Vol. 13/ Núm. 3 2026 pág. 1506
https://doi.org/10.69639/arandu.v13i3.2485
The Impact of LuzIA on Academic Writing in English as
Foreign Language Learners
El impacto de LuzIA en la escritura académica de estudiantes de inglés como lengua
extranjera
Eleyvi Castellón Vladés
ecastellonv@ube.edu.ec
https://orcid.org/0009-0003-0481-6111
Universidad Bolivariana del Ecuador
Durán – Ecuador
Jaime David Sáenz Herdoiza
jdsaenzh@ube.edu.ec
https://orcid.org/0009-0003-7317-8563
Universidad Bolivariana del Ecuador
Durán – Ecuador
Josué Reinaldo Bonilla Tenesaca
jrbonillat@ube.edu.ec
https://orcid.org/0000-0002-6748-2345
Universidad Bolivariana del Ecuador
Durán – Ecuador
Jacqueline Elizabeth Lopéz Lopéz
jelopezl@ube.edu.ec
https://orcid.org/0000-0002-1765-8103
Universidad Bolivariana del Ecuador
Durán – Ecuador
Artículo recibido: 10 julio 2026- Aceptado para publicación:16 agosto 2026
Conflictos de intereses: Ninguno que declarar.
ABSTRACT
Academic writing is one of the most difficult skills for English as a Foreign Language (EFL)
learners to develop because it requires not only linguistic competence but also the ability to
organize ideas and achieve coherence. This challenge is particularly relevant in multilingual
environments, where learners must manage these demands simultaneously. This study analyzes
the influence of LuzIA, a GPT-powered conversational artificial intelligence assistant integrated
into WhatsApp, on academic writing, writing anxiety, and learner autonomy among EFL students
at a bilingual secondary school in Ecuador. A sequential explanatory mixed-methods design was
employed with 145 students from six intact classroom cohorts in Grades 8–12, whose expected
English proficiency ranged from CEFR B2 to C1 according to their curricular placement. LuzIA
was used for 12 weeks as a writing support tool in three English courses within the curriculum.
Writing performance was assessed using the analytic rubric developed by Jacobs et al. (1981),
which includes the higher-order components of Content and Organization and the lower-order
Vol. 13/ Núm. 3 2026 pág. 1507
components of Vocabulary, Grammar, and Mechanics. Writing anxiety (SLWAI), learner
autonomy, and the Technology Acceptance Model (TAM) were also assessed through an 18-item
questionnaire. The results showed a statistically significant increase in writing performance (ΔM
= 16.4 points; t (144) = 14.23, p < .001, d = 0.90). The largest gains were observed in Grammar
(d = 1.12) and Vocabulary (d = 1.05), while significant but smaller gains were found in Content
(d = 0.72) and Organization (d = 0.68). Writing anxiety decreased significantly (ΔM = −1.24; d
= 1.35), whereas learner autonomy increased (ΔM = +0.87; d = 0.92). TAM results were also
positive, with Perceived Ease of Use reaching 4.47/5 and Perceived Usefulness reaching 4.12/5.
The study concludes that the use of LuzIA may provide benefits for EFL academic writing when
it is used as a support tool rather than as a replacement for learners’ own writing. The strongest
changes were observed in language-level aspects of writing. In addition, accessing the tool
directly through WhatsApp may facilitate its implementation.
Keywords: LuzIA, EFL academic writing, GPT conversational AI, writing anxiety, learner
autonomy
RESUMEN
La escritura académica es la habilidad más difícil de desarrollar para los estudiantes de Inglés
como Lengua Extranjera (EFL) debido a la competencia lingüística requerida pero también por
la necesidad de organizar los pensamientos y lograr la coherencia. Este problema se acentúa
especialmente en un entorno multilingüe donde se juntan todos estos requerimientos. Este trabajo
analiza la influencia de LuzIA, un asistente de inteligencia artificial conversacional basado en la
tecnología GPT integrado en la aplicación WhatsApp en la escritura académica, la ansiedad por
la escritura y la autonomía del estudiante de EFL de una escuela secundaria bilingüe de Ecuador.
En esta investigación se aplicó un diseño explicativo secuencial mixto y se seleccionaron 145
estudiantes de seis grupos completos de 8º a 12º grado que supuestamente tienen un nivel de
inglés entre B2 y C1 del MCER dependiendo de su ubicación curricular. LuzIA se utilizó durante
12 semanas como un ayudante en el proceso de escritura en tres cursos de inglés del plan de
estudios. Los resultados de la escritura fueron evaluados mediante la rúbrica analítica de Jacobs
et al. (1981) la cual incluye componentes de alto nivel de contenido y organización así como
elementos de bajo nivel de vocabulario, gramática y mecánica. Además se evaluó la ansiedad por
la escritura (SLWAI), la autonomía del estudiante y el Modelo de Aceptación de la Tecnología
(TAM) mediante un cuestionario de 18 ítems. Como resultado se encontró un aumento
estadísticamente significativo en el desempeño de la escritura (ΔM = 16.4 puntos; t(144) = 14.23,
p < .001, d = 0.90). Los mayores aumentos se observaron en la gramática (d = 1.12) y el
vocabulario (d = 1.05). Hubo ganancias significativas pero algo menores en el contenido (d =
0.72) y la organización (d = 0.68). Aunque hubo tal mejora la ansiedad por la escritura disminuyó
significativamente (ΔM = -1.24; d = 1.35) y la autonomía del estudiante aumentó positivamente
components of Vocabulary, Grammar, and Mechanics. Writing anxiety (SLWAI), learner
autonomy, and the Technology Acceptance Model (TAM) were also assessed through an 18-item
questionnaire. The results showed a statistically significant increase in writing performance (ΔM
= 16.4 points; t (144) = 14.23, p < .001, d = 0.90). The largest gains were observed in Grammar
(d = 1.12) and Vocabulary (d = 1.05), while significant but smaller gains were found in Content
(d = 0.72) and Organization (d = 0.68). Writing anxiety decreased significantly (ΔM = −1.24; d
= 1.35), whereas learner autonomy increased (ΔM = +0.87; d = 0.92). TAM results were also
positive, with Perceived Ease of Use reaching 4.47/5 and Perceived Usefulness reaching 4.12/5.
The study concludes that the use of LuzIA may provide benefits for EFL academic writing when
it is used as a support tool rather than as a replacement for learners’ own writing. The strongest
changes were observed in language-level aspects of writing. In addition, accessing the tool
directly through WhatsApp may facilitate its implementation.
Keywords: LuzIA, EFL academic writing, GPT conversational AI, writing anxiety, learner
autonomy
RESUMEN
La escritura académica es la habilidad más difícil de desarrollar para los estudiantes de Inglés
como Lengua Extranjera (EFL) debido a la competencia lingüística requerida pero también por
la necesidad de organizar los pensamientos y lograr la coherencia. Este problema se acentúa
especialmente en un entorno multilingüe donde se juntan todos estos requerimientos. Este trabajo
analiza la influencia de LuzIA, un asistente de inteligencia artificial conversacional basado en la
tecnología GPT integrado en la aplicación WhatsApp en la escritura académica, la ansiedad por
la escritura y la autonomía del estudiante de EFL de una escuela secundaria bilingüe de Ecuador.
En esta investigación se aplicó un diseño explicativo secuencial mixto y se seleccionaron 145
estudiantes de seis grupos completos de 8º a 12º grado que supuestamente tienen un nivel de
inglés entre B2 y C1 del MCER dependiendo de su ubicación curricular. LuzIA se utilizó durante
12 semanas como un ayudante en el proceso de escritura en tres cursos de inglés del plan de
estudios. Los resultados de la escritura fueron evaluados mediante la rúbrica analítica de Jacobs
et al. (1981) la cual incluye componentes de alto nivel de contenido y organización así como
elementos de bajo nivel de vocabulario, gramática y mecánica. Además se evaluó la ansiedad por
la escritura (SLWAI), la autonomía del estudiante y el Modelo de Aceptación de la Tecnología
(TAM) mediante un cuestionario de 18 ítems. Como resultado se encontró un aumento
estadísticamente significativo en el desempeño de la escritura (ΔM = 16.4 puntos; t(144) = 14.23,
p < .001, d = 0.90). Los mayores aumentos se observaron en la gramática (d = 1.12) y el
vocabulario (d = 1.05). Hubo ganancias significativas pero algo menores en el contenido (d =
0.72) y la organización (d = 0.68). Aunque hubo tal mejora la ansiedad por la escritura disminuyó
significativamente (ΔM = -1.24; d = 1.35) y la autonomía del estudiante aumentó positivamente
Vol. 13/ Núm. 3 2026 pág. 1508
(ΔM = +0.87; d = 0.92). Los resultados del TAM también fueron positivos: la puntuación de
usabilidad fue 4.47/5 y la puntuación de utilidad percibida fue 4.12/5. Concluyendo este estudio
se puede afirmar que el uso de LuzIA puede proporcionar ciertos beneficios a la escritura
académica en EFL pero solo si no se utiliza como un reemplazo sino como un ayudante en la
búsqueda de algún objetivo. Los cambios más significativos se hicieron en la parte lingüística de
la escritura. Además el hecho de usar el programa directamente en la aplicación WhatsApp hace
más fácil su implementación.
Palabras Clave: LuzIA, escritura académica EFL, IA conversacional GPT, ansiedad por la
escritura, autonomía del estudiante
Todo el contenido de la Revista Científica Internacional Arandu UTIC publicado en este sitio está disponible bajo
licencia Creative Commons Atribution 4.0 International.
(ΔM = +0.87; d = 0.92). Los resultados del TAM también fueron positivos: la puntuación de
usabilidad fue 4.47/5 y la puntuación de utilidad percibida fue 4.12/5. Concluyendo este estudio
se puede afirmar que el uso de LuzIA puede proporcionar ciertos beneficios a la escritura
académica en EFL pero solo si no se utiliza como un reemplazo sino como un ayudante en la
búsqueda de algún objetivo. Los cambios más significativos se hicieron en la parte lingüística de
la escritura. Además el hecho de usar el programa directamente en la aplicación WhatsApp hace
más fácil su implementación.
Palabras Clave: LuzIA, escritura académica EFL, IA conversacional GPT, ansiedad por la
escritura, autonomía del estudiante
Todo el contenido de la Revista Científica Internacional Arandu UTIC publicado en este sitio está disponible bajo
licencia Creative Commons Atribution 4.0 International.
Vol. 13/ Núm. 3 2026 pág. 1509
INTRODUCTION
English academic writing acts as the gateway for university entrance and worldwide
certification, and the learning of the skill among students whose native tongue is not English
(English as a Foreign Language students) becomes extremely complicated. Academic writing
requires learners to manage multiple demands simultaneously, including grammatical accuracy,
appropriate vocabulary, logical organization of ideas, genre conventions, and academic style. For
EFL learners, these demands may be particularly challenging because they are developing
academic writing skills while continuing to develop proficiency in the target language (Ghafar &
Raheem, 2025). There is one more challenge in the learning process of English academic writing.
Providing continuous individualized feedback in conventional classroom settings may be difficult
because of practical constraints such as limited instructional time (Sanchez, 2024).
Fortunately, one more tool for language development appeared due to the rapid progress in
artificial intelligence. AI-powered LLMs are capable of giving instant feedback on grammar,
vocabulary, organization, and other writing-related issues without any necessity of teacher
intervention during all steps. Recent research has provided growing evidence that AI-assisted
tools can support EFL writing development. Experimental and comparative studies have reported
improvements in different dimensions of writing following AI-assisted instruction or revision,
while systematic reviews have identified broader patterns of linguistic and cognitive scaffolding
across EFL contexts (Song & Song, 2023; Tsai et al., 2024; Li & Wilson, 2025). In particular,
AI-supported writing appears to offer useful assistance with language-related aspects such as
grammar and vocabulary, while also providing opportunities for revision, idea development, and
organization.
There are many different AI-assistive tools. One of the unique features of the LuzIA tool
is the fact that it can work in WhatsApp. It becomes important because of the fact that WhatsApp
is integrated in daily practices of many Latin American teenagers. As a result, it will be possible
to receive AI help using the application with which they are already familiar. LuzIA may reduce
some barriers to AI access because it can be used through familiar digital environments such as
WhatsApp, potentially reducing the need for students to adopt additional specialized platforms.
The tool also provides language-related functions, including writing assistance, translation,
grammatical correction, and explanations, which may support its use in EFL learning contexts.
However, accessibility does not automatically lead to educational effectiveness. Recent
studies have reported promising applications of AI-based conversational tools in EFL speaking
development, including improvements in oral performance, pronunciation, confidence,
willingness to communicate, and reduced speaking anxiety (Isabella et al., 2026; Siagian et al.,
2026). Nevertheless, these studies examined DOLA AI rather than LuzIA and focused primarily
on oral language development. To date, there appears to be limited empirical evidence regarding
INTRODUCTION
English academic writing acts as the gateway for university entrance and worldwide
certification, and the learning of the skill among students whose native tongue is not English
(English as a Foreign Language students) becomes extremely complicated. Academic writing
requires learners to manage multiple demands simultaneously, including grammatical accuracy,
appropriate vocabulary, logical organization of ideas, genre conventions, and academic style. For
EFL learners, these demands may be particularly challenging because they are developing
academic writing skills while continuing to develop proficiency in the target language (Ghafar &
Raheem, 2025). There is one more challenge in the learning process of English academic writing.
Providing continuous individualized feedback in conventional classroom settings may be difficult
because of practical constraints such as limited instructional time (Sanchez, 2024).
