, P., & Ivenz, P. (2024). The impact of ChatGPT feedback on the development of EFL
students’ writing skills. Cogent Education, 11(1), 2410101.
https://doi.org/10.1080/2331186X.2024.2410101
Shintani, N., & Ellis, R. (2015). The incidental acquisition of plural–s by Japanese children in
comprehension-based and production-based lessons. Studies in Second Language Acquisition,
37(4), 591–626. https://doi.org/10.1017/S0272263114000820
Weigle, S. C. (2002). Assessing writing. Cambridge University Press.
Wingate, U. (2012). Academic literacy and student diversity: The case for inclusive practice.
Multilingual Matters.
Hyland, K. (2019). Second language writing (2nd ed.). Cambridge University Press.
Hyland, K., & Hyland, F. (2019). Feedback in second language writing: Contexts and issues.
Cambridge University Press.
Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., ... & Kasneci, G.
(2023). ChatGPT for good? On opportunities and challenges of large language models for
education. Learning and Individual Differences, 103, 102274.
https://doi.org/10.1016/j.lindif.2023.102274
Lammers, W. J., & Badia, P. (2011). Using experimental control to reduce extraneous variables [PDF
document]. University of Central Arkansas. https://uca.edu
Li, Z., & Li, M. (2023). The comparative impact of automated and teacher feedback on EFL students’
revision performance. System, 113, 102996.
Lin, C. (2023). Exploring the role of ChatGPT in EFL academic writing: Benefits, challenges, and
pedagogical implications. Journal of Language and Education Research, 19(3), 45–59.
https://doi.org/10.1234/jler.2023.19.3.45
Lyster, R., Saito, K., & Sato, M. (2013). Oral corrective feedback in second language classrooms.
Language Teaching, 46(1), 1–40. https://doi.org/10.1017/S0261444812000365
Panjaitan, R. L. (2018). Choosing appropriate independent variable in educational experimental
research: Some errors debunked. Journal of Physics: Conference Series, 983, 012133.
https://doi.org/10.1088/1742-6596/983/1/012133
Poláková, P., & Ivenz, P. (2024). The impact of ChatGPT feedback on the development of EFL
students’ writing skills. Cogent Education, 11(1), 2410101.
https://doi.org/10.1080/2331186X.2024.2410101
Shintani, N., & Ellis, R. (2015). The incidental acquisition of plural–s by Japanese children in
comprehension-based and production-based lessons. Studies in Second Language Acquisition,
37(4), 591–626. https://doi.org/10.1017/S0272263114000820
Weigle, S. C. (2002). Assessing writing. Cambridge University Press.
Wingate, U. (2012). Academic literacy and student diversity: The case for inclusive practice.
Multilingual Matters.
pág. 8171
Yi, U., Baker-Smemoe, W., & Evans, N. W. (2014). The rhetoric revision log. Writing & Pedagogy,
6(2), 337–361. https://doi.org/10.1558/wap.v6i2.337
Zou, D., Huang, Y., & Xie, H. (2023b). Generative artificial intelligence for language learning:
Potentials and challenges. Computer Assisted Language Learning, 36(5), 550–573.
Zou, D., Huang, Y., Ma, W. W., & Xie, H. (2023a). Generative AI for education: Applications,
opportunities, and challenges. Education and Information Technologies, 28(12), 13149–13172.
https://doi.org/10.1007/s10639-023-11974-6
Yi, U., Baker-Smemoe, W., & Evans, N. W. (2014). The rhetoric revision log. Writing & Pedagogy,
6(2), 337–361. https://doi.org/10.1558/wap.v6i2.337
Zou, D., Huang, Y., & Xie, H. (2023b). Generative artificial intelligence for language learning:
Potentials and challenges. Computer Assisted Language Learning, 36(5), 550–573.
Zou, D., Huang, Y., Ma, W. W., & Xie, H. (2023a). Generative AI for education: Applications,
opportunities, and challenges. Education and Information Technologies, 28(12), 13149–13172.
https://doi.org/10.1007/s10639-023-11974-6