Vol. 13/ Núm. 1 2026 pág. 861
https://doi.org/10.69639/arandu.v13i1.1950
The Use of Artificial Intelligence to Improve Speaking
Fluency in English as a Foreign Language (EFL) Learning
El uso de la Inteligencia Artificial para mejorar la fluidez oral en el aprendizaje
del inglés como lengua extranjera
Freddy Stefano Chacón Guerrero
https://orcid.org/0009-0006-9610-5940
Universidad Bolivariana del Ecuador
Ecuador
Luiggi Andrés Panta Guillén
https://orcid.org/0009-0003-7184-5377
Universidad Bolivariana del Ecuador
Ecuador
Josué Bonilla Tenesaca
https://orcid.org/0000-0002-6748-2345
Universidad Bolivariana del Ecuador
Ecuador
Diana Egas Herrera
https://orcid.org/0000-0003-2878-0689
Universidad Bolivariana del Ecuador
Ecuador
Artículo recibido: 10 diciembre 2025 -Aceptado para publicación: 18 enero 2026
Conflictos de intereses: Ninguno que declarar.
ABSTRACT
This action research examined the extent to which short AI-mediated speaking tasks can enhance
oral performance in an A2 EFL class. Fifteen students, selected by convenience from a regular
English course, completed an eight-session intervention over two weeks in which ChatGPT was
used to simulate guided conversations and provide immediate, text-based feedback. Speaking
performance was measured through a pretest–posttest design using the Cambridge A2 Speaking
Assessment descriptors: grammar and vocabulary, pronunciation, and interactive communication.
Descriptive results showed clear gains in all three dimensions: grammar and vocabulary increased
from M = 2.8 to M = 4.0, interactive communication from M = 3.4 to M = 4.2, and pronunciation
from M = 2.7 to M = 3.5, with reduced dispersion in two of the three areas. Paired-samples t tests
confirmed that the improvements were statistically significant in grammar and vocabulary (p
= .0009, d ≈ .50), pronunciation (p = .0053, d ≈ .40) and interactive communication (p = .0125, d
≈ .50). These findings indicate that AI-powered tools can operate as low-anxiety rehearsal spaces
that promote more frequent oral practice, support self-correction, and increase learners perceived
fluency and willingness to communicate. Although the study was small-scale and context-bound,
Vol. 13/ Núm. 1 2026 pág. 862
it provides empirical support for integrating AI within structured, teacher-mediated speaking
instruction as a complementary pathway to develop cognitive, utterance, and perceived fluency.
Keywords: Artificial Intelligence, speaking fluency, EFL, chatbots, adaptive learning
RESUMEN
Esta investigación–acción examinó hasta qué punto las tareas orales breves mediadas por IA
pueden mejorar el desempeño oral en una clase de inglés como lengua extranjera (A2). Quince
estudiantes, seleccionados por conveniencia de un curso regular de inglés, participaron en una
intervención de ocho sesiones durante dos semanas en la que se utilizó ChatGPT para simular
conversaciones guiadas y proporcionar retroalimentación inmediata por escrito. El desempeño
oral se midió con un diseño pretest–posttest empleando los descriptores de evaluación oral de
Cambridge para el nivel A2: gramática y vocabulario, pronunciación y comunicación interactiva.
Los resultados descriptivos mostraron incrementos claros en las tres dimensiones: gramática y
vocabulario pasó de M = 2.8 a M = 4.0, comunicación interactiva de M = 3.4 a M = 4.2 y
pronunciación de M = 2.7 a M = 3.5, con una disminución de la dispersión en dos de las tres áreas.
Las pruebas t para muestras relacionadas confirmaron que las mejoras fueron estadísticamente
significativas en gramática y vocabulario (p = .0009, d ≈ .50), pronunciación (p = .0053, d ≈ .40)
y comunicación interactiva (p = .0125, d ≈ .50). Estos hallazgos indican que las herramientas
basadas en IA pueden funcionar como espacios de práctica de bajo nivel de ansiedad que
favorecen una mayor frecuencia de producción oral, apoyan la autorregulación y aumentan la
fluidez percibida y la disposición a comunicarse. Aunque el estudio fue de pequeña escala y
situado en un contexto específico, aporta evidencia empírica para integrar la IA en una enseñanza
oral estructurada y mediada por el docente como vía complementaria para desarrollar la fluidez
cognitiva, la fluidez del enunciado y la fluidez percibida.
Palabras claves: inteligencia artificial, fluidez oral, inglés como lengua extranjera,
chatbots, aprendizaje adaptativo
Todo el contenido de la Revista Científica Internacional Arandu UTIC publicado en este sitio está disponible bajo
licencia Creative Commons Atribution 4.0 International.
it provides empirical support for integrating AI within structured, teacher-mediated speaking
instruction as a complementary pathway to develop cognitive, utterance, and perceived fluency.
Keywords: Artificial Intelligence, speaking fluency, EFL, chatbots, adaptive learning
RESUMEN
Esta investigación–acción examinó hasta qué punto las tareas orales breves mediadas por IA
pueden mejorar el desempeño oral en una clase de inglés como lengua extranjera (A2). Quince
estudiantes, seleccionados por conveniencia de un curso regular de inglés, participaron en una
intervención de ocho sesiones durante dos semanas en la que se utilizó ChatGPT para simular
conversaciones guiadas y proporcionar retroalimentación inmediata por escrito. El desempeño
oral se midió con un diseño pretest–posttest empleando los descriptores de evaluación oral de
Cambridge para el nivel A2: gramática y vocabulario, pronunciación y comunicación interactiva.
Los resultados descriptivos mostraron incrementos claros en las tres dimensiones: gramática y
vocabulario pasó de M = 2.8 a M = 4.0, comunicación interactiva de M = 3.4 a M = 4.2 y
pronunciación de M = 2.7 a M = 3.5, con una disminución de la dispersión en dos de las tres áreas.