Fortunately, one more tool for language development appeared due to the rapid progress in
artificial intelligence. AI-powered LLMs are capable of giving instant feedback on grammar,
vocabulary, organization, and other writing-related issues without any necessity of teacher
intervention during all steps. Recent research has provided growing evidence that AI-assisted
tools can support EFL writing development. Experimental and comparative studies have reported
improvements in different dimensions of writing following AI-assisted instruction or revision,
while systematic reviews have identified broader patterns of linguistic and cognitive scaffolding
across EFL contexts (Song & Song, 2023; Tsai et al., 2024; Li & Wilson, 2025). In particular,
AI-supported writing appears to offer useful assistance with language-related aspects such as
grammar and vocabulary, while also providing opportunities for revision, idea development, and
organization.
There are many different AI-assistive tools. One of the unique features of the LuzIA tool
is the fact that it can work in WhatsApp. It becomes important because of the fact that WhatsApp
is integrated in daily practices of many Latin American teenagers. As a result, it will be possible
to receive AI help using the application with which they are already familiar. LuzIA may reduce
some barriers to AI access because it can be used through familiar digital environments such as
WhatsApp, potentially reducing the need for students to adopt additional specialized platforms.
The tool also provides language-related functions, including writing assistance, translation,
grammatical correction, and explanations, which may support its use in EFL learning contexts.
However, accessibility does not automatically lead to educational effectiveness. Recent
studies have reported promising applications of AI-based conversational tools in EFL speaking
development, including improvements in oral performance, pronunciation, confidence,
willingness to communicate, and reduced speaking anxiety (Isabella et al., 2026; Siagian et al.,
2026). Nevertheless, these studies examined DOLA AI rather than LuzIA and focused primarily
on oral language development. To date, there appears to be limited empirical evidence regarding
Vol. 13/ Núm. 3 2026 pág. 1510
the use of LuzIA to support EFL academic writing. Consequently, it remains unclear whether
LuzIA can support not only lower-order concerns, such as grammar and vocabulary, but also
higher-order concerns, including content development and the organization of ideas.
To address this research gap, the present study examines the use of LuzIA in a multilingual
secondary-school EFL context in Ecuador. The study evaluates multiple dimensions of the writing
process, such as grammar, vocabulary, organization and ideation of the texts along with writing
anxiety and learner autonomy as important writing-related factors.
Thus, there are three contributions that this study makes to science. First, it focuses
precisely on the use of LuzIA as the tool for improving academic writing. Second, it includes the
evaluation of AI assistance in Latin American secondary education, which is the context that is
relatively neglected in most of the existing research. Finally, the study evaluates the effect of AI
assistance on higher order and lower order concerns in order to understand what type of help can
be provided efficiently.
Research Objective
The main objective of this study was to evaluate the influence of LuzIA-assisted writing
support on EFL learners’ academic writing performance, writing anxiety, and learner autonomy
in a multilingual secondary-school context in Ecuador, with particular attention to changes in
Higher-Order Concerns (HOCs) and Lower-Order Concerns (LOCs) in students’ academic
writing.
Literature review
Learning English as a Foreign Language (EFL)
English as a Foreign Language (EFL) generally refers to the teaching and learning of
English in contexts where English is not the primary language of everyday communication and is
mainly learned through formal educational settings. It creates specific challenges for the learners
because the vast majority of all the experience acquired by EFL learners happens inside the
classroom. Thus, the amount of natural exposure that they may receive outside the classroom
setting might be minimal, and, as a result, the opportunity to get the necessary meaningful input
decreases significantly (Krashen, 1982). The same concerns writing because without getting the
exposure to academic English outside the classroom writing development becomes even harder.
However, in the case of Ecuador, things are even a little bit more complicated. English is a
mandatory foreign language throughout primary and secondary education, while Spanish serves
as a prevalent national language there. German, Kichwa, and other tongues are spoken in the
specific communities. In the private educational establishment where this research will be
conducted students learn English along with German, and therefore, the multi-language
environment is formed. For the learners, it requires managing several different systems
simultaneously, what increases the metacognitive and cognitive load for them during the learning
process (Ayala-Pazmiño & Alvarado-Lucas, 2023).
the use of LuzIA to support EFL academic writing. Consequently, it remains unclear whether
LuzIA can support not only lower-order concerns, such as grammar and vocabulary, but also
higher-order concerns, including content development and the organization of ideas.
To address this research gap, the present study examines the use of LuzIA in a multilingual
secondary-school EFL context in Ecuador. The study evaluates multiple dimensions of the writing
process, such as grammar, vocabulary, organization and ideation of the texts along with writing
anxiety and learner autonomy as important writing-related factors.
Thus, there are three contributions that this study makes to science. First, it focuses
precisely on the use of LuzIA as the tool for improving academic writing. Second, it includes the
evaluation of AI assistance in Latin American secondary education, which is the context that is
relatively neglected in most of the existing research. Finally, the study evaluates the effect of AI
assistance on higher order and lower order concerns in order to understand what type of help can
be provided efficiently.
Research Objective
The main objective of this study was to evaluate the influence of LuzIA-assisted writing
support on EFL learners’ academic writing performance, writing anxiety, and learner autonomy
in a multilingual secondary-school context in Ecuador, with particular attention to changes in
Higher-Order Concerns (HOCs) and Lower-Order Concerns (LOCs) in students’ academic
writing.
Literature review
Learning English as a Foreign Language (EFL)
English as a Foreign Language (EFL) generally refers to the teaching and learning of
English in contexts where English is not the primary language of everyday communication and is
mainly learned through formal educational settings. It creates specific challenges for the learners
because the vast majority of all the experience acquired by EFL learners happens inside the
classroom. Thus, the amount of natural exposure that they may receive outside the classroom
setting might be minimal, and, as a result, the opportunity to get the necessary meaningful input
decreases significantly (Krashen, 1982). The same concerns writing because without getting the
exposure to academic English outside the classroom writing development becomes even harder.
However, in the case of Ecuador, things are even a little bit more complicated. English is a
mandatory foreign language throughout primary and secondary education, while Spanish serves
as a prevalent national language there. German, Kichwa, and other tongues are spoken in the
specific communities. In the private educational establishment where this research will be
conducted students learn English along with German, and therefore, the multi-language
environment is formed. For the learners, it requires managing several different systems
simultaneously, what increases the metacognitive and cognitive load for them during the learning
process (Ayala-Pazmiño & Alvarado-Lucas, 2023).
Vol. 13/ Núm. 3 2026 pág. 1511
The complexity of the described situation becomes most prominent in secondary education
because by the eleventh grade students prepare for Cambridge B2 First, with high-performing
candidates potentially demonstrating performance at C1 on the Cambridge English Scale. In
addition, writing plays a crucial role during such examinations since learners need to not only
create grammatically correct sentences but also develop ideas, structure them logically, use the
necessary vocabulary, and adapt the created texts depending on the purpose. These competencies
are just the one that it may be quite difficult to develop in the framework of conventional class
learning.
However, it seems that similar issues with the development of writing happen among EFL
students in other Latin American countries. Research has identified academic writing as a
particularly challenging area for EFL learners. These difficulties are associated with the need to
manage several aspects of writing simultaneously, including grammatical accuracy, vocabulary
use, organization, coherence, and the development of ideas. Limited opportunities for writing
practice and individualized feedback may further increase these challenges (Ghafar & Raheem,
2025).
Thus, from this perspective, it makes sense to use AI to develop learners' writing skills in
the classroom. The main benefit of such a tool is that it allows getting an immediate and repeated
feedback during the text creation process. Using it in this way can add additional opportunities
for revising grammatical structures, selecting different vocabulary and developing ideas since it
is not possible to do so before the teacher's feedback is provided.
Academic Writing in the EFL Context
In contrast, academic writing entails skills beyond generating correct sentences. As EFL
writers think about grammar and vocabulary, they also organize their thoughts, make connections,
and meet requirements characteristic of argumentative or expository styles. Coordinating all these
processes places considerable demands on working memory, as students are still far from
acquiring a high level of proficiency in English (Ghafar & Raheem, 2025). The issues associated
with writing can be divided into two groups called Lower-Order Concerns (LOCs) and Higher-
Order Concerns (HOCs). While LOCs relate to grammar, spelling, punctuation, and vocabulary
selection, HOCs entail such things as formulating a thesis statement, organizing one's arguments,
and choosing language depending on purpose and audience (Jacobs et al., 1981).
This distinction is particularly relevant to understanding the role of AI in EFL writing
instruction. Recent research suggests that AI-assisted writing can be particularly effective in
supporting lower-order concerns (LOCs), including grammar and vocabulary, by providing
learners with immediate language-focused feedback and opportunities for revision (Tsai et al.,
2024). When it comes to higher-order concerns, on the other hand, such judgments require certain
understanding of the text being written – something that is often lacking in computer programs.
The complexity of the described situation becomes most prominent in secondary education
because by the eleventh grade students prepare for Cambridge B2 First, with high-performing
candidates potentially demonstrating performance at C1 on the Cambridge English Scale. In
addition, writing plays a crucial role during such examinations since learners need to not only
create grammatically correct sentences but also develop ideas, structure them logically, use the
necessary vocabulary, and adapt the created texts depending on the purpose. These competencies
are just the one that it may be quite difficult to develop in the framework of conventional class
learning.
However, it seems that similar issues with the development of writing happen among EFL
students in other Latin American countries. Research has identified academic writing as a
particularly challenging area for EFL learners. These difficulties are associated with the need to
manage several aspects of writing simultaneously, including grammatical accuracy, vocabulary
use, organization, coherence, and the development of ideas. Limited opportunities for writing
practice and individualized feedback may further increase these challenges (Ghafar & Raheem,
2025).
Thus, from this perspective, it makes sense to use AI to develop learners' writing skills in
the classroom. The main benefit of such a tool is that it allows getting an immediate and repeated
feedback during the text creation process. Using it in this way can add additional opportunities
for revising grammatical structures, selecting different vocabulary and developing ideas since it
is not possible to do so before the teacher's feedback is provided.
Academic Writing in the EFL Context
In contrast, academic writing entails skills beyond generating correct sentences. As EFL
writers think about grammar and vocabulary, they also organize their thoughts, make connections,
and meet requirements characteristic of argumentative or expository styles. Coordinating all these
processes places considerable demands on working memory, as students are still far from
acquiring a high level of proficiency in English (Ghafar & Raheem, 2025). The issues associated
with writing can be divided into two groups called Lower-Order Concerns (LOCs) and Higher-
Order Concerns (HOCs). While LOCs relate to grammar, spelling, punctuation, and vocabulary
selection, HOCs entail such things as formulating a thesis statement, organizing one's arguments,
and choosing language depending on purpose and audience (Jacobs et al., 1981).
This distinction is particularly relevant to understanding the role of AI in EFL writing
instruction. Recent research suggests that AI-assisted writing can be particularly effective in
supporting lower-order concerns (LOCs), including grammar and vocabulary, by providing
learners with immediate language-focused feedback and opportunities for revision (Tsai et al.,
2024). When it comes to higher-order concerns, on the other hand, such judgments require certain
understanding of the text being written – something that is often lacking in computer programs.
Vol. 13/ Núm. 3 2026 pág. 1512
For this reason, AI should be regarded as writing assistance rather than substitute of a
person. It will allow helping students to identify and eliminate language problems, so instructors
would have more time to focus on other important things such as arguments, organization of ideas,
purposes of writing, and meanings conveyed through the text. In this way, two means of help
might complement each other.
Another aspect of writing is emotional. Writing essays in a second language might cause
feelings of insecurity, mainly due to the fact that every mistake will be taken into consideration.
As identified by Cheng (2004) via Second Language Writing Anxiety Inventory, there are three
kinds of it, namely somatic, cognitive, and avoidance anxiety. All these types of anxiety might
make students unwilling to do any writing at all, avoid it, delay writing, etc.
The use of AI in learning environments, on the contrary, gives an opportunity to experiment
with writing, to generate questions and get answers to them, revise the text, get feedback – without
making it public at first. That way, the pressure caused by evaluating is somewhat reduced. Still,
this reduction alone will not make students write better; it might motivate them to write more, so
they would have more chances to practice.
Defining LuzIA: Design, Accessibility, and Educational Potential
LuzIA is a conversational artificial intelligence assistant designed to provide accessible AI-
based support through different digital environments. The platform allows users to interact with
the assistant through text or voice and can respond in the language used by the user. Importantly
for the context of the present study, LuzIA can be accessed through WhatsApp, allowing students
to interact with an AI assistant within a digital environment that may already be familiar to them
(LuzIA, n.d.-a).
LuzIA is a general-purpose AI assistant rather than a tool developed specifically for EFL
writing. Its documented functionalities include answering questions, assisting with writing and
study-related tasks, translating texts, transcribing audio, reading documents, explaining concepts,
and supporting language-learning activities (LuzIA, n.d.-a, n.d.-b). These functions make the
platform potentially relevant to EFL learning; however, the availability of these features does not
in itself demonstrate their effectiveness for developing academic writing.
In the present study, LuzIA was therefore used as a conversational scaffold rather than as
a tool for generating complete texts on behalf of learners. Its use was organized around three areas
related to the study's HOC–LOC framework. At the lower-order level, students used the assistant
to obtain support with grammar, vocabulary, spelling, punctuation, and alternative wording. At
the higher-order level, students could request guidance related to paragraph organization, idea
development, thesis statements, and cohesion. Students remained responsible for evaluating the
suggestions provided by the AI and deciding whether and how to incorporate them into their own
writing.
For this reason, AI should be regarded as writing assistance rather than substitute of a
person. It will allow helping students to identify and eliminate language problems, so instructors
would have more time to focus on other important things such as arguments, organization of ideas,
purposes of writing, and meanings conveyed through the text. In this way, two means of help
might complement each other.
Another aspect of writing is emotional. Writing essays in a second language might cause
feelings of insecurity, mainly due to the fact that every mistake will be taken into consideration.
As identified by Cheng (2004) via Second Language Writing Anxiety Inventory, there are three
kinds of it, namely somatic, cognitive, and avoidance anxiety. All these types of anxiety might
make students unwilling to do any writing at all, avoid it, delay writing, etc.
The use of AI in learning environments, on the contrary, gives an opportunity to experiment
with writing, to generate questions and get answers to them, revise the text, get feedback – without
making it public at first. That way, the pressure caused by evaluating is somewhat reduced. Still,
this reduction alone will not make students write better; it might motivate them to write more, so
they would have more chances to practice.
Defining LuzIA: Design, Accessibility, and Educational Potential
LuzIA is a conversational artificial intelligence assistant designed to provide accessible AI-
based support through different digital environments. The platform allows users to interact with
the assistant through text or voice and can respond in the language used by the user. Importantly
for the context of the present study, LuzIA can be accessed through WhatsApp, allowing students
to interact with an AI assistant within a digital environment that may already be familiar to them
(LuzIA, n.d.-a).
LuzIA is a general-purpose AI assistant rather than a tool developed specifically for EFL
writing. Its documented functionalities include answering questions, assisting with writing and
study-related tasks, translating texts, transcribing audio, reading documents, explaining concepts,
and supporting language-learning activities (LuzIA, n.d.-a, n.d.-b). These functions make the
platform potentially relevant to EFL learning; however, the availability of these features does not
in itself demonstrate their effectiveness for developing academic writing.
In the present study, LuzIA was therefore used as a conversational scaffold rather than as
a tool for generating complete texts on behalf of learners. Its use was organized around three areas
related to the study's HOC–LOC framework. At the lower-order level, students used the assistant
to obtain support with grammar, vocabulary, spelling, punctuation, and alternative wording. At
the higher-order level, students could request guidance related to paragraph organization, idea
development, thesis statements, and cohesion. Students remained responsible for evaluating the
suggestions provided by the AI and deciding whether and how to incorporate them into their own
writing.
Vol. 13/ Núm. 3 2026 pág. 1513
This distinction is central to the present research because it allows LuzIA to be examined
not simply as a text-generation tool but as a form of writing support. The study therefore evaluates
both the extent to which AI-assisted interaction is associated with improvements in HOCs and
LOCs and the areas in which learners may continue to require teacher guidance.
LuzIA AI in EFL Academic Writing
Nevertheless, despite some preliminary efforts to explore the effectiveness of using LuzIA
in writing English as a Foreign Language (EFL), the relevant literature has yet to develop
substantially. However, one can infer the potential benefits of the app from a number of empirical
findings related to AI writing support in general. Some of them seem especially relevant to the
theoretical framework of this research.
Song and Song (2023) examined the use of ChatGPT to support academic writing among
Chinese EFL university students. In their mixed-methods study, 50 learners were assigned to
either an AI-assisted group using ChatGPT or a control group receiving traditional writing
instruction. The AI-assisted group demonstrated significantly stronger writing performance,
including improvements in content, organization, grammar, and vocabulary. Their findings
suggest that immediate AI-supported feedback and opportunities for revision may contribute to
the development of EFL writing skills.
Marzuki et al. (2023) provide complementary qualitative evidence from EFL teachers’
perspectives. Based on semi-structured interviews with four EFL teachers from three Indonesian
universities, the study examined how teachers perceived the influence of AI writing tools on
students’ writing. The findings indicated perceived benefits for content development and
organization, including the generation and structuring of ideas. However, because the study relied
on teachers’ perceptions rather than direct experimental measurement of students’ writing gains,
its findings should be interpreted as qualitative evidence of the perceived educational potential of
AI writing tools.
Tsai et al. (2024) provide further evidence regarding the potential of ChatGPT-assisted
revision in EFL writing. In their paired-comparison study, 44 EFL English majors submitted
original essays and ChatGPT-assisted revised versions, which were independently evaluated by
two graders. The ChatGPT-assisted versions received significantly higher writing scores,
suggesting that AI-supported revision can improve assessed writing performance. However, the
authors also raised concerns about the implications of AI-assisted revision for fairness and
conventional approaches to writing assessment. This fact, obviously, makes this result especially
relevant for the current study, as if AI is capable of quickly identifying issues in language, then
similar effects may be expected from LuzIA when it is used as a writing scaffold.
Further support for the educational potential of AI-assisted writing can be found in Li and
Wilson (2025), who conducted a systematic review of AI-integrated scaffolding among K–12
English language learners. Their findings suggest that AI tools can provide linguistic, cognitive,
This distinction is central to the present research because it allows LuzIA to be examined
not simply as a text-generation tool but as a form of writing support. The study therefore evaluates
both the extent to which AI-assisted interaction is associated with improvements in HOCs and
LOCs and the areas in which learners may continue to require teacher guidance.
LuzIA AI in EFL Academic Writing
Nevertheless, despite some preliminary efforts to explore the effectiveness of using LuzIA
in writing English as a Foreign Language (EFL), the relevant literature has yet to develop
substantially. However, one can infer the potential benefits of the app from a number of empirical
findings related to AI writing support in general. Some of them seem especially relevant to the
theoretical framework of this research.
Song and Song (2023) examined the use of ChatGPT to support academic writing among
Chinese EFL university students. In their mixed-methods study, 50 learners were assigned to
either an AI-assisted group using ChatGPT or a control group receiving traditional writing
instruction. The AI-assisted group demonstrated significantly stronger writing performance,
including improvements in content, organization, grammar, and vocabulary. Their findings
suggest that immediate AI-supported feedback and opportunities for revision may contribute to
the development of EFL writing skills.
Marzuki et al. (2023) provide complementary qualitative evidence from EFL teachers’
perspectives. Based on semi-structured interviews with four EFL teachers from three Indonesian
universities, the study examined how teachers perceived the influence of AI writing tools on
students’ writing. The findings indicated perceived benefits for content development and
organization, including the generation and structuring of ideas. However, because the study relied
on teachers’ perceptions rather than direct experimental measurement of students’ writing gains,
its findings should be interpreted as qualitative evidence of the perceived educational potential of
AI writing tools.
Tsai et al. (2024) provide further evidence regarding the potential of ChatGPT-assisted
revision in EFL writing. In their paired-comparison study, 44 EFL English majors submitted
original essays and ChatGPT-assisted revised versions, which were independently evaluated by
two graders. The ChatGPT-assisted versions received significantly higher writing scores,
suggesting that AI-supported revision can improve assessed writing performance. However, the
authors also raised concerns about the implications of AI-assisted revision for fairness and
conventional approaches to writing assessment. This fact, obviously, makes this result especially
relevant for the current study, as if AI is capable of quickly identifying issues in language, then
similar effects may be expected from LuzIA when it is used as a writing scaffold.
Further support for the educational potential of AI-assisted writing can be found in Li and
Wilson (2025), who conducted a systematic review of AI-integrated scaffolding among K–12
English language learners. Their findings suggest that AI tools can provide linguistic, cognitive,
Vol. 13/ Núm. 3 2026 pág. 1514
and creative scaffolding by supporting processes such as grammar correction, vocabulary
development, idea generation, and content organization. However, the authors also emphasize the
importance of maintaining learner agency and avoiding excessive dependence on AI-generated
assistance. These findings are particularly relevant to the present study because they suggest that
AI can support different stages of the writing process while students remain responsible for
evaluating and applying the assistance they receive.
Thus, taking into account the results discussed above, one can come to the conclusion that
the usage of AI-assistant in writing usually provides useful, instant assistance during drafting and
revision stages, leading to substantial improvements in language-level aspects.
Theoretical framework
In total, three theories complement each other and address different aspects of EFL writing
under AI support – the development of learners who have guidance, gradual emergence of learner
autonomy and reasons behind the continued usage of certain technologies.
One way to start analyzing the intervention and understand its underlying processes is using
Vygotsky's Sociocultural Theory (1978) and specifically its aspect known as Zone of Proximal
Development (ZPD). The ZPD concept means the difference between what learners are able to
do independently and what they are able to do having the right help. Traditionally, this help comes
from a teacher, a peer who is better at something or generally More Knowledgeable Other. In this
situation, the role of a mediating tool may partially belong to LuzIA as it can analyze texts of
students, find mistakes and give suggestions that may allow moving beyond the limits of their
own capability.
It is very important to note the limit that should be taken into account in this case. The idea
is not that the machine does the job, and the student takes the results. On the contrary, it is
important to note that even when the AI provides certain help, the student remains responsible for
the decision whether to accept its suggestions or reject them.
The theory by Holec (1981) of learner autonomy provides another point of view for this
issue. In general, the theory claims that the learner takes responsibility for the process of his/her
own learning. For example, the learner sets the aims, chooses strategies, monitors himself/herself,
evaluates the outcome of the process. This is especially important in cases when learners deal
with AI since the process of getting help only starts with the suggestion. Next, students need to
decide what to do with it.
As a result, the usage of LuzIA may go beyond grammar/vocabulary corrections in certain
moments since the learner compares options, analyzes suggestions, applies the changes and
evaluates the options – all of this is metacognitive strategies of writing. Some of them may
become an integral part of the strategy of students' revision. Initially, the process happens between
a student and a machine but gradually some of those actions become a part of learners'
and creative scaffolding by supporting processes such as grammar correction, vocabulary
development, idea generation, and content organization. However, the authors also emphasize the
importance of maintaining learner agency and avoiding excessive dependence on AI-generated
assistance. These findings are particularly relevant to the present study because they suggest that
AI can support different stages of the writing process while students remain responsible for
evaluating and applying the assistance they receive.
Thus, taking into account the results discussed above, one can come to the conclusion that
the usage of AI-assistant in writing usually provides useful, instant assistance during drafting and
revision stages, leading to substantial improvements in language-level aspects.
Theoretical framework
In total, three theories complement each other and address different aspects of EFL writing
under AI support – the development of learners who have guidance, gradual emergence of learner
autonomy and reasons behind the continued usage of certain technologies.
One way to start analyzing the intervention and understand its underlying processes is using
Vygotsky's Sociocultural Theory (1978) and specifically its aspect known as Zone of Proximal
Development (ZPD). The ZPD concept means the difference between what learners are able to
do independently and what they are able to do having the right help. Traditionally, this help comes
from a teacher, a peer who is better at something or generally More Knowledgeable Other. In this
situation, the role of a mediating tool may partially belong to LuzIA as it can analyze texts of
students, find mistakes and give suggestions that may allow moving beyond the limits of their
own capability.
It is very important to note the limit that should be taken into account in this case. The idea
is not that the machine does the job, and the student takes the results. On the contrary, it is
important to note that even when the AI provides certain help, the student remains responsible for
the decision whether to accept its suggestions or reject them.
The theory by Holec (1981) of learner autonomy provides another point of view for this
issue. In general, the theory claims that the learner takes responsibility for the process of his/her
own learning. For example, the learner sets the aims, chooses strategies, monitors himself/herself,
evaluates the outcome of the process. This is especially important in cases when learners deal
with AI since the process of getting help only starts with the suggestion. Next, students need to
decide what to do with it.
As a result, the usage of LuzIA may go beyond grammar/vocabulary corrections in certain
moments since the learner compares options, analyzes suggestions, applies the changes and
evaluates the options – all of this is metacognitive strategies of writing. Some of them may
become an integral part of the strategy of students' revision. Initially, the process happens between
a student and a machine but gradually some of those actions become a part of learners'
Vol. 13/ Núm. 3 2026 pág. 1515
autonomous practice – identification of a problem, consideration of alternatives and evaluation of
the final option.
Such an approach is natural in terms of sociocultural theory since the support that comes
initially externally can gradually become internal over time. The idea is not the permanent
dependency on a particular machine but rather learning the strategies to work independently.
Finally, TAM of Davis (1989) deals with another issue related to learners. It is about
motivation behind the consistent use of the same technology in order to continue learning.
Technology Acceptance Model focuses on the following aspects – Perceived Usefulness (PU)
and Perceived Ease of Use (PEOU). The first one describes learners' opinion about the extent to
which the technology helps to perform meaningful tasks. The latter one refers to the ease of use
of the technology in question.
It is important to mention the aspect related to PEOU that the tool appears in the context of
an already used application. It means that the learners do not have to spend additional time on
learning how to use a new technology in order to be able to communicate with the machine. PEOU
concerns learners' perceptions of how easy the technology is to access and use, whereas PU
concerns their beliefs about its usefulness for improving writing performance.
MATERIALS AND METHODS
Research Design
The study employed a Sequential Explanatory Mixed-Methods Design consisting of an
initial quantitative phase followed by a qualitative explanatory pase (Maforah & Leburu-Masigo,
2018). This design was selected considering that numbers alone cannot suffice in explanation of
processes occurring as students incorporate AI into writing practice. Thus, the investigation takes
place through two interrelated phases, with the qualitative phase assisting in the interpretation of
the first phase quantitative findings.
During the first phase, a quantitative analysis is conducted. Writing scores before and after
the intervention were compared using an analysis rubric alongside with composite Likert-scale
questionnaire scores. The data was used to reveal whether any changes take place after the
implementation of the LuzIA intervention and, most importantly, in what way these changes were
statistically significant and in which direction.
At the second stage of the study, the experiences associated with the numeric data were
analyzed in greater detail using semi-structured interviews and reflective journals as data
collection instruments. This stage does not aim at verification of the previous findings but at more
profound interpretation. More precisely, the aim is to explore students' experience related to
changes that have taken place during the intervention.
The quasi-experimental design is used rather than a completely randomized one. As the
research is performed with intact existing groups of students in one school, there is no possibility
autonomous practice – identification of a problem, consideration of alternatives and evaluation of
the final option.
Such an approach is natural in terms of sociocultural theory since the support that comes
initially externally can gradually become internal over time. The idea is not the permanent
dependency on a particular machine but rather learning the strategies to work independently.
Finally, TAM of Davis (1989) deals with another issue related to learners. It is about
motivation behind the consistent use of the same technology in order to continue learning.
Technology Acceptance Model focuses on the following aspects – Perceived Usefulness (PU)
and Perceived Ease of Use (PEOU). The first one describes learners' opinion about the extent to
which the technology helps to perform meaningful tasks. The latter one refers to the ease of use
of the technology in question.