Las pruebas t para muestras relacionadas confirmaron que las mejoras fueron estadísticamente
significativas en gramática y vocabulario (p = .0009, d ≈ .50), pronunciación (p = .0053, d ≈ .40)
y comunicación interactiva (p = .0125, d ≈ .50). Estos hallazgos indican que las herramientas
basadas en IA pueden funcionar como espacios de práctica de bajo nivel de ansiedad que
favorecen una mayor frecuencia de producción oral, apoyan la autorregulación y aumentan la
fluidez percibida y la disposición a comunicarse. Aunque el estudio fue de pequeña escala y
situado en un contexto específico, aporta evidencia empírica para integrar la IA en una enseñanza
oral estructurada y mediada por el docente como vía complementaria para desarrollar la fluidez
cognitiva, la fluidez del enunciado y la fluidez percibida.
Palabras claves: inteligencia artificial, fluidez oral, inglés como lengua extranjera,
chatbots, aprendizaje adaptativo
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. 1 2026 pág. 863
INTRODUCTION
The integration of Artificial Intelligence (AI) into educational settings has offered
innovative tools to support and enhance communicative competence. Within English as a Foreign
Language (EFL) instruction, AI technologies such as intelligent tutoring systems, speech
recognition applications, and virtual conversation agents have emerged as valuable resources to
improve speaking fluency with their capacity to provide immediate feedback, simulate authentic
dialogues, and personalize learning experiences, presenting new opportunities to address the
persistent challenges associated with oral language development.
This action research study aims to explore the use of an Al-powered tool to enhance
speaking fluency in EFL learners; It explores the impact of these tools on various dimensions of
speaking fluency, including accuracy, speech rate, pronunciation, and learner confidence by
implementing short speaking activities and evaluating their effects on students' oral performance
and confidence.
Speaking skills
Speaking skills involve the capacity to express thoughts, opinions, and emotions clearly
and effectively through verbal communication, both in face-to-face interactions and mediated
contexts. As Rao (2019) underscores, speaking is the most important among the four basic
language skills—listening, speaking, reading, and writing because it directly influences an
individual’s ability to engage in effective communication in a globalized world. He notes that in
the context of international communication, speaking skills serve not only as a tool for interaction
but also as a determinant of professional and academic success.
The acquisition of speaking skills demands more than simple memorization or rote
learning. According to Mishra (2017), the development of speaking competence encompasses
several linguistic elements namely phonetics, stress, pitch, and intonation that shape the clarity,
meaning, and naturalness of spoken language. Phonetics analyze how speech sounds are
produced, transmitted, and received. It is divided into articulatory phonetics (how speech sounds
are produced by the organs of speech), acoustic phonetics (the physical properties of sound
waves), and auditory phonetics (how sounds are perceived by the listener). A foundational
knowledge of phonetics allows learners to articulate sounds accurately, distinguish between
similar phonemes, and improve their overall pronunciation and intelligibility.
Stress refers to the emphasis placed on particular syllables or words within speech.
According to Mishra (2017) stress patterns influence the rhythm of English and often alter
meaning. For instance, the placement of stress in a word can differentiate between a noun and a
verb (e.g., ’record as a noun vs. record as a verb), while stress at the sentence level can change
the focus or implication of a statement. Meanwhile, pitch refers to the perceived highness or
lowness of the voice, to convey emotional nuance and speaker attitude. He notes that factors such
INTRODUCTION
The integration of Artificial Intelligence (AI) into educational settings has offered
innovative tools to support and enhance communicative competence. Within English as a Foreign
Language (EFL) instruction, AI technologies such as intelligent tutoring systems, speech
recognition applications, and virtual conversation agents have emerged as valuable resources to
improve speaking fluency with their capacity to provide immediate feedback, simulate authentic
dialogues, and personalize learning experiences, presenting new opportunities to address the
persistent challenges associated with oral language development.
This action research study aims to explore the use of an Al-powered tool to enhance
speaking fluency in EFL learners; It explores the impact of these tools on various dimensions of
speaking fluency, including accuracy, speech rate, pronunciation, and learner confidence by
implementing short speaking activities and evaluating their effects on students' oral performance
and confidence.
Speaking skills
Speaking skills involve the capacity to express thoughts, opinions, and emotions clearly
and effectively through verbal communication, both in face-to-face interactions and mediated
contexts. As Rao (2019) underscores, speaking is the most important among the four basic
language skills—listening, speaking, reading, and writing because it directly influences an
individual’s ability to engage in effective communication in a globalized world. He notes that in
the context of international communication, speaking skills serve not only as a tool for interaction
but also as a determinant of professional and academic success.
The acquisition of speaking skills demands more than simple memorization or rote
learning. According to Mishra (2017), the development of speaking competence encompasses
several linguistic elements namely phonetics, stress, pitch, and intonation that shape the clarity,
meaning, and naturalness of spoken language. Phonetics analyze how speech sounds are
produced, transmitted, and received. It is divided into articulatory phonetics (how speech sounds
are produced by the organs of speech), acoustic phonetics (the physical properties of sound
waves), and auditory phonetics (how sounds are perceived by the listener). A foundational
knowledge of phonetics allows learners to articulate sounds accurately, distinguish between
similar phonemes, and improve their overall pronunciation and intelligibility.
Stress refers to the emphasis placed on particular syllables or words within speech.
According to Mishra (2017) stress patterns influence the rhythm of English and often alter
meaning. For instance, the placement of stress in a word can differentiate between a noun and a
verb (e.g., ’record as a noun vs. record as a verb), while stress at the sentence level can change
the focus or implication of a statement. Meanwhile, pitch refers to the perceived highness or
lowness of the voice, to convey emotional nuance and speaker attitude. He notes that factors such
Vol. 13/ Núm. 1 2026 pág. 864
as fatigue or surprise can influence a speaker’s pitch, thereby affecting the listener’s interpretation
of the message. For example, a rise in pitch might signal excitement or inquiry, while a lower
pitch suggests seriousness or authority.