It is important to mention the aspect related to PEOU that the tool appears in the context of
an already used application. It means that the learners do not have to spend additional time on
learning how to use a new technology in order to be able to communicate with the machine. PEOU
concerns learners' perceptions of how easy the technology is to access and use, whereas PU
concerns their beliefs about its usefulness for improving writing performance.
MATERIALS AND METHODS
Research Design
The study employed a Sequential Explanatory Mixed-Methods Design consisting of an
initial quantitative phase followed by a qualitative explanatory pase (Maforah & Leburu-Masigo,
2018). This design was selected considering that numbers alone cannot suffice in explanation of
processes occurring as students incorporate AI into writing practice. Thus, the investigation takes
place through two interrelated phases, with the qualitative phase assisting in the interpretation of
the first phase quantitative findings.
During the first phase, a quantitative analysis is conducted. Writing scores before and after
the intervention were compared using an analysis rubric alongside with composite Likert-scale
questionnaire scores. The data was used to reveal whether any changes take place after the
implementation of the LuzIA intervention and, most importantly, in what way these changes were
statistically significant and in which direction.
At the second stage of the study, the experiences associated with the numeric data were
analyzed in greater detail using semi-structured interviews and reflective journals as data
collection instruments. This stage does not aim at verification of the previous findings but at more
profound interpretation. More precisely, the aim is to explore students' experience related to
changes that have taken place during the intervention.
The quasi-experimental design is used rather than a completely randomized one. As the
research is performed with intact existing groups of students in one school, there is no possibility
Vol. 13/ Núm. 3 2026 pág. 1516
of randomization of individual participants to distinct groups. Hence, the focus of statistical
comparisons is within-participant comparisons between pre and post test. The initial difference in
writing ability across grade levels is controlled by means of ANCOVA.
Thus, the quasi-experimental nature of the design limits the strength of the causal
inferences that can be drawn. Nonetheless, it reflects the actual conditions under which
educational intervention research is conducted, with pre-existing classes and student schedules
already established. This approach is consistent with conducting educational research in authentic
settings (Creswell & Creswell, 2023) and provides evidence of how AI-assisted writing support
operates when implemented within an actual EFL instructional context.
Context and Participants
The study was carried out in a private school located in Quito within a bilingual secondary-
school context offering a multilingual program that includes Spanish, German, English, and
French. As part of its educational program, the school combines Ecuadorian and German
educational systems, including the German Abitur program, resulting in substantial demands
related to linguistic accuracy, critical thinking, and academic writing.
English language instruction begins at approximately age ten and becomes progressively
more advanced throughout secondary education. By the upper secondary grades, students prepare
for international English certification, with expected proficiency generally ranging from B2 to C1
according to the Common European Framework of Reference for Languages (CEFR). This setting
is particularly relevant to the present study because academic writing forms part of both students'
language development and their preparation for international certification.
The study included 145 EFL learners enrolled in six intact classroom cohorts ranging from
8th to 12th grade. All eligible students from these cohorts who met the study's participation
requirements and for whom the required consent and assent had been obtained were included in
the study. Therefore, rather than selecting a subset of students through stratified random sampling,
the study included the accessible population across the participating cohorts. This approach
ensured representation across grade levels and gender while preserving the naturally occurring
classroom groups in which the intervention was implemented.
An a priori power analysis was used to assess whether the available number of participants
would provide adequate statistical power for the planned analyses (Cohen, 1988). Assuming α =
.05 and a medium effect size (d = 0.50), the available sample of 145 participants exceeded the
minimum requirement for 80% statistical power. Participants represented different stages of
secondary education, with expected English proficiency ranging primarily from B2 to C1. Table
1 presents the distribution of participants across the six classroom cohorts.
Inclusion and Exclusion Criteria
Students were eligible for inclusion if they were enrolled in one of the six participating
classroom cohorts, were actively attending the English courses involved in the study, and had
of randomization of individual participants to distinct groups. Hence, the focus of statistical
comparisons is within-participant comparisons between pre and post test. The initial difference in
writing ability across grade levels is controlled by means of ANCOVA.
Thus, the quasi-experimental nature of the design limits the strength of the causal
inferences that can be drawn. Nonetheless, it reflects the actual conditions under which
educational intervention research is conducted, with pre-existing classes and student schedules
already established. This approach is consistent with conducting educational research in authentic
settings (Creswell & Creswell, 2023) and provides evidence of how AI-assisted writing support
operates when implemented within an actual EFL instructional context.
Context and Participants
The study was carried out in a private school located in Quito within a bilingual secondary-
school context offering a multilingual program that includes Spanish, German, English, and
French. As part of its educational program, the school combines Ecuadorian and German
educational systems, including the German Abitur program, resulting in substantial demands
related to linguistic accuracy, critical thinking, and academic writing.
English language instruction begins at approximately age ten and becomes progressively
more advanced throughout secondary education. By the upper secondary grades, students prepare
for international English certification, with expected proficiency generally ranging from B2 to C1
according to the Common European Framework of Reference for Languages (CEFR). This setting
is particularly relevant to the present study because academic writing forms part of both students'
language development and their preparation for international certification.
The study included 145 EFL learners enrolled in six intact classroom cohorts ranging from
8th to 12th grade. All eligible students from these cohorts who met the study's participation
requirements and for whom the required consent and assent had been obtained were included in
the study. Therefore, rather than selecting a subset of students through stratified random sampling,
the study included the accessible population across the participating cohorts. This approach
ensured representation across grade levels and gender while preserving the naturally occurring
classroom groups in which the intervention was implemented.
An a priori power analysis was used to assess whether the available number of participants
would provide adequate statistical power for the planned analyses (Cohen, 1988). Assuming α =
.05 and a medium effect size (d = 0.50), the available sample of 145 participants exceeded the
minimum requirement for 80% statistical power. Participants represented different stages of
secondary education, with expected English proficiency ranging primarily from B2 to C1. Table
1 presents the distribution of participants across the six classroom cohorts.
Inclusion and Exclusion Criteria
Students were eligible for inclusion if they were enrolled in one of the six participating
classroom cohorts, were actively attending the English courses involved in the study, and had
Vol. 13/ Núm. 3 2026 pág. 1517
obtained written informed consent from a parent or legal guardian and provided their own assent
to participate. Participants were also required to complete the pre-test and post-test writing
assessments and participate in the intervention activities.
Students were excluded from the final analysis if parental or guardian consent or student
assent was not obtained, if they did not complete either the pre-test or post-test assessment, or if
they did not participate sufficiently in the intervention to provide complete data for analysis. No
students were excluded on the basis of gender or expected CEFR proficiency level.
Table 1
Sample Distribution by Grade Level, Gender, and English-Medium Subjects
Grade / Level Males Females n Subjects in English Expected CEFR
Level
8th Grade EGB A 20 12 32 English; Social Studies B2
9th Grade EGB A 8 10 18 English; Social Studies B2
10th Grade EGB A 10 11 21 English; Social Studies;
Reading
B2
1st Year BGU
(Sciences)
12 7 19 English; Reading B2–C1
3rd Year BGU A 14 16 30 English; Reading; Intl.
Cert. Prep.
B2–C1
3rd Year BGU B 9 16 25 English; Reading; Intl.
Cert. Prep.
B2–C1
Total 73 72 145*
Note. N = 145. Participants represented six intact classroom cohorts and ranged in age from
approximately 12 to 18 years. CEFR classifications represent expected curricular proficiency
levels established by the school and were not independently assessed as part of the present study.
Instruments
Data collection was carried out using five devices that catered for both qualitative and
quantitative aspects investigated in the study.
Pre-test Prompt
The pre-test consisted of the 60-minute argumentative writing test and followed the
structure of the writing that is expected from students when doing Cambridge FCE. The task
included writing of 140 to 190 words, discussing two different opinions about a sociological-
educational problem, presenting one's point of view and organizing the writing in accordance with
typical academic essay outline: introduction, body paragraphs and conclusion. The prompt
specified: “Some people think that technology makes students independent learners. Others argue
that technology makes students dependent on the external assistance. Discuss both sides and give
obtained written informed consent from a parent or legal guardian and provided their own assent
to participate. Participants were also required to complete the pre-test and post-test writing
assessments and participate in the intervention activities.
Students were excluded from the final analysis if parental or guardian consent or student
assent was not obtained, if they did not complete either the pre-test or post-test assessment, or if
they did not participate sufficiently in the intervention to provide complete data for analysis. No
students were excluded on the basis of gender or expected CEFR proficiency level.
Table 1
Sample Distribution by Grade Level, Gender, and English-Medium Subjects
Grade / Level Males Females n Subjects in English Expected CEFR
Level
8th Grade EGB A 20 12 32 English; Social Studies B2
9th Grade EGB A 8 10 18 English; Social Studies B2
10th Grade EGB A 10 11 21 English; Social Studies;
Reading
B2
1st Year BGU
(Sciences)
12 7 19 English; Reading B2–C1
3rd Year BGU A 14 16 30 English; Reading; Intl.
Cert. Prep.
B2–C1
3rd Year BGU B 9 16 25 English; Reading; Intl.
Cert. Prep.
B2–C1
Total 73 72 145*
Note. N = 145. Participants represented six intact classroom cohorts and ranged in age from
approximately 12 to 18 years. CEFR classifications represent expected curricular proficiency
levels established by the school and were not independently assessed as part of the present study.
Instruments
Data collection was carried out using five devices that catered for both qualitative and
quantitative aspects investigated in the study.
Pre-test Prompt
The pre-test consisted of the 60-minute argumentative writing test and followed the
structure of the writing that is expected from students when doing Cambridge FCE. The task
included writing of 140 to 190 words, discussing two different opinions about a sociological-
educational problem, presenting one's point of view and organizing the writing in accordance with
typical academic essay outline: introduction, body paragraphs and conclusion. The prompt
specified: “Some people think that technology makes students independent learners. Others argue
that technology makes students dependent on the external assistance. Discuss both sides and give
Vol. 13/ Núm. 3 2026 pág. 1518
your opinion”. The selected topic was considered appropriate for the range of proficiency
expected across the participating cohorts and allowed more advanced students to develop
increasingly complex arguments. Students had no possibility to use LuzIA and any other AI tool.
Post-test Prompt
12 weeks later, the participants performed another similar argumentative writing task, also
in the conditions: 60 minutes, between 140 to 190 words, same writing requirements as pre-test
and a different topic to minimize influence of familiarity with the first prompt. As in the pre-test,
participants completed the post-test without access to LuzIA or any other AI-assisted writing tool.
The second prompt stated: “There are some educators who are in favor of using AI tools in
language learning to improve students’ results. There are others who say that AI hinders the
development of independent language skills. Discuss both sides and give your opinion.” Just like
before, this test included considering different viewpoints, stating and defending one's opinion on
the issue. However, unlike the pre-test prompt, this task was specifically connected to their
experiences during the previous 12 weeks, during which they engaged in using LuzIA tool, so
they had a tangible starting point for discussion of the advantages and limitations of the tool in
language learning.
Pilot testing
Prior to conducting data collection, both prompts underwent pilot testing during Week 1
among 12 students of similar proficiency level who did not take part in the current study. The aim
of the pilot testing was to ensure the comparability of the two tasks, their performability in the
allotted time frame and their interpretability for scoring.
Analytic Scoring Rubric
The analysis of the written work of students was carried out through an analytic rubric
based on the ESL Composition Profile created by Jacobs et al. (1981). As opposed to assigning
one rating based on one’s general impression of the writing performance, the rubric offered a
possibility to analyze each essay according to five specific dimensions based on HOC and LOC
framework used in this study.
Content and Organization were the HOC dimensions, for which 30 and 20 points were
assigned, respectively. Vocabulary, Language Use/Grammar and Mechanics were LOC
dimensions, with 20, 25 and 5 points. Thus, the total writing score could reach the value of 100
points when combining the results for all the five dimensions.
Each of the dimensions was rated for four performance bands – Excellent, Very Good,
Good and Very Poor. The behavioral descriptors that corresponded to each band were formulated
and served to guide rater during the evaluation.
In order to evaluate reliability of rater judgments, two trained raters analyzed a stratified
random sample including 30% of pre- and post-test essays. Inter-rater reliability was computed
using Cohen’s weighted kappa (κ), with the pre-established minimum acceptable level κ = .70.
your opinion”. The selected topic was considered appropriate for the range of proficiency
expected across the participating cohorts and allowed more advanced students to develop
increasingly complex arguments. Students had no possibility to use LuzIA and any other AI tool.
Post-test Prompt
12 weeks later, the participants performed another similar argumentative writing task, also
in the conditions: 60 minutes, between 140 to 190 words, same writing requirements as pre-test
and a different topic to minimize influence of familiarity with the first prompt. As in the pre-test,
participants completed the post-test without access to LuzIA or any other AI-assisted writing tool.
The second prompt stated: “There are some educators who are in favor of using AI tools in
language learning to improve students’ results. There are others who say that AI hinders the
development of independent language skills. Discuss both sides and give your opinion.” Just like
before, this test included considering different viewpoints, stating and defending one's opinion on
the issue. However, unlike the pre-test prompt, this task was specifically connected to their
experiences during the previous 12 weeks, during which they engaged in using LuzIA tool, so
they had a tangible starting point for discussion of the advantages and limitations of the tool in
language learning.
Pilot testing
Prior to conducting data collection, both prompts underwent pilot testing during Week 1
among 12 students of similar proficiency level who did not take part in the current study. The aim
of the pilot testing was to ensure the comparability of the two tasks, their performability in the
allotted time frame and their interpretability for scoring.
Analytic Scoring Rubric
The analysis of the written work of students was carried out through an analytic rubric
based on the ESL Composition Profile created by Jacobs et al. (1981). As opposed to assigning
one rating based on one’s general impression of the writing performance, the rubric offered a
possibility to analyze each essay according to five specific dimensions based on HOC and LOC
framework used in this study.
Content and Organization were the HOC dimensions, for which 30 and 20 points were
assigned, respectively. Vocabulary, Language Use/Grammar and Mechanics were LOC
dimensions, with 20, 25 and 5 points. Thus, the total writing score could reach the value of 100
points when combining the results for all the five dimensions.
Each of the dimensions was rated for four performance bands – Excellent, Very Good,
Good and Very Poor. The behavioral descriptors that corresponded to each band were formulated
and served to guide rater during the evaluation.
In order to evaluate reliability of rater judgments, two trained raters analyzed a stratified
random sample including 30% of pre- and post-test essays. Inter-rater reliability was computed
using Cohen’s weighted kappa (κ), with the pre-established minimum acceptable level κ = .70.
Vol. 13/ Núm. 3 2026 pág. 1519
The disagreements in case of the difference in at least one performance band for any particular
dimension between raters were discussed until consensus was reached.
Finally, the inter-rater reliability reached the value of κ = .83 and exceeded the required
level.
Table 2
Analytic Scoring Rubric: HOC and LOC Domains, Maximum Points, and Constructs Measured
Category Domain Max.
Points
Construct Measured
HOCs (50
pts)
Content 30 Thesis development, topic relevance, depth
of argumentation
Organization 20 Logical sequencing, cohesion, transition
deployment
LOCs (50
pts)
Vocabulary 20 Lexical range, word choice accuracy,
academic register
Language Use /
Grammar
25 Morphosyntactic accuracy, tense,
agreement, complexity
Mechanics 5 Spelling, punctuation, capitalization,
paragraphing
Total Score 100 Overall academic writing proficiency
Notes. HOCs: Higher Order Concerns; LOCs: Lower Order Concerns. All domains were rated on four performance
bands with behavioral descriptions.
Student Perception Questionnaire
An 18-item questionnaire using a five-point Likert scale (1 = Strongly Disagree to 5 =
Strongly Agree) was used to examine three variables related to students’ experiences throughout
the intervention; these are writing anxiety, learner autonomy, and technology acceptance.
The first part of the questionnaire consisted of items (1-5), which focused on writing
anxiety, based on the Second Language Writing Anxiety Inventory (SLWAI) by Cheng (2004).
The second section (items 6-10) examined learner autonomy based on Holec (1981)'s
conceptualization, focusing on observable learning actions rather than seeing autonomy only as
an abstract construct (for example, conducting an individual brainstorm, conducting revision
independently, actively searching for support).
Third section consisted of items 11-18 and focused on TAM developed by Davis (1989);
this model examines two factors, namely Perceived Usefulness (PU) and Perceived Ease of Use
(PEOU).
The writing anxiety and learner autonomy sections of the questionnaire were administered
twice, at pre-intervention and post-intervention, whereas the TAM section assessing Perceived
The disagreements in case of the difference in at least one performance band for any particular
dimension between raters were discussed until consensus was reached.
Finally, the inter-rater reliability reached the value of κ = .83 and exceeded the required
level.
Table 2
Analytic Scoring Rubric: HOC and LOC Domains, Maximum Points, and Constructs Measured
Category Domain Max.
Points
Construct Measured
HOCs (50
pts)
Content 30 Thesis development, topic relevance, depth
of argumentation
Organization 20 Logical sequencing, cohesion, transition
deployment
LOCs (50
pts)
Vocabulary 20 Lexical range, word choice accuracy,
academic register
Language Use /
Grammar
25 Morphosyntactic accuracy, tense,
agreement, complexity
Mechanics 5 Spelling, punctuation, capitalization,
paragraphing
Total Score 100 Overall academic writing proficiency
Notes. HOCs: Higher Order Concerns; LOCs: Lower Order Concerns. All domains were rated on four performance
bands with behavioral descriptions.
Student Perception Questionnaire
An 18-item questionnaire using a five-point Likert scale (1 = Strongly Disagree to 5 =
Strongly Agree) was used to examine three variables related to students’ experiences throughout
the intervention; these are writing anxiety, learner autonomy, and technology acceptance.
The first part of the questionnaire consisted of items (1-5), which focused on writing
anxiety, based on the Second Language Writing Anxiety Inventory (SLWAI) by Cheng (2004).
The second section (items 6-10) examined learner autonomy based on Holec (1981)'s
conceptualization, focusing on observable learning actions rather than seeing autonomy only as
an abstract construct (for example, conducting an individual brainstorm, conducting revision
independently, actively searching for support).
Third section consisted of items 11-18 and focused on TAM developed by Davis (1989);
this model examines two factors, namely Perceived Usefulness (PU) and Perceived Ease of Use
(PEOU).
The writing anxiety and learner autonomy sections of the questionnaire were administered
twice, at pre-intervention and post-intervention, whereas the TAM section assessing Perceived
Vol. 13/ Núm. 3 2026 pág. 1520
Usefulness (PU) and Perceived Ease of Use (PEOU) was administered only at post-intervention,
after students had experienced using LuzIA. Reliability was measured using Cronbach's Alpha,
where value above α ≥. 70 was considered to be adequate (Nunnally, 1978). Reliability of all
sections was satisfactory – SLWAI α=. 82, Learner Autonomy α=. 79, PU α=. 88, and PEOU α=.
91.
Technology Acceptance Model (TAM) Perceptions of LuzIA
The questions included in the TAM portion were designed to measure students’ views
regarding LuzIA as a support for their writing tasks. In particular, four questions measured PU,
measuring whether LuzIA helped improve writing, identify grammar mistakes, expand their
vocabulary, and help improve their organizational skills.
Four more questions focused on PEOU and evaluated whether students considered LuzIA
easy to reach through WhatsApp, user-friendly, not complicated by unnecessary technicalities,
and quickly attainable. This adaptation is consistent with the Technology Acceptance Model
(Davis, 1989) and with recent research on AI-supported language learning through chatbot-based
environments (Behforouz & Al Ghaithi, 2024).
Specifically, WhatsApp seems particularly relevant in the study as students did not have to
master another platform before starting working with AI. Familiar environment seemed to be
expected to result in improved PEOU.
EFA demonstrated the expected two-factor model of PU and PEOU. All factors were
loaded on the appropriate scale at λ ≥ .62, with no cross-loading being equal to or exceeding .30.
Data Collection Procedure
For each stage, data were collected for the entire twelve-weeks period. In Week 1-2 (Stage
1), the pretest was taken with all the essays done under uniform test conditions at the school
computer lab without LuzIA or any other AI access. Immediately after that, participants took the
baseline questionnaire. Participants then took part in two guided workshops on the use of LuzIA
where topics of formulating purposeful prompts, evaluating AI recommendations, deciding
whether to accept or reject AI suggestions, recognizing possible errors or hallucination by AI, as
well as the limits of academic integrity were addressed. Right from the very beginning, it was
made clear that the aim of the use of LuzIA is to support the process of students' writing but not
to do the writing for them. Participants were given an alphanumeric code and were made aware
of the voluntary and confidential nature of their participation in the research.
In Weeks 3-10 (Stage 2), the tool started to be incorporated into students' writing process
in three subjects which were Academic Writing, Grammar & Composition and Advanced
Writing. In each of these disciplines, it had a different purpose – developing arguments in the
essay, grammatical correction, vocabulary choosing and more sophisticated writing tasks.
Students used LuzIA both during in-class writing activities and outside school in individual
writing practice. The frequency, duration of the work with the tool and the primary task performed
Usefulness (PU) and Perceived Ease of Use (PEOU) was administered only at post-intervention,
after students had experienced using LuzIA. Reliability was measured using Cronbach's Alpha,
where value above α ≥. 70 was considered to be adequate (Nunnally, 1978). Reliability of all
sections was satisfactory – SLWAI α=. 82, Learner Autonomy α=. 79, PU α=. 88, and PEOU α=.
91.
Technology Acceptance Model (TAM) Perceptions of LuzIA
The questions included in the TAM portion were designed to measure students’ views
regarding LuzIA as a support for their writing tasks. In particular, four questions measured PU,
measuring whether LuzIA helped improve writing, identify grammar mistakes, expand their
vocabulary, and help improve their organizational skills.
Four more questions focused on PEOU and evaluated whether students considered LuzIA
easy to reach through WhatsApp, user-friendly, not complicated by unnecessary technicalities,
and quickly attainable. This adaptation is consistent with the Technology Acceptance Model
(Davis, 1989) and with recent research on AI-supported language learning through chatbot-based
environments (Behforouz & Al Ghaithi, 2024).
Specifically, WhatsApp seems particularly relevant in the study as students did not have to
master another platform before starting working with AI. Familiar environment seemed to be
expected to result in improved PEOU.
EFA demonstrated the expected two-factor model of PU and PEOU. All factors were
loaded on the appropriate scale at λ ≥ .62, with no cross-loading being equal to or exceeding .30.
Data Collection Procedure
For each stage, data were collected for the entire twelve-weeks period. In Week 1-2 (Stage
1), the pretest was taken with all the essays done under uniform test conditions at the school
computer lab without LuzIA or any other AI access. Immediately after that, participants took the
baseline questionnaire. Participants then took part in two guided workshops on the use of LuzIA
where topics of formulating purposeful prompts, evaluating AI recommendations, deciding
whether to accept or reject AI suggestions, recognizing possible errors or hallucination by AI, as
well as the limits of academic integrity were addressed. Right from the very beginning, it was
made clear that the aim of the use of LuzIA is to support the process of students' writing but not
to do the writing for them. Participants were given an alphanumeric code and were made aware
of the voluntary and confidential nature of their participation in the research.
In Weeks 3-10 (Stage 2), the tool started to be incorporated into students' writing process
in three subjects which were Academic Writing, Grammar & Composition and Advanced
Writing. In each of these disciplines, it had a different purpose – developing arguments in the
essay, grammatical correction, vocabulary choosing and more sophisticated writing tasks.
Students used LuzIA both during in-class writing activities and outside school in individual
writing practice. The frequency, duration of the work with the tool and the primary task performed
Vol. 13/ Núm. 3 2026 pág. 1521
(either grammatical correction, finding vocabulary, structural support or translation) were
recorded in WhatsApp interaction logs. Besides, students kept weekly reflective journals
containing information on the use of LuzIA and attempts at writing.
Finally, in Week 11-12 (Stage 3), the post-test was taken under the same standardized
conditions and then followed by the post intervention questionnaire. Selected 20 students were
interviewed in a semi-structured manner. A maximum variation sampling approach made sure
that interviews included participants of different grades, genders, and proficiency bands.
Data Analysis
Quantitative data analysis entailed five stages. First, descriptive statistics (means, SD,
medians, ranges, etc.) for overall writing scores, all rubric dimensions, and all questionnaire
constructs were calculated. When sample sizes allowed it, comparisons between subgroups were
performed in accordance with their grade level and gender. Tests for normality (Shapiro-Wilk
tests) were run before selecting the inferential methods.
Second, paired-samples t-tests or Wilcoxon signed-rank test was selected for comparing
pretest/posttest performance based on the rubric total score as well as five writing dimensions. As
five different writing rubrics were compared, the significance threshold was adjusted (Bonferroni
procedure, α=.01). Significant p-value only tells about the presence or absence of difference.
Therefore, Cohen's d effect size was calculated as Mdiff/SDdiff for all of them according to the
criteria suggested by Plonsky and Oswald (2014) for EFL studies: small (d = .40), medium (d =
.70), and large (d = 1.00).
Third, in order to establish whether post-test writing performance differed among the six
classroom cohorts, a one-way ANCOVA was performed while controlling for initial writing
performance. Post-test total score was entered as the dependent variable, classroom cohort as the
between-subjects factor, and pre-test total score as the covariate. Homogeneity of regression
slopes and normality of residuals were checked in advance.
Fourth, Pearson correlation analyses were conducted to examine the relationships between
students’ perceptions of LuzIA and the intervention outcomes. Specifically, the association
between Perceived Usefulness (PU) and writing-score improvement was examined, as well as the
relationship between Perceived Ease of Use (PEOU) and PU.
Fifth, changes in learners' writing anxiety and autonomy were estimated using paired-
samples t-tests or Wilcoxon tests depending on the distributions. The TAM questionnaire items
were analyzed using exploratory factor analysis. Cronbach's alpha was calculated for each
questionnaire construct.
As for the qualitative data analysis, it followed Braun and Clarke's (2022) framework of
Reflexive Thematic Analysis. Journals and interviews were analyzed in six stages: familiarizing
with data, generating initial codes, searching for themes, reviewing themes, defining them and
producing report.
(either grammatical correction, finding vocabulary, structural support or translation) were
recorded in WhatsApp interaction logs. Besides, students kept weekly reflective journals
containing information on the use of LuzIA and attempts at writing.
Finally, in Week 11-12 (Stage 3), the post-test was taken under the same standardized
conditions and then followed by the post intervention questionnaire. Selected 20 students were
interviewed in a semi-structured manner. A maximum variation sampling approach made sure
that interviews included participants of different grades, genders, and proficiency bands.
Data Analysis
Quantitative data analysis entailed five stages. First, descriptive statistics (means, SD,
medians, ranges, etc.) for overall writing scores, all rubric dimensions, and all questionnaire
constructs were calculated. When sample sizes allowed it, comparisons between subgroups were
performed in accordance with their grade level and gender. Tests for normality (Shapiro-Wilk
tests) were run before selecting the inferential methods.
Second, paired-samples t-tests or Wilcoxon signed-rank test was selected for comparing
pretest/posttest performance based on the rubric total score as well as five writing dimensions. As
five different writing rubrics were compared, the significance threshold was adjusted (Bonferroni
procedure, α=.01). Significant p-value only tells about the presence or absence of difference.
Therefore, Cohen's d effect size was calculated as Mdiff/SDdiff for all of them according to the
criteria suggested by Plonsky and Oswald (2014) for EFL studies: small (d = .40), medium (d =
.70), and large (d = 1.00).
Third, in order to establish whether post-test writing performance differed among the six
classroom cohorts, a one-way ANCOVA was performed while controlling for initial writing
performance. Post-test total score was entered as the dependent variable, classroom cohort as the
between-subjects factor, and pre-test total score as the covariate. Homogeneity of regression
slopes and normality of residuals were checked in advance.
Fourth, Pearson correlation analyses were conducted to examine the relationships between
students’ perceptions of LuzIA and the intervention outcomes. Specifically, the association
between Perceived Usefulness (PU) and writing-score improvement was examined, as well as the
relationship between Perceived Ease of Use (PEOU) and PU.
Fifth, changes in learners' writing anxiety and autonomy were estimated using paired-
samples t-tests or Wilcoxon tests depending on the distributions. The TAM questionnaire items
were analyzed using exploratory factor analysis. Cronbach's alpha was calculated for each
questionnaire construct.
As for the qualitative data analysis, it followed Braun and Clarke's (2022) framework of
Reflexive Thematic Analysis. Journals and interviews were analyzed in six stages: familiarizing
with data, generating initial codes, searching for themes, reviewing themes, defining them and
producing report.
Vol. 13/ Núm. 3 2026 pág. 1522
Auditing trail was compiled to record researchers' codes and reflections.
The joint display framework made it possible to integrate qualitative and quantitative
strands of data through comparing their findings, explaining one type of results with the other and
finding out the discrepancies in the experiences of the respondents and statistical data.
Ethical Considerations
Given that a substantial proportion of the participants were between 12 and 18 years of age,
the protection of minors was an integral component of the research design. The study followed
internationally recognized principles for research involving human participants, with particular
attention to informed consent and assent, voluntary participation, confidentiality, data protection,
and participants’ right to withdraw from the study without penalty.
Prior to data collection, authorization to conduct the study was obtained through the
institutional procedures established by the participating school. Because the study involved
minors, a two-level consent procedure was implemented. Written informed consent was obtained
from parents or legal guardians, while students provided their own assent to participate. The
consent and assent forms were provided in Spanish and explained the purpose and procedures of
the study, the use and protection of the collected data, the voluntary nature of participation, and
the right to withdraw without academic penalty. Only students for whom parental or guardian
consent and student assent had been obtained were included in the research procedures.
To protect participants’ identities, alphanumeric codes were assigned to all research
records. Personally identifying information was stored separately from the research data in a
protected file accessible only to the research team. Research data were securely stored and
managed in accordance with the data-protection and retention requirements established by the
participating institution.
RESULTS
Descriptive Analysis of the Pre-test and Post-test
Table 3 shows the summary statistics for the pre-test and post-test among the overall sample
(n = 145), comprising the scores on all of the five dimensions of writing. Before the intervention,
students scored an average Total Score of 51.4 (SD = 12.3) from 100. At the end of the 12-week
period, the average score increased to 67.8 (SD = 10.7) after the intervention, which is indicative
of an increase of 16.4 in their scores in the pre-test and post-test comparison. It provides a
preliminary picture of the effects of the intervention. Further statistical analysis seeks to examine
whether there is a significant effect, what the size of the effect is, and whether the effect can be
observed at the Lower and Higher-Order dimensions level separately.
Auditing trail was compiled to record researchers' codes and reflections.
The joint display framework made it possible to integrate qualitative and quantitative
strands of data through comparing their findings, explaining one type of results with the other and
finding out the discrepancies in the experiences of the respondents and statistical data.
Ethical Considerations
Given that a substantial proportion of the participants were between 12 and 18 years of age,
the protection of minors was an integral component of the research design. The study followed
internationally recognized principles for research involving human participants, with particular
attention to informed consent and assent, voluntary participation, confidentiality, data protection,
and participants’ right to withdraw from the study without penalty.
Prior to data collection, authorization to conduct the study was obtained through the
institutional procedures established by the participating school. Because the study involved
minors, a two-level consent procedure was implemented. Written informed consent was obtained
from parents or legal guardians, while students provided their own assent to participate. The
consent and assent forms were provided in Spanish and explained the purpose and procedures of
the study, the use and protection of the collected data, the voluntary nature of participation, and
the right to withdraw without academic penalty. Only students for whom parental or guardian
consent and student assent had been obtained were included in the research procedures.
To protect participants’ identities, alphanumeric codes were assigned to all research
records. Personally identifying information was stored separately from the research data in a
protected file accessible only to the research team. Research data were securely stored and
managed in accordance with the data-protection and retention requirements established by the
participating institution.
RESULTS
Descriptive Analysis of the Pre-test and Post-test
Table 3 shows the summary statistics for the pre-test and post-test among the overall sample
(n = 145), comprising the scores on all of the five dimensions of writing. Before the intervention,
students scored an average Total Score of 51.4 (SD = 12.3) from 100. At the end of the 12-week
period, the average score increased to 67.8 (SD = 10.7) after the intervention, which is indicative
of an increase of 16.4 in their scores in the pre-test and post-test comparison. It provides a
preliminary picture of the effects of the intervention. Further statistical analysis seeks to examine
whether there is a significant effect, what the size of the effect is, and whether the effect can be
observed at the Lower and Higher-Order dimensions level separately.
Vol. 13/ Núm. 3 2026 pág. 1523
Table 3
Descriptive Statistics for Rubric Domain Scores: Pre-test and Post-test (n = 145)
Domain Max. Pre M (SD) Post M (SD) ΔM t df p Cohen's d
Content 30 18.2 (4.1) 22.1 (3.4) +3.9 10.24 144 < .001 0.72
Organization 20 12.4 (3.2) 15.8 (2.7) +3.4 9.87 144 < .001 0.68
Vocabulary 20 10.3 (3.5) 14.2 (2.9) +3.9 13.45 144 < .001 1.05
Grammar 25 8.1 (3.8) 12.4 (3.2) +4.3 14.71 144 < .001 1.12
Mechanics 5 2.4 (0.9) 3.3 (0.8) +0.9 11.33 144 < .001 0.88
Total Score 100 51.4 (12.3) 67.8 (10.7) +16.4 14.23 144 < .001 0.90
Note. M = mean; SD = standard deviation; ΔM = post-test minus pre-test mean gain; t = paired-samples t-statistic; df
= degrees of freedom; p = two-tailed significance value; Cohen's d interpreted against Plonsky & Oswald (2014) EFL
benchmarks: small = 0.40, medium = 0.70, large = 1.00. Bonferroni-corrected α = .01 applied to domain-level
comparisons.
It becomes clear that all dimensions of the individual writing have similar dynamics. It
should be noted that there were the most impressive improvements of Lower-Order Concerns
(LOCs): Grammar was improved to d = 1.12, while Vocabulary gained a result of d = 1.05.
Additionally, Mechanics showed improvement to the extent of d = 0.88. Higher-Order Concerns
(HOCs) also became better but in comparison with LOCs, this progress is less obvious. Content
showed a result of d = 0.72, and Organization improved to d = 0.68.
It should be mentioned that the discrepancy is especially noticeable in the case of LOCs vs.
HOCs: The largest effects were observed for Grammar (d = 1.12) and Vocabulary (d = 1.05),
followed by Mechanics (d = 0.88). However, students improved both surface-level and deeper
dimensions of their writing: both Content and Organization dimensions showed some progress
despite the difference in d values. In general, all five dimensions, as well as the Total Score value
became significantly higher after the intervention under the Bonferroni-adjusted criterion of
significance (α = .01).
The distribution of scores was normally distributed almost for all groups. One exception
should be discussed here. There were no normal distributions in the 9th Grade EGB group (n =
18) because the Shapiro–Wilk test reported deviation from normality (p = .038). Thus, Wilcoxon
signed-rank test was used in the analysis for this group. The significant difference in the Total
Score (Z = 3.68, p < .001) means that the same pattern of improvement can be traced even with
the use of the non-parametric statistical procedure.
ANCOVA has helped to get more details about differences between various classroom
cohorts. While controlling for the pretest writing scores, the classroom cohort had a significant
impact on post-test performance of students, F(5, 138) = 8.74, p < .001, partial η² = .24. The result
shows that there are still differences between various classroom cohorts of students even after
taking into account writing performance of participants.
Table 3
Descriptive Statistics for Rubric Domain Scores: Pre-test and Post-test (n = 145)
Domain Max. Pre M (SD) Post M (SD) ΔM t df p Cohen's d
Content 30 18.2 (4.1) 22.1 (3.4) +3.9 10.24 144 < .001 0.72
Organization 20 12.4 (3.2) 15.8 (2.7) +3.4 9.87 144 < .001 0.68
Vocabulary 20 10.3 (3.5) 14.2 (2.9) +3.9 13.45 144 < .001 1.05
Grammar 25 8.1 (3.8) 12.4 (3.2) +4.3 14.71 144 < .001 1.12
Mechanics 5 2.4 (0.9) 3.3 (0.8) +0.9 11.33 144 < .001 0.88
Total Score 100 51.4 (12.3) 67.8 (10.7) +16.4 14.23 144 < .001 0.90
Note. M = mean; SD = standard deviation; ΔM = post-test minus pre-test mean gain; t = paired-samples t-statistic; df
= degrees of freedom; p = two-tailed significance value; Cohen's d interpreted against Plonsky & Oswald (2014) EFL
benchmarks: small = 0.40, medium = 0.70, large = 1.00. Bonferroni-corrected α = .01 applied to domain-level
comparisons.
It becomes clear that all dimensions of the individual writing have similar dynamics. It
should be noted that there were the most impressive improvements of Lower-Order Concerns
(LOCs): Grammar was improved to d = 1.12, while Vocabulary gained a result of d = 1.05.
Additionally, Mechanics showed improvement to the extent of d = 0.88. Higher-Order Concerns
(HOCs) also became better but in comparison with LOCs, this progress is less obvious. Content
showed a result of d = 0.72, and Organization improved to d = 0.68.
It should be mentioned that the discrepancy is especially noticeable in the case of LOCs vs.
HOCs: The largest effects were observed for Grammar (d = 1.12) and Vocabulary (d = 1.05),
followed by Mechanics (d = 0.88). However, students improved both surface-level and deeper
dimensions of their writing: both Content and Organization dimensions showed some progress
despite the difference in d values. In general, all five dimensions, as well as the Total Score value
became significantly higher after the intervention under the Bonferroni-adjusted criterion of
significance (α = .01).
The distribution of scores was normally distributed almost for all groups. One exception
should be discussed here. There were no normal distributions in the 9th Grade EGB group (n =
18) because the Shapiro–Wilk test reported deviation from normality (p = .038). Thus, Wilcoxon
signed-rank test was used in the analysis for this group. The significant difference in the Total
Score (Z = 3.68, p < .001) means that the same pattern of improvement can be traced even with
the use of the non-parametric statistical procedure.
ANCOVA has helped to get more details about differences between various classroom
cohorts. While controlling for the pretest writing scores, the classroom cohort had a significant
impact on post-test performance of students, F(5, 138) = 8.74, p < .001, partial η² = .24. The result
shows that there are still differences between various classroom cohorts of students even after
taking into account writing performance of participants.
Vol. 13/ Núm. 3 2026 pág. 1524
The Bonferroni-adjusted post-hoc test showed that students in 3rd Year BGU scored
significantly higher adjusted post-test scores compared to those from 8th Grade EGB. The
difference in mean values between these two groups was 11.3 points (95% CI [7.2, 15.4]).
It should be highlighted that the improvement was observed not only among students in the
upper-grade cohorts. All six groups of students were at different stages of English acquisition at
the beginning and showed diverse writing skills at the end of intervention. However, all students
made significant progress during the experiment.Table 4 presents descriptive statistics for the
three questionnaire constructs measured at pre- and post-intervention.
Table 4
Questionnaire Construct Composite Scores: Pre- and Post-Intervention (n = 145)
Construct Items α Pre M
(SD)
Post M
(SD)
ΔM p Cohen's
d
Writing Anxiety
(SLWAI)
5 .82 3.42
(0.73)
2.18
(0.61)
−1.24 < .001 1.35
Learner Autonomy 5 .79 2.89
(0.68)
3.76
(0.57)
+0.87 < .001 0.92
Perceived Usefulness
(PU)
4 .88 — 4.12
(0.64)
— — —
Perceived Ease of Use
(PEOU)
4 .91 — 4.47
(0.52)
— — —
Note. α = Cronbach's alpha internal consistency coefficient; M = mean composite score (5-point Likert); SD = standard
deviation; ΔM = post-minus-pre mean change. SLWAI: higher scores = greater anxiety; post-intervention reduction
indicates anxiety amelioration. Writing anxiety and learner autonomy items were administered at both pre-intervention
and post-intervention, whereas the TAM constructs (PU and PEOU) were administered only after the intervention
because they assessed students’ perceptions of LuzIA following actual use of the tool.
Correlation analyses showed a moderate positive association between Perceived
Usefulness and improvement in writing scores (r = .48, p < .001), indicating that students who
perceived LuzIA as more useful tended to show greater gains in writing performance. Perceived
Ease of Use was also positively associated with Perceived Usefulness (r = .39, p < .001),
indicating that students who perceived LuzIA as easier to use also tended to report greater
perceived usefulness.
DISCUSSION
As shown in the table of results above, there is a definite trend regarding writing scores,
questionnaire outcomes, and students' reports about experiences with the AI-based tool during the
intervention. These results indicate where the support provided by LuzIA worked best and where
it proved less useful.
Positive effects occurred most strongly in the Lower-Order Concerns (LOCs), particularly
Grammar (d = 1.12), Vocabulary (d = 1.05), and Mechanics (d = 0.88). This pattern suggests that
LuzIA may have been particularly useful in providing immediate language-level support.