Intonation, closely related to both stress and pitch, is described by Mishra (2017) as the
melody of speech. It involves the variation in pitch across an utterance and is used to express
meaning beyond the literal words spoken. Intonation helps indicate the speaker’s intention,
emotional state, or the communicative function of the sentence. There are two primary patterns:
the falling tune, typically used in statements or commands, and the rising tune, often employed in
yes-no questions or expressions of uncertainty. Mastery of intonation is essential for learners to
sound natural and to avoid unintended communicative breakdowns.
Speaking fluency
Speaking fluency in a foreign language extends beyond mere linguistic knowledge to
encompass the smooth, coherent, and rapid production of speech in real-time communicative
situations. According to Richards (2006), fluency involves the use of naturally occurring language
in meaningful interactions were communication flows despite occasional gaps in grammatical
accuracy. This view is reinforced by Baily (2003), who describes fluency as the ability to use
language quickly and confidently with minimal hesitation or unnatural pauses, as a fluent speaker
can produce speech without frequent interruptions to think about what to say next.
According to Ghasemi-Mozaheb (2021), fluency entails both general oral proficiency and
the specific temporal aspects of speech, such as speech rate, pause length, and hesitation.
Disfluent speakers struggle to maintain their interlocutor’s attention, often leading to breakdowns
in communication. The cognitive basis for fluency involves proceduralization where linguistic
knowledge is automatized allowing speakers to allocate cognitive resources efficiently during
real-time speech. Thus, the more automatized a learner’s linguistic knowledge becomes, the less
cognitive effort is needed to produce speech, enabling faster and more fluid communication.
The development of fluency, as conceptualized by Segalowitz (2022), is best understood
through a triadic model comprising cognitive fluency, utterance fluency, and perceived fluency
each representing a distinct aspect of speech production. Cognitive fluency refers to the internal,
mental efficiency with which a speaker accesses and processes language. It involves the speed
and coordination of the cognitive mechanisms such as lexical access, grammatical encoding, and
phonological assembly responsible for planning, retrieving, and articulating speech. In second
language (L2) speakers, cognitive fluency lags behind that of native speakers due to a greater
reliance on controlled processing rather than automatic retrieval. This cognitive load leads to
slower speech, more frequent pauses, and increased self-monitoring.
Utterance fluency, in contrast, focuses on observable and measurable speech
characteristics. It includes quantitative measures such as speech rate (words or syllables per
minute), articulation rate (excluding pauses), mean length of run (number of syllables between
as fatigue or surprise can influence a speaker’s pitch, thereby affecting the listener’s interpretation
of the message. For example, a rise in pitch might signal excitement or inquiry, while a lower
pitch suggests seriousness or authority.
Intonation, closely related to both stress and pitch, is described by Mishra (2017) as the
melody of speech. It involves the variation in pitch across an utterance and is used to express
meaning beyond the literal words spoken. Intonation helps indicate the speaker’s intention,
emotional state, or the communicative function of the sentence. There are two primary patterns:
the falling tune, typically used in statements or commands, and the rising tune, often employed in
yes-no questions or expressions of uncertainty. Mastery of intonation is essential for learners to
sound natural and to avoid unintended communicative breakdowns.
Speaking fluency
Speaking fluency in a foreign language extends beyond mere linguistic knowledge to
encompass the smooth, coherent, and rapid production of speech in real-time communicative
situations. According to Richards (2006), fluency involves the use of naturally occurring language
in meaningful interactions were communication flows despite occasional gaps in grammatical
accuracy. This view is reinforced by Baily (2003), who describes fluency as the ability to use
language quickly and confidently with minimal hesitation or unnatural pauses, as a fluent speaker
can produce speech without frequent interruptions to think about what to say next.
According to Ghasemi-Mozaheb (2021), fluency entails both general oral proficiency and
the specific temporal aspects of speech, such as speech rate, pause length, and hesitation.
Disfluent speakers struggle to maintain their interlocutor’s attention, often leading to breakdowns
in communication. The cognitive basis for fluency involves proceduralization where linguistic
knowledge is automatized allowing speakers to allocate cognitive resources efficiently during
real-time speech. Thus, the more automatized a learner’s linguistic knowledge becomes, the less
cognitive effort is needed to produce speech, enabling faster and more fluid communication.
The development of fluency, as conceptualized by Segalowitz (2022), is best understood
through a triadic model comprising cognitive fluency, utterance fluency, and perceived fluency
each representing a distinct aspect of speech production. Cognitive fluency refers to the internal,
mental efficiency with which a speaker accesses and processes language. It involves the speed
and coordination of the cognitive mechanisms such as lexical access, grammatical encoding, and
phonological assembly responsible for planning, retrieving, and articulating speech. In second
language (L2) speakers, cognitive fluency lags behind that of native speakers due to a greater
reliance on controlled processing rather than automatic retrieval. This cognitive load leads to
slower speech, more frequent pauses, and increased self-monitoring.
Utterance fluency, in contrast, focuses on observable and measurable speech
characteristics. It includes quantitative measures such as speech rate (words or syllables per
minute), articulation rate (excluding pauses), mean length of run (number of syllables between
Vol. 13/ Núm. 1 2026 pág. 865
pauses), pause frequency, and repair frequency (e.g., self-corrections, false starts). Utterance
fluency improves with increased proficiency, as speakers reduce the length and number of pauses
and become more efficient in delivering connected speech. (Tavakoli et al., 2020)
Perceived fluency pertains to how a speaker’s fluency is judged by listeners. It is a
subjective evaluation, made based on how smooth, natural, and effortless the speech sounds.
Interestingly, perceived fluency does not always align perfectly with utterance fluency. For
example, a speaker may have a slower speech rate but still be judged as fluent if their delivery is
coherent, confident, and appropriately paced. Listeners are particularly sensitive to the rhythm,
intonation, hesitation patterns, and the placement of pauses (Segalowitz, 2022). As such, even
minor disfluencies can negatively affect fluency perception, especially if they occur in the middle
of a clause or disrupt the listener’s understanding of meaning
Speaking fluency is hindered by several common challenges: limited vocabulary, lack of
listening and speaking practice, and cognitive barriers such as thinking in the native language
during speech (Alaraj, 2017). Also, learner attitudes and affective factors such as anxiety,
motivation, and mindset influence fluency outcomes. These affective variables do not directly
change a learner’s linguistic competence but significantly influence how that competence is
expressed in real-time communication. Negative beliefs about language ability, often rooted in
fixed mindsets or previous failure experiences, can inhibit participation and risk-taking in oral
tasks (Bárkányi, 2021).
Foreign language anxiety, particularly in speaking contexts, can disrupt the automatic
retrieval of vocabulary and increase self-monitoring, hindering smooth speech production.
According to Pabro-Maquidato (2021), high anxiety levels lead to more frequent pauses,
hesitations, and errors, thus reducing utterance fluency. Students often report being afraid of
making mistakes, especially in front of peers or teachers, and this fear causes them to avoid
speaking altogether; In that regard, fostering a growth mindset and building a supportive
classroom environment helps reduce the fear of pupils of making mistakes.
From a pedagogical perspective, developing fluency requires intentional design of
learning environments and activities, including vocabulary expansion routines and authentic
listening tasks. Brown (2017) emphasizes that effective fluency development involves creating
psychologically safe environments for speaking practice where learners engage with familiar
content and focus on meaning, while experiencing pressure to perform at speed, and are able to
produce large volumes of language.
Use of Artificial Intelligence in EFL
The use of Artificial Intelligence (AI) in English as a Foreign Language (EFL) education
has grown significantly in recent years, enabling more personalized, efficient, and engaging
learning experiences. These advancements are driven by the integration of Natural Language
pauses), pause frequency, and repair frequency (e.g., self-corrections, false starts). Utterance
fluency improves with increased proficiency, as speakers reduce the length and number of pauses
and become more efficient in delivering connected speech. (Tavakoli et al., 2020)
Perceived fluency pertains to how a speaker’s fluency is judged by listeners. It is a
subjective evaluation, made based on how smooth, natural, and effortless the speech sounds.
Interestingly, perceived fluency does not always align perfectly with utterance fluency. For
example, a speaker may have a slower speech rate but still be judged as fluent if their delivery is
coherent, confident, and appropriately paced. Listeners are particularly sensitive to the rhythm,
intonation, hesitation patterns, and the placement of pauses (Segalowitz, 2022). As such, even
minor disfluencies can negatively affect fluency perception, especially if they occur in the middle
of a clause or disrupt the listener’s understanding of meaning
Speaking fluency is hindered by several common challenges: limited vocabulary, lack of
listening and speaking practice, and cognitive barriers such as thinking in the native language
during speech (Alaraj, 2017). Also, learner attitudes and affective factors such as anxiety,
motivation, and mindset influence fluency outcomes. These affective variables do not directly
change a learner’s linguistic competence but significantly influence how that competence is
expressed in real-time communication. Negative beliefs about language ability, often rooted in
fixed mindsets or previous failure experiences, can inhibit participation and risk-taking in oral
tasks (Bárkányi, 2021).
Foreign language anxiety, particularly in speaking contexts, can disrupt the automatic
retrieval of vocabulary and increase self-monitoring, hindering smooth speech production.
According to Pabro-Maquidato (2021), high anxiety levels lead to more frequent pauses,
hesitations, and errors, thus reducing utterance fluency. Students often report being afraid of
making mistakes, especially in front of peers or teachers, and this fear causes them to avoid
speaking altogether; In that regard, fostering a growth mindset and building a supportive
classroom environment helps reduce the fear of pupils of making mistakes.
From a pedagogical perspective, developing fluency requires intentional design of
learning environments and activities, including vocabulary expansion routines and authentic
listening tasks. Brown (2017) emphasizes that effective fluency development involves creating
psychologically safe environments for speaking practice where learners engage with familiar
content and focus on meaning, while experiencing pressure to perform at speed, and are able to
produce large volumes of language.
Use of Artificial Intelligence in EFL
The use of Artificial Intelligence (AI) in English as a Foreign Language (EFL) education
has grown significantly in recent years, enabling more personalized, efficient, and engaging
learning experiences. These advancements are driven by the integration of Natural Language
Vol. 13/ Núm. 1 2026 pág. 866
Processing (NLP), Machine Learning (ML), Artificial Neural Networks (ANNs), and Affective
Computing (AC), (Jiang, 2022).
AI fosters individualized learning through lIntelligent Tutoring Systems (ITSs), AI-
powered chatbots, and Automated Evaluation Systems (AESs). These systems provide learners
with immediate feedback tailored to their language proficiency and learning pace. For example,
chatbots facilitate real-time interaction and help develop oral and written communication skills,
while AESs support writing development through automated grammar correction and content
suggestions (Sumakul et al, 2022). Tools such as ChatGPT, Google Translate, and Grammarly
have been found to improve language production in writing and speaking, as they offer
immediate, customized feedback that traditional methods often lack (Al-Raimi et al., 2024).
The adoption of AI in EFL has also led to improved student attitudes and engagement.
According to Darwin et al. (2024) students respond positively to AI applications and appreciate
the autonomy and individualized pace these tools offer, particularly due to their flexibility,
interactivity, and capacity to simulate real-life communication. Also, AI tools with speech
recognition technologies reduce language anxiety while supporting motivation and sustained
attention in learning activities, as they help learners enhance pronunciation and fluency through
repeated, low-stress interaction, encouraging more natural language use without the fear of human
judgment (Alshumaimeri & Alshememry, 2024).
Furthermore, AI-based mobile applications and virtual environments extend learning
beyond the classroom by simulating real-world communication scenarios and offering flexible
access to resources. However, AI integration in EFL raises concerns related to data privacy,
algorithmic bias, and increased reliance on technology. According to Alghamdy (2023), the use
of facial recognition for proctoring or AI-generated assessments brings up questions about student
surveillance and the handling of sensitive information.
Moreover, dependence on AI tools may undermine the development of critical thinking
and reduce opportunities for authentic human interaction. Teachers have expressed concerns
regarding the limitations of AI tools in capturing cultural and contextual nuances of language use,
as well as their potential to dehumanize the learning process. Some AI systems, although effective
in providing grammatical feedback, lack the ability to offer deeper, creative insights or socio-
cultural appropriateness in language usage (Sumakul et al., 2022).
Given the cognitive and emotional demands associated with oral fluency, particularly in
EFL contexts, innovative pedagogical approaches are needed to support learners in overcoming
these challenges. In this regard, the incorporation of Artificial Intelligence offers promising
possibilities by leveraging adaptive feedback, real-time interaction, and simulated communicative
environments. AI technologies present a new opportunity to foster fluency development in a
responsive, individualized, and scalable means.
Processing (NLP), Machine Learning (ML), Artificial Neural Networks (ANNs), and Affective
Computing (AC), (Jiang, 2022).
AI fosters individualized learning through lIntelligent Tutoring Systems (ITSs), AI-
powered chatbots, and Automated Evaluation Systems (AESs). These systems provide learners
with immediate feedback tailored to their language proficiency and learning pace. For example,
chatbots facilitate real-time interaction and help develop oral and written communication skills,
while AESs support writing development through automated grammar correction and content
suggestions (Sumakul et al, 2022). Tools such as ChatGPT, Google Translate, and Grammarly
have been found to improve language production in writing and speaking, as they offer
immediate, customized feedback that traditional methods often lack (Al-Raimi et al., 2024).
The adoption of AI in EFL has also led to improved student attitudes and engagement.
According to Darwin et al. (2024) students respond positively to AI applications and appreciate
the autonomy and individualized pace these tools offer, particularly due to their flexibility,
interactivity, and capacity to simulate real-life communication. Also, AI tools with speech
recognition technologies reduce language anxiety while supporting motivation and sustained
attention in learning activities, as they help learners enhance pronunciation and fluency through
repeated, low-stress interaction, encouraging more natural language use without the fear of human
judgment (Alshumaimeri & Alshememry, 2024).
Furthermore, AI-based mobile applications and virtual environments extend learning
beyond the classroom by simulating real-world communication scenarios and offering flexible
access to resources. However, AI integration in EFL raises concerns related to data privacy,
algorithmic bias, and increased reliance on technology. According to Alghamdy (2023), the use
of facial recognition for proctoring or AI-generated assessments brings up questions about student
surveillance and the handling of sensitive information.
Moreover, dependence on AI tools may undermine the development of critical thinking
and reduce opportunities for authentic human interaction. Teachers have expressed concerns
regarding the limitations of AI tools in capturing cultural and contextual nuances of language use,
as well as their potential to dehumanize the learning process. Some AI systems, although effective
in providing grammatical feedback, lack the ability to offer deeper, creative insights or socio-
cultural appropriateness in language usage (Sumakul et al., 2022).
Given the cognitive and emotional demands associated with oral fluency, particularly in
EFL contexts, innovative pedagogical approaches are needed to support learners in overcoming
these challenges. In this regard, the incorporation of Artificial Intelligence offers promising
possibilities by leveraging adaptive feedback, real-time interaction, and simulated communicative
environments. AI technologies present a new opportunity to foster fluency development in a
responsive, individualized, and scalable means.
Vol. 13/ Núm. 1 2026 pág. 867
Types of AI tools used to support speaking in EFL contexts,
AI-Powered Chatbots appear as the most extensively used tools. These are conversational
agents embedded with natural language processing (NLP), speech recognition, and multimodal
feedback systems, that are capable of simulating real-life dialogue, often used in mobile apps or
integrated into learning platforms to support learners in structured and unstructured speaking
tasks.
ChatGPT
ChatGPT stands out as a generative AI model capable of maintaining coherent, extended
dialogues that mirror real-life conversation patterns. Its multimodal features text-based interaction
and optional audio integration make it a flexible tool for learners seeking structured or
spontaneous speaking practice. ChatGPT provides learners with scaffolded conversation
opportunities, including feedback on delivery, coherence, and pronunciation.
Replika
This is a conversational AI designed as an empathetic companion, offering both text and
voice interaction modes. The tool employs ASR (automatic speech recognition) to process learner
input and engage in real-time voice-based exchanges, supporting spontaneity and turn-taking. The
avatar mode also enhances engagement by adding visual cues and embodied interaction, helping
learners gain confidence in speaking and reducing hesitation.
Lora
This is a mobile-based AI speaking tutor specifically designed to improve pronunciation
and oral fluency through structured tasks. It provides adaptive feedback, fluency scores, and real-
time correction, making it a robust tool for academic EFL contexts. Its automated scoring system
allows learners to monitor progress, while targeted suggestions support iterative improvement.
EnglishBot
It is a task-specific chatbot designed to simulate academic speaking activities, particularly
those found in standardized tests like TOEFL. It supports fluency by guiding learners through
structured prompts such as summaries, descriptions, and arguments. The tool offers feedback on
lexical richness, response length, and semantic relevance, contributing to the development of
coherent and fluent oral output.
Andy
It is a beginner-friendly chatbot that uses pre-scripted dialogues and vocabulary support
to encourage basic speaking practice. It provides a safe space for early-stage learners to rehearse
language forms without fear of judgment, making it suitable for foundational fluency
development but insufficient for higher order speaking skills such as argumentation or narrative
construction.
Types of AI tools used to support speaking in EFL contexts,
AI-Powered Chatbots appear as the most extensively used tools. These are conversational
agents embedded with natural language processing (NLP), speech recognition, and multimodal
feedback systems, that are capable of simulating real-life dialogue, often used in mobile apps or
integrated into learning platforms to support learners in structured and unstructured speaking
tasks.
ChatGPT
ChatGPT stands out as a generative AI model capable of maintaining coherent, extended
dialogues that mirror real-life conversation patterns. Its multimodal features text-based interaction
and optional audio integration make it a flexible tool for learners seeking structured or
spontaneous speaking practice. ChatGPT provides learners with scaffolded conversation
opportunities, including feedback on delivery, coherence, and pronunciation.
Replika
This is a conversational AI designed as an empathetic companion, offering both text and
voice interaction modes. The tool employs ASR (automatic speech recognition) to process learner
input and engage in real-time voice-based exchanges, supporting spontaneity and turn-taking. The
avatar mode also enhances engagement by adding visual cues and embodied interaction, helping
learners gain confidence in speaking and reducing hesitation.
Lora
This is a mobile-based AI speaking tutor specifically designed to improve pronunciation
and oral fluency through structured tasks. It provides adaptive feedback, fluency scores, and real-
time correction, making it a robust tool for academic EFL contexts. Its automated scoring system
allows learners to monitor progress, while targeted suggestions support iterative improvement.
EnglishBot
It is a task-specific chatbot designed to simulate academic speaking activities, particularly
those found in standardized tests like TOEFL. It supports fluency by guiding learners through
structured prompts such as summaries, descriptions, and arguments. The tool offers feedback on
lexical richness, response length, and semantic relevance, contributing to the development of
coherent and fluent oral output.
Andy
It is a beginner-friendly chatbot that uses pre-scripted dialogues and vocabulary support
to encourage basic speaking practice. It provides a safe space for early-stage learners to rehearse
language forms without fear of judgment, making it suitable for foundational fluency
development but insufficient for higher order speaking skills such as argumentation or narrative
construction.
Vol. 13/ Núm. 1 2026 pág. 868
Google Assistant
Google Assistant has been utilized in EFL classrooms for impromptu speaking practice.
Learners engage in voice-based interactions by asking questions, seeking clarifications, or
simulating mini-dialogues. Its real-time voice responses supported pronunciation practice and
improved learners’ comfort with oral interaction.
Table 1
AI tools to promote speaking fluency
Tool Strengths Limitations
ChatGPT Context-aware, flexible, feedback-
rich dialogue in text/audio
No built-in voice interface; limited
real-time speech processing
Replika Realistic, emotionally supportive
interaction; real-time voice exchange
Superficial linguistic feedback;
better for affective than corrective
goals
Lora Pronunciation correction, fluency
scoring, structured practice
Limited open-ended conversation;
requires intermediate-level
proficiency
EnglishBot Academic task alignment, semantic
feedback, ASR integration
Less spontaneous; focused on
formal language production
Andy Beginner-friendly, vocabulary
scaffolding, fixed dialogues
Minimal feedback; not suited for
extended discourse
Google Assistant Natural voice interaction, highly
accessible
No educational scaffolding or
feedback
Note: Author’s elaboration, 2025
Contrastingly, Intelligent Personal Assistants (IPAs) and voice interfaces rely on
automatic speech recognition (ASR) and natural language processing (NLP) to facilitate oral
practice in both classroom and mobile-based learning environments.
Lyra Virtual Assistant
This is a mobile application that enables learners to rehearse structured speaking tasks
through voice interaction. The assistant’s real-time feedback allows learners to compare their
pronunciation to a model and adjust accordingly, promoting self-paced learning and greater
motivation. However, its limitations include a lack of advanced feedback on discourse-level
features, making it more suitable for beginners.
iLEAP
It is an enabled pronunciation and speaking coach designed for dual language learners
(DLLs). It integrates ASR technology, phoneme-level intelligibility assessment, and animated
visual feedback to guide learners in adjusting their articulation. Its interface features lip
animations and emoji-based responses that make pronunciation training engaging for young
learners.
Pronunciation and fluency feedback tools are designed to offer real-time, automated, and
individualized feedback on learners’ oral performance, particularly targeting pronunciation
accuracy, fluency rate, intonation, and other delivery-related metrics.
Google Assistant
Google Assistant has been utilized in EFL classrooms for impromptu speaking practice.
Learners engage in voice-based interactions by asking questions, seeking clarifications, or
simulating mini-dialogues. Its real-time voice responses supported pronunciation practice and
improved learners’ comfort with oral interaction.
Table 1
AI tools to promote speaking fluency
Tool Strengths Limitations
ChatGPT Context-aware, flexible, feedback-
rich dialogue in text/audio
No built-in voice interface; limited
real-time speech processing
Replika Realistic, emotionally supportive
interaction; real-time voice exchange
Superficial linguistic feedback;
better for affective than corrective
goals
Lora Pronunciation correction, fluency
scoring, structured practice
Limited open-ended conversation;
requires intermediate-level
proficiency
EnglishBot Academic task alignment, semantic
feedback, ASR integration
Less spontaneous; focused on
formal language production
Andy Beginner-friendly, vocabulary
scaffolding, fixed dialogues
Minimal feedback; not suited for
extended discourse
Google Assistant Natural voice interaction, highly
accessible
No educational scaffolding or
feedback
Note: Author’s elaboration, 2025
Contrastingly, Intelligent Personal Assistants (IPAs) and voice interfaces rely on
automatic speech recognition (ASR) and natural language processing (NLP) to facilitate oral
practice in both classroom and mobile-based learning environments.
Lyra Virtual Assistant
This is a mobile application that enables learners to rehearse structured speaking tasks
through voice interaction. The assistant’s real-time feedback allows learners to compare their
pronunciation to a model and adjust accordingly, promoting self-paced learning and greater
motivation. However, its limitations include a lack of advanced feedback on discourse-level
features, making it more suitable for beginners.
iLEAP
It is an enabled pronunciation and speaking coach designed for dual language learners
(DLLs). It integrates ASR technology, phoneme-level intelligibility assessment, and animated
visual feedback to guide learners in adjusting their articulation. Its interface features lip
animations and emoji-based responses that make pronunciation training engaging for young
learners.
Pronunciation and fluency feedback tools are designed to offer real-time, automated, and
individualized feedback on learners’ oral performance, particularly targeting pronunciation
accuracy, fluency rate, intonation, and other delivery-related metrics.
Vol. 13/ Núm. 1 2026 pág. 869
Speeko
This is a mobile application that functions as a public speaking coach. It is designed to
provide detailed feedback on articulation, pacing, and vocal tone through an AI-powered speech
analysis engine. The tool provides feedback on elements such as filler word use, pitch variation,
and clarity, making it especially valuable for learners working on delivery refinement and
expressive fluency.
Fluent
This is an AI writing tool designed primarily for individuals who stutter, but it has
implications for fluency development more broadly. Unlike other tools that assess speech after
delivery, Fluent assists users in preparing phonologically optimized scripts for oral delivery. The
system identifies potential “trigger words” that might disrupt fluency and offers personalized
synonym suggestions based on phonetic structure and speaker preferences. This forward-planning
approach allows users to rehearse with greater confidence, especially in high-stakes contexts such
as public speaking.
EAP Talk
It is designed for academic English learners and provides feedback aligned with IELTS
speaking band descriptors. This tool offers feedback on fluency, grammar, pronunciation, and
coherence through AI-generated scoring and example responses, helping learners become
familiar with standardized speaking criteria and benchmark their performance accordingly.
Liulishuo
Liulishuo, is another speech evaluation app that integrates ASR technology and AI
algorithms to provide automatic scoring on pronunciation, fluency, and grammatical accuracy.
The app allows for speaking practice in both scripted and spontaneous formats, offering detailed
feedback and recommendations for improvement. Learners benefit from immediate insights into
their fluency rate, articulation, and delivery style.
Adaptive learning platforms use artificial intelligence to personalize the learning
experience based on individual learner performance, promoting incremental improvement in
fluency through integrated tasks that include speaking, listening, vocabulary, and grammar
practice.
Duolingo
Duolingo is one of the most widely used adaptive language learning applications
worldwide. Its pedagogical model is grounded in the integration of machine learning algorithms,
gamification, and real-time feedback loops. The AI system behind Duolingo dynamically adjusts
the difficulty level and type of tasks presented to the learner based on their ongoing performance.
Shanbay
This is a Chinese language learning platform that includes modules for vocabulary and
pronunciation training. It uses speech recognition to evaluate learners’ pronunciation and fluency
Speeko
This is a mobile application that functions as a public speaking coach. It is designed to
provide detailed feedback on articulation, pacing, and vocal tone through an AI-powered speech
analysis engine. The tool provides feedback on elements such as filler word use, pitch variation,
and clarity, making it especially valuable for learners working on delivery refinement and
expressive fluency.
Fluent
This is an AI writing tool designed primarily for individuals who stutter, but it has
implications for fluency development more broadly. Unlike other tools that assess speech after
delivery, Fluent assists users in preparing phonologically optimized scripts for oral delivery. The
system identifies potential “trigger words” that might disrupt fluency and offers personalized
synonym suggestions based on phonetic structure and speaker preferences. This forward-planning
approach allows users to rehearse with greater confidence, especially in high-stakes contexts such
as public speaking.
EAP Talk
It is designed for academic English learners and provides feedback aligned with IELTS
speaking band descriptors. This tool offers feedback on fluency, grammar, pronunciation, and
coherence through AI-generated scoring and example responses, helping learners become
familiar with standardized speaking criteria and benchmark their performance accordingly.
Liulishuo
Liulishuo, is another speech evaluation app that integrates ASR technology and AI
algorithms to provide automatic scoring on pronunciation, fluency, and grammatical accuracy.
The app allows for speaking practice in both scripted and spontaneous formats, offering detailed
feedback and recommendations for improvement. Learners benefit from immediate insights into
their fluency rate, articulation, and delivery style.
Adaptive learning platforms use artificial intelligence to personalize the learning
experience based on individual learner performance, promoting incremental improvement in
fluency through integrated tasks that include speaking, listening, vocabulary, and grammar
practice.
Duolingo
Duolingo is one of the most widely used adaptive language learning applications
worldwide. Its pedagogical model is grounded in the integration of machine learning algorithms,
gamification, and real-time feedback loops. The AI system behind Duolingo dynamically adjusts
the difficulty level and type of tasks presented to the learner based on their ongoing performance.
Shanbay
This is a Chinese language learning platform that includes modules for vocabulary and
pronunciation training. It uses speech recognition to evaluate learners’ pronunciation and fluency
Vol. 13/ Núm. 1 2026 pág. 870
based on target sentences. The system offers color-coded feedback, pronunciation scores, and
sample audio for imitation, helping learners identify specific phonetic segments that need
improvement. Its scaffolded format allows learners to track their progress over time and focus on
repeated practice of problematic sounds.
Reported impact of AI tools on various components of speaking fluency
The qualitative findings indicate that AI tools support speaking fluency across three key
dimensions cognitive fluency, utterance fluency, and perceived fluency, aligned with
Segalowitz’s (2010) framework.
Cognitive fluency
AI tools particularly those integrated into chatbots, intelligent tutors, and adaptive
learning platforms significantly contribute to the development of cognitive fluency, as repeated
exposure, patterned practice, and error-sensitive feedback offered by these technologies support
learners in internalizing common language structures, rehearsing syntactic frames, and improving
their ability to express complex ideas efficiently.
For example, learners using Lora, Replika, and Lyra were found to develop stronger
conceptualization strategies. They began to articulate thoughts more fluidly and showed reduced
reliance on code-switching or their first language (L1) when engaging in oral communication.
This was particularly evident in structured and semi-structured tasks where learners were required
to synthesize information and formulate extended responses. Tools like iLEAP, which include
built-in tracking of pronunciation and performance, further enhanced learners’ mental processing
efficiency by identifying recurring errors and adapting instructional input accordingly.
One of the clearest demonstrations of AI’s support for cognitive fluency comes from Qiao
and Zhao’s (2023) study on Duolingo. Learners who used this platform demonstrated higher
levels of self-regulated learning, which translated into increased task awareness, goal setting, and
sustained verbal output. Through repeated, bite-sized oral exercises and vocabulary drills, learners
were able to automate lexical retrieval and grammatical structuring, which in turn improved their
performance in free-speaking tasks.
Another significant contribution to cognitive fluency was noted in Celik et al. (2025),
where learners practicing with ChatGPT reported greater ease in planning and organizing oral
responses. ChatGPT’s dialogic scaffolding supported learners in rehearsing responses multiple
times, improving their ability to formulate ideas on the spot. Importantly, this structured rehearsal
reduced cognitive load, allowing for greater focus on message delivery rather than form
construction.
Additionally, learners using EnglishBot and similar chatbot systems reported an increased
ability to regulate their participation, initiate speech, and extend conversation turns. Learners who
had greater control over pacing and task structure also displayed more spontaneous, coherent
production, indicating better on-the-fly planning and monitoring of speech.
based on target sentences. The system offers color-coded feedback, pronunciation scores, and
sample audio for imitation, helping learners identify specific phonetic segments that need
improvement. Its scaffolded format allows learners to track their progress over time and focus on
repeated practice of problematic sounds.
Reported impact of AI tools on various components of speaking fluency
The qualitative findings indicate that AI tools support speaking fluency across three key
dimensions cognitive fluency, utterance fluency, and perceived fluency, aligned with
Segalowitz’s (2010) framework.
Cognitive fluency
AI tools particularly those integrated into chatbots, intelligent tutors, and adaptive
learning platforms significantly contribute to the development of cognitive fluency, as repeated
exposure, patterned practice, and error-sensitive feedback offered by these technologies support
learners in internalizing common language structures, rehearsing syntactic frames, and improving
their ability to express complex ideas efficiently.
For example, learners using Lora, Replika, and Lyra were found to develop stronger
conceptualization strategies. They began to articulate thoughts more fluidly and showed reduced
reliance on code-switching or their first language (L1) when engaging in oral communication.
This was particularly evident in structured and semi-structured tasks where learners were required
to synthesize information and formulate extended responses. Tools like iLEAP, which include
built-in tracking of pronunciation and performance, further enhanced learners’ mental processing
efficiency by identifying recurring errors and adapting instructional input accordingly.
One of the clearest demonstrations of AI’s support for cognitive fluency comes from Qiao
and Zhao’s (2023) study on Duolingo. Learners who used this platform demonstrated higher
levels of self-regulated learning, which translated into increased task awareness, goal setting, and
sustained verbal output. Through repeated, bite-sized oral exercises and vocabulary drills, learners
were able to automate lexical retrieval and grammatical structuring, which in turn improved their
performance in free-speaking tasks.
Another significant contribution to cognitive fluency was noted in Celik et al. (2025),
where learners practicing with ChatGPT reported greater ease in planning and organizing oral
responses. ChatGPT’s dialogic scaffolding supported learners in rehearsing responses multiple
times, improving their ability to formulate ideas on the spot. Importantly, this structured rehearsal
reduced cognitive load, allowing for greater focus on message delivery rather than form
construction.
Additionally, learners using EnglishBot and similar chatbot systems reported an increased
ability to regulate their participation, initiate speech, and extend conversation turns. Learners who
had greater control over pacing and task structure also displayed more spontaneous, coherent
production, indicating better on-the-fly planning and monitoring of speech.
Vol. 13/ Núm. 1 2026 pág. 871
Utterance fluency
The use of AI tools fosters smoother speech delivery and reduced disfluencies by offering
structured repetition, real-time corrective feedback, and interactive dialogue formats that
promoted sustained verbal output. In Azizimajd’s (2023) study, learners who practiced with
Replika, demonstrated statistically significant improvements in both speech rate and pause
reduction. The repetitive and dialogic structure of this tool encouraged learners to extend their
spoken responses while minimizing hesitation markers, enabling longer and more fluent speech
runs. Similarly, Mudawy’s (2025) participants noted their improved capacity to “talk and think
more quickly,” supported by observable reductions in filler words and self-corrections.
Additional support for these findings is provided by Zou et al. (2023) and Junaidi et al.
(2020), whose studies reported that users of Lyra and Liulishuo experienced improvements in
rhythm, intonation, and grammatical control. The tools’ built-in feedback mechanisms helped
learners adjust the pacing of their speech and refine stress placement, contributing to more natural
and cohesive oral production. Lyra’s systematic voice-based corrections, in particular, enabled
learners to internalize prosodic patterns that are essential for producing fluid and comprehensible
speech.
EnglishBot, examined in Ruan et al. (2021), also demonstrated a notable impact on
utterance fluency. Learners engaged in semi-structured script-based dialogues achieved higher
scores in lexical resource and coherence during oral tasks, suggesting increased fluency and
planning efficiency. The platform’s non-threatening environment and structured format allowed
learners to focus on organizing their ideas without frequent interruptions or reformulations. Also,
learners using iLEAP showed improved articulation and reduced reliance on fillers or pauses after
repeated oral interactions. This suggests that the corrective functions of AI tools support more
seamless