Previous research has similarly shown that LLM-based writing tools can provide immediate
feedback on grammar, vocabulary use, sentence structure, and alternative phrasing, creating
The Bonferroni-adjusted post-hoc test showed that students in 3rd Year BGU scored
significantly higher adjusted post-test scores compared to those from 8th Grade EGB. The
difference in mean values between these two groups was 11.3 points (95% CI [7.2, 15.4]).
It should be highlighted that the improvement was observed not only among students in the
upper-grade cohorts. All six groups of students were at different stages of English acquisition at
the beginning and showed diverse writing skills at the end of intervention. However, all students
made significant progress during the experiment.Table 4 presents descriptive statistics for the
three questionnaire constructs measured at pre- and post-intervention.
Table 4
Questionnaire Construct Composite Scores: Pre- and Post-Intervention (n = 145)
Construct Items α Pre M
(SD)
Post M
(SD)
ΔM p Cohen's
d
Writing Anxiety
(SLWAI)
5 .82 3.42
(0.73)
2.18
(0.61)
−1.24 < .001 1.35
Learner Autonomy 5 .79 2.89
(0.68)
3.76
(0.57)
+0.87 < .001 0.92
Perceived Usefulness
(PU)
4 .88 — 4.12
(0.64)
— — —
Perceived Ease of Use
(PEOU)
4 .91 — 4.47
(0.52)
— — —
Note. α = Cronbach's alpha internal consistency coefficient; M = mean composite score (5-point Likert); SD = standard
deviation; ΔM = post-minus-pre mean change. SLWAI: higher scores = greater anxiety; post-intervention reduction
indicates anxiety amelioration. Writing anxiety and learner autonomy items were administered at both pre-intervention
and post-intervention, whereas the TAM constructs (PU and PEOU) were administered only after the intervention
because they assessed students’ perceptions of LuzIA following actual use of the tool.
Correlation analyses showed a moderate positive association between Perceived
Usefulness and improvement in writing scores (r = .48, p < .001), indicating that students who
perceived LuzIA as more useful tended to show greater gains in writing performance. Perceived
Ease of Use was also positively associated with Perceived Usefulness (r = .39, p < .001),
indicating that students who perceived LuzIA as easier to use also tended to report greater
perceived usefulness.
DISCUSSION
As shown in the table of results above, there is a definite trend regarding writing scores,
questionnaire outcomes, and students' reports about experiences with the AI-based tool during the
intervention. These results indicate where the support provided by LuzIA worked best and where
it proved less useful.
Positive effects occurred most strongly in the Lower-Order Concerns (LOCs), particularly
Grammar (d = 1.12), Vocabulary (d = 1.05), and Mechanics (d = 0.88). This pattern suggests that
LuzIA may have been particularly useful in providing immediate language-level support.
Previous research has similarly shown that LLM-based writing tools can provide immediate
feedback on grammar, vocabulary use, sentence structure, and alternative phrasing, creating
Vol. 13/ Núm. 3 2026 pág. 1525
opportunities for learners to revise their writing (Song & Song, 2023). More broadly, the stronger
post-revision writing performance reported by Tsai et al. (2024) provides additional evidence that
ChatGPT-assisted revision can contribute to improved assessed writing performance, although
their study did not establish the same LOC-specific effect pattern observed in the present research.
This trend was also illustrated by qualitative findings as learners stated grammar correction
and vocabulary substitution as primary motives for using LuzIA in reflective journals and
interviews. Requesting assistance related to argumentation and essay organization appeared in
students' responses too, yet not consistently.
In Higher-Level Concerns category the improvement is meaningful yet smaller, with d =
0.72 in terms of Content and d = 0.68 with regards to Organization. However, this aspect of the
results is crucial since developing an argument is a completely different matter compared to
correcting a sentence. Developing a powerful thesis, a counter-argument, and good paragraphs
require context, purpose, audience, and rhetorical awareness – the areas where the current LLMs
appear to provide less consistent support for higher-order rhetorical decisions.
Qualitative data also confirmed this aspect – while some students found the suggestions
concerning organization helpful (it came in handy when learners did not know how to structure
their arguments and connect their ideas), the others viewed them as too generic or unsuited to the
academic tone.
Thus, the experience with regards to higher-level concerns can be considered
unpredictable, confirming the idea of applying the tool in combination with the teacher, but not
as a replacement. AI can provide immediate language-level support while the teacher plays the
role of helping students develop argumentation and rhetorical skills needed for writing good
essays.
Perhaps, the most impressive result of this study was not related to grammar or rhetoric,
but to emotion. The writing anxiety decreased with the largest measured effect (d = 1.35).
Qualitative findings offer two plausible explanations for this pattern. Firstly, it is possible
that learners could communicate with the tool without feeling immediate criticism of their
mistakes since they had private interaction with it. Secondly, it is possible to correct errors in
numerous revisions – revise a sentence, submit the next version, reflect on feedback and continue
working without exposing failed attempts to public judgment.
The opportunity to make mistakes in a private and non-judgmental environment may be
particularly important in high-stakes writing contexts. This interpretation is consistent with
Siagian et al. (2026), who reported that EFL learners using DOLA AI for speaking practice
experienced reduced fear of negative evaluation and greater emotional comfort during AI-
mediated interaction. A similar mechanism may help explain the reduction in writing anxiety
observed in the present study, as learners were able to experiment with language and receive
opportunities for learners to revise their writing (Song & Song, 2023). More broadly, the stronger
post-revision writing performance reported by Tsai et al. (2024) provides additional evidence that
ChatGPT-assisted revision can contribute to improved assessed writing performance, although
their study did not establish the same LOC-specific effect pattern observed in the present research.
This trend was also illustrated by qualitative findings as learners stated grammar correction
and vocabulary substitution as primary motives for using LuzIA in reflective journals and
interviews. Requesting assistance related to argumentation and essay organization appeared in
students' responses too, yet not consistently.
In Higher-Level Concerns category the improvement is meaningful yet smaller, with d =
0.72 in terms of Content and d = 0.68 with regards to Organization. However, this aspect of the
results is crucial since developing an argument is a completely different matter compared to
correcting a sentence. Developing a powerful thesis, a counter-argument, and good paragraphs
require context, purpose, audience, and rhetorical awareness – the areas where the current LLMs
appear to provide less consistent support for higher-order rhetorical decisions.
Qualitative data also confirmed this aspect – while some students found the suggestions
concerning organization helpful (it came in handy when learners did not know how to structure
their arguments and connect their ideas), the others viewed them as too generic or unsuited to the
academic tone.
Thus, the experience with regards to higher-level concerns can be considered
unpredictable, confirming the idea of applying the tool in combination with the teacher, but not
as a replacement. AI can provide immediate language-level support while the teacher plays the
role of helping students develop argumentation and rhetorical skills needed for writing good
essays.
Perhaps, the most impressive result of this study was not related to grammar or rhetoric,
but to emotion. The writing anxiety decreased with the largest measured effect (d = 1.35).
Qualitative findings offer two plausible explanations for this pattern. Firstly, it is possible
that learners could communicate with the tool without feeling immediate criticism of their
mistakes since they had private interaction with it. Secondly, it is possible to correct errors in
numerous revisions – revise a sentence, submit the next version, reflect on feedback and continue
working without exposing failed attempts to public judgment.
The opportunity to make mistakes in a private and non-judgmental environment may be
particularly important in high-stakes writing contexts. This interpretation is consistent with
Siagian et al. (2026), who reported that EFL learners using DOLA AI for speaking practice
experienced reduced fear of negative evaluation and greater emotional comfort during AI-
mediated interaction. A similar mechanism may help explain the reduction in writing anxiety
observed in the present study, as learners were able to experiment with language and receive
Vol. 13/ Núm. 3 2026 pág. 1526
support without immediate peer or teacher evaluation. In certification-oriented writing contexts,
this private space may be particularly relevant because written errors remain visible and may
subsequently be evaluated.
Autonomy of students also improved significantly (d = 0.92). The reason why it is an
important point is the ultimate goal of introducing an AI into writing course, which is making the
students less dependent on it in the long run.
According to reflective journals, students began to move toward increased autonomy in the
later stages of the intervention. During the last two weeks, students started to describe themselves
doing grammar check, vocabulary search and organization review independently without
immediately resorting to LuzIA.
This finding is consistent with the developmental path proposed by Vygotsky's ZPD and
learner autonomy concept of Holec (1981). An external support can trigger this path and some
strategies which have been developed under such conditions can slowly become part of students'
practices. The conclusion is not about discontinuing using LuzIA after 12 weeks but rather about
the evolution of students' practices: students begin to utilize the tool not automatically looking up
the answers there but using it among the other tools available to them.
It is important to mention results of TAM questionnaire. Perceived ease of use turned out
to be high (M = 4.47/5), which confirms the idea of decreasing the effort required for adopting
the educational tool through its access through already familiar platform (WhatsApp).
Perceived usefulness was also high (M = 4.12/5) and was positively associated with writing
improvement (r = .48, p < .001). Students reporting greater perceived usefulness tended to show
larger gains in their writing scores. Perceived ease of use was also positively associated with
perceived usefulness (r = .39, p < .001), which is consistent with Davis's (1989) TAM.
Thus, it turns out that usefulness of LuzIA does not consist in excelling in all components
of academic writing, but consists in the fact that it provides the language-level support along with
reduction of anxiety in students and their increased independence in the later stages of revision
practices. Teacher remains irreplaceable in helping with more complex rhetorical questions and
decision-making.
CONCLUSIONS
This research analyzes the effect of using LuzIA, an artificial intelligence assistant based
on the use of GPT and accessible on WhatsApp for improving academic writing, writing anxiety,
and learner autonomy among EFL learners in a multilingual secondary-school environment in
Ecuador. Combining quantitative results with the learners' experiences leads to three key
conclusions.
To begin with, the use of LuzIA is related to statistically significant improvements in the
area of academic writing skills. In this case, the strongest increases were noticed in Lower-Order
support without immediate peer or teacher evaluation. In certification-oriented writing contexts,
this private space may be particularly relevant because written errors remain visible and may
subsequently be evaluated.
Autonomy of students also improved significantly (d = 0.92). The reason why it is an
important point is the ultimate goal of introducing an AI into writing course, which is making the
students less dependent on it in the long run.
According to reflective journals, students began to move toward increased autonomy in the
later stages of the intervention. During the last two weeks, students started to describe themselves
doing grammar check, vocabulary search and organization review independently without
immediately resorting to LuzIA.
This finding is consistent with the developmental path proposed by Vygotsky's ZPD and
learner autonomy concept of Holec (1981). An external support can trigger this path and some
strategies which have been developed under such conditions can slowly become part of students'
practices. The conclusion is not about discontinuing using LuzIA after 12 weeks but rather about
the evolution of students' practices: students begin to utilize the tool not automatically looking up
the answers there but using it among the other tools available to them.
It is important to mention results of TAM questionnaire. Perceived ease of use turned out
to be high (M = 4.47/5), which confirms the idea of decreasing the effort required for adopting
the educational tool through its access through already familiar platform (WhatsApp).
Perceived usefulness was also high (M = 4.12/5) and was positively associated with writing
improvement (r = .48, p < .001). Students reporting greater perceived usefulness tended to show
larger gains in their writing scores. Perceived ease of use was also positively associated with
perceived usefulness (r = .39, p < .001), which is consistent with Davis's (1989) TAM.
Thus, it turns out that usefulness of LuzIA does not consist in excelling in all components
of academic writing, but consists in the fact that it provides the language-level support along with
reduction of anxiety in students and their increased independence in the later stages of revision
practices. Teacher remains irreplaceable in helping with more complex rhetorical questions and
decision-making.
CONCLUSIONS
This research analyzes the effect of using LuzIA, an artificial intelligence assistant based
on the use of GPT and accessible on WhatsApp for improving academic writing, writing anxiety,
and learner autonomy among EFL learners in a multilingual secondary-school environment in
Ecuador. Combining quantitative results with the learners' experiences leads to three key
conclusions.
To begin with, the use of LuzIA is related to statistically significant improvements in the
area of academic writing skills. In this case, the strongest increases were noticed in Lower-Order
Vol. 13/ Núm. 3 2026 pág. 1527
Concerns (LOCs), especially in such areas as Grammar (d = 1.12) and Vocabulary (d = 1.05).
Moreover, Content (d = 0.72) and Organization (d = 0.68) also become better. These effect sizes
can be compared positively to the benchmarks for EFL interventions provided by Plonsky &
Oswald (2014).
It is also important that different improvements for LOCs and HOCs have particular
pedagogical meaning. It appears that LuzIA is beneficial to deal with writing issues that many
learners find difficult to monitor while developing an argument – grammar, vocabulary, spelling,
and other linguistic issues. In this way, with the help of some of such tasks being delegated to AI,
students can develop the ability to pay more attention to the intended messages and structures of
texts. However, the data do not imply that the role of AI will allow replacing the entire writing
process as it still needs contextual decision-making in which teacher intervention is indispensable.
Therefore, the most beneficial role of AI in the case of LuzIA is to serve as assistance for teachers.
Secondly, it is worth noticing that using LuzIA causes the reduction of writing anxiety (d
= 1.35). In this case, it is essential to consider the fact that it does not merely reflect the negative
feeling but also the decision about writing. Moreover, the anxiety can affect the extent to which
a student will exercise his/her writing abilities. Lack of exercises will lead to poor improvement,
as it was mentioned before.
In fact, LuzIA allows giving students a personal space, which is very difficult to create in
a crowded classroom, where they can try, ask, correct mistakes, etc. In this way, the opportunity
to deal with a text multiple times seems to cause the relief related to the process of academic
writing. So, it means that not only the corrections can be valuable in the case of this tool but also
the positive feeling related to writing and correcting the text can encourage more practice.
Finally, one more important thing to mention is the increase in learner autonomy (d = 0.92).
In this case, qualitative analysis shows a specific dynamic in terms of the development of the use
of strategies: in the second half of the intervention, students start reporting their use of
independently performed strategy that they initially practiced with the help of LuzIA (like
checking grammatical issues, searching for alternatives of vocabulary and the organization
structure).
This does not mean the automatic elimination of AI dependence. The dependence depends
on how the tool is introduced and used in the class. In the current research, the approach implies
that students needed to question the answer of AI, decide which suggestion is suitable for him/her,
etc. So, it means that simple acceptance of AI advice can decrease thinking required in writing.
Otherwise, analyzing and implementing the feedback is able to be included in the process of
learning revision strategies.
What can be considered the distinctive feature of the current study is that LuzIA operates
via WhatsApp, a tool familiar for all students. In this way, there is no need to introduce any new
interface and platform as there is no special software that requires installation. Moreover, there is
Concerns (LOCs), especially in such areas as Grammar (d = 1.12) and Vocabulary (d = 1.05).
Moreover, Content (d = 0.72) and Organization (d = 0.68) also become better. These effect sizes
can be compared positively to the benchmarks for EFL interventions provided by Plonsky &
Oswald (2014).
It is also important that different improvements for LOCs and HOCs have particular
pedagogical meaning. It appears that LuzIA is beneficial to deal with writing issues that many
learners find difficult to monitor while developing an argument – grammar, vocabulary, spelling,
and other linguistic issues. In this way, with the help of some of such tasks being delegated to AI,
students can develop the ability to pay more attention to the intended messages and structures of
texts. However, the data do not imply that the role of AI will allow replacing the entire writing
process as it still needs contextual decision-making in which teacher intervention is indispensable.
Therefore, the most beneficial role of AI in the case of LuzIA is to serve as assistance for teachers.
Secondly, it is worth noticing that using LuzIA causes the reduction of writing anxiety (d
= 1.35). In this case, it is essential to consider the fact that it does not merely reflect the negative
feeling but also the decision about writing. Moreover, the anxiety can affect the extent to which
a student will exercise his/her writing abilities. Lack of exercises will lead to poor improvement,
as it was mentioned before.
In fact, LuzIA allows giving students a personal space, which is very difficult to create in
a crowded classroom, where they can try, ask, correct mistakes, etc. In this way, the opportunity
to deal with a text multiple times seems to cause the relief related to the process of academic
writing. So, it means that not only the corrections can be valuable in the case of this tool but also
the positive feeling related to writing and correcting the text can encourage more practice.
Finally, one more important thing to mention is the increase in learner autonomy (d = 0.92).
In this case, qualitative analysis shows a specific dynamic in terms of the development of the use
of strategies: in the second half of the intervention, students start reporting their use of
independently performed strategy that they initially practiced with the help of LuzIA (like
checking grammatical issues, searching for alternatives of vocabulary and the organization
structure).
This does not mean the automatic elimination of AI dependence. The dependence depends
on how the tool is introduced and used in the class. In the current research, the approach implies
that students needed to question the answer of AI, decide which suggestion is suitable for him/her,
etc. So, it means that simple acceptance of AI advice can decrease thinking required in writing.
Otherwise, analyzing and implementing the feedback is able to be included in the process of
learning revision strategies.
What can be considered the distinctive feature of the current study is that LuzIA operates
via WhatsApp, a tool familiar for all students. In this way, there is no need to introduce any new
interface and platform as there is no special software that requires installation. Moreover, there is
Vol. 13/ Núm. 3 2026 pág. 1528
no additional platform that should be integrated in the class before usage. This aspect can be very
helpful to use the technology as there are cases in which schools are unable to afford it.
Thus, the main finding of the current research is that technology can provide effective
support of EFL learners in improving language-level accuracy of the academic texts while
teachers still bear full responsibility for developing arguments and rhetorical issues of students.
In fact, the goal of this use is not to train students to write properly only when they use AI but
make the process such as they will become better writers through AI support.
Many important issues need further investigation. Future research is supposed to
investigate whether such effects will last if students have reduced access to LuzIA, to compare
LuzIA and alternative AI writing assistants under similar conditions, and to conduct the
randomized controlled design of the current intervention where possible in such an environment.
In addition, longitudinal studies will be extremely useful to check if the learned strategies become
autonomous.
no additional platform that should be integrated in the class before usage. This aspect can be very
helpful to use the technology as there are cases in which schools are unable to afford it.
Thus, the main finding of the current research is that technology can provide effective
support of EFL learners in improving language-level accuracy of the academic texts while
teachers still bear full responsibility for developing arguments and rhetorical issues of students.
In fact, the goal of this use is not to train students to write properly only when they use AI but
make the process such as they will become better writers through AI support.
Many important issues need further investigation. Future research is supposed to
investigate whether such effects will last if students have reduced access to LuzIA, to compare
LuzIA and alternative AI writing assistants under similar conditions, and to conduct the
randomized controlled design of the current intervention where possible in such an environment.
In addition, longitudinal studies will be extremely useful to check if the learned strategies become
autonomous.
Vol. 13/ Núm. 3 2026 pág. 1529
REFERENCES
Ayala-Pazmiño, M. F., & Alvarado-Lucas, K. I. (2023). Integración de la inteligencia artificial en
la educación del idioma inglés en Ecuador: Un camino para mejorar los resultados del
aprendizaje. 593 Digital Publisher CEIT, 8(3-1), 679–687.
https://doi.org/10.33386/593dp.2023.3-1.1862
Behforouz, B., & Al Ghaithi, A. (2024). The impact of using interactive chatbots on self-directed
learning. Studies in Self-Access Learning Journal, 15(3), 317–344.
https://doi.org/10.37237/150302
Braun, V., & Clarke, V. (2022). Thematic analysis: A practical guide. SAGE Publications.
Cheng, Y.-S. (2004). A measure of second language writing anxiety: Scale development and
preliminary validation. Journal of Second Language Writing, 13(4), 313–335.
https://doi.org/10.1016/j.jslw.2004.07.001
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence
Erlbaum Associates.
Creswell, J. W., & Creswell, J. D. (2023). Research design: Qualitative, quantitative, and mixed
methods approaches (6th ed.). SAGE Publications.
Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of
information technology. MIS Quarterly, 13(3), 319–340. https://doi.org/10.2307/249008
Ghafar, Z. N., & Raheem, B. R. (2025). Understanding EFL learners’ academic writing problems
and the benefits of artificial intelligence tools: A descriptive review. Journal of Language
and Pragmatics Studies, 4(3), 187–193. https://doi.org/10.58881/jlps.v4i3.141
Holec, H. (1981). Autonomy and foreign language learning. Pergamon Press.
Isabella, P., Ferdiyanto, F., & Zainuddin, Z. (2026). Implementation of Dola Artificial
Intelligence as a learning tool to improve students’ English speaking skills. Journal of
English Language and Education, 11(2), 640–650.
https://doi.org/10.31004/jele.v11i2.2264
Jacobs, H. L., Zinkgraf, S. A., Wormuth, D. R., Hartfiel, V. F., & Hughey, J. B. (1981). Testing
ESL composition: A practical approach. Newbury House.
Krashen, S. D. (1982). Principles and practice in second language acquisition. Pergamon Press.
Li, M., & Wilson, J. (2025). AI-integrated scaffolding to enhance agency and creativity in K-12
English language learners: A systematic review. Information, 16(7), 519.
https://doi.org/10.3390/info16070519
LuzIA. (n.d.-a). LuzIA: Your intelligent assistant at a click. https://www.luzia.com/en
LuzIA. (n.d.-b). Help: Frequently asked questions. https://www.luzia.com/ayuda
REFERENCES
Ayala-Pazmiño, M. F., & Alvarado-Lucas, K. I. (2023). Integración de la inteligencia artificial en
la educación del idioma inglés en Ecuador: Un camino para mejorar los resultados del
aprendizaje. 593 Digital Publisher CEIT, 8(3-1), 679–687.
https://doi.org/10.33386/593dp.2023.3-1.1862
Behforouz, B., & Al Ghaithi, A. (2024). The impact of using interactive chatbots on self-directed
learning. Studies in Self-Access Learning Journal, 15(3), 317–344.
https://doi.org/10.37237/150302
Braun, V., & Clarke, V. (2022). Thematic analysis: A practical guide. SAGE Publications.
Cheng, Y.-S. (2004). A measure of second language writing anxiety: Scale development and
preliminary validation. Journal of Second Language Writing, 13(4), 313–335.
https://doi.org/10.1016/j.jslw.2004.07.001
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence
Erlbaum Associates.
Creswell, J. W., & Creswell, J. D. (2023). Research design: Qualitative, quantitative, and mixed
methods approaches (6th ed.). SAGE Publications.
Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of
information technology. MIS Quarterly, 13(3), 319–340. https://doi.org/10.2307/249008
Ghafar, Z. N., & Raheem, B. R. (2025). Understanding EFL learners’ academic writing problems
and the benefits of artificial intelligence tools: A descriptive review. Journal of Language
and Pragmatics Studies, 4(3), 187–193. https://doi.org/10.58881/jlps.v4i3.141
Holec, H. (1981). Autonomy and foreign language learning. Pergamon Press.
Isabella, P., Ferdiyanto, F., & Zainuddin, Z. (2026). Implementation of Dola Artificial
Intelligence as a learning tool to improve students’ English speaking skills. Journal of
English Language and Education, 11(2), 640–650.
https://doi.org/10.31004/jele.v11i2.2264
Jacobs, H. L., Zinkgraf, S. A., Wormuth, D. R., Hartfiel, V. F., & Hughey, J. B. (1981). Testing
ESL composition: A practical approach. Newbury House.
Krashen, S. D. (1982). Principles and practice in second language acquisition. Pergamon Press.
Li, M., & Wilson, J. (2025). AI-integrated scaffolding to enhance agency and creativity in K-12
English language learners: A systematic review. Information, 16(7), 519.
https://doi.org/10.3390/info16070519
LuzIA. (n.d.-a). LuzIA: Your intelligent assistant at a click. https://www.luzia.com/en
LuzIA. (n.d.-b). Help: Frequently asked questions. https://www.luzia.com/ayuda
Vol. 13/ Núm. 3 2026 pág. 1530
Maforah, N., & Leburu-Masigo, G. (2018). Application of the mixed methods research using
sequential explanatory design. In ICERI2018 Proceedings (pp. 9710–9715). IATED.
https://doi.org/10.21125/iceri.2018.0797
Marzuki, M., Widiati, U., Rusdin, D., Darwin, D., & Indrawati, I. (2023). The impact of AI
writing tools on the content and organization of students’ writing: EFL teachers’
perspective. Cogent Education, 10(2), 2236469.
https://doi.org/10.1080/2331186X.2023.2236469
Nunnally, J. C. (1978). Psychometric theory (2nd ed.). McGraw-Hill.
Plonsky, L., & Oswald, F. L. (2014). How big is “big”? Interpreting effect sizes in L2 research.
Language Learning, 64(4), 878–912. https://doi.org/10.1111/lang.12079
Sanchez, H. S. (2024). Teacher-written feedback in L2 writing education: Insights into student
tensions and responses. Feedback Research in Second Language, 2, 58–80.
https://doi.org/10.32038/frsl.2024.02.04
Siagian, L. P., Kharisma, A. J., & Syahputri, D. (2026). Mitigating EFL speaking anxiety with
Dola AI: Enhancing oral proficiency. Journal of Social, Humanity, and Education, 6(2),
255–266. https://doi.org/10.35912/jshe.v6i2.3926
Song, C., & Song, Y. (2023). Enhancing academic writing skills and motivation: Assessing the
efficacy of ChatGPT in AI-assisted language learning for EFL students. Frontiers in
Psychology, 14, 1260843. https://doi.org/10.3389/fpsyg.2023.1260843
Tsai, C.-Y., Lin, Y.-T., & Brown, I. K. (2024). Impacts of ChatGPT-assisted writing for EFL
English majors: Feasibility and challenges. Education and Information Technologies,
29(17), 22427–22445. https://doi.org/10.1007/s10639-024-12722-y
Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes.
Harvard University Press.
Maforah, N., & Leburu-Masigo, G. (2018). Application of the mixed methods research using
sequential explanatory design. In ICERI2018 Proceedings (pp. 9710–9715). IATED.
https://doi.org/10.21125/iceri.2018.0797
Marzuki, M., Widiati, U., Rusdin, D., Darwin, D., & Indrawati, I. (2023). The impact of AI
writing tools on the content and organization of students’ writing: EFL teachers’
perspective. Cogent Education, 10(2), 2236469.
https://doi.org/10.1080/2331186X.2023.2236469
Nunnally, J. C. (1978). Psychometric theory (2nd ed.). McGraw-Hill.
Plonsky, L., & Oswald, F. L. (2014). How big is “big”? Interpreting effect sizes in L2 research.
Language Learning, 64(4), 878–912. https://doi.org/10.1111/lang.12079
Sanchez, H. S. (2024). Teacher-written feedback in L2 writing education: Insights into student
tensions and responses. Feedback Research in Second Language, 2, 58–80.
https://doi.org/10.32038/frsl.2024.02.04
Siagian, L. P., Kharisma, A. J., & Syahputri, D. (2026). Mitigating EFL speaking anxiety with
Dola AI: Enhancing oral proficiency. Journal of Social, Humanity, and Education, 6(2),
255–266. https://doi.org/10.35912/jshe.v6i2.3926
Song, C., & Song, Y. (2023). Enhancing academic writing skills and motivation: Assessing the
efficacy of ChatGPT in AI-assisted language learning for EFL students. Frontiers in
Psychology, 14, 1260843. https://doi.org/10.3389/fpsyg.2023.1260843
Tsai, C.-Y., Lin, Y.-T., & Brown, I. K. (2024). Impacts of ChatGPT-assisted writing for EFL
English majors: Feasibility and challenges. Education and Information Technologies,
29(17), 22427–22445. https://doi.org/10.1007/s10639-024-12722-y
Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes.
Harvard University Press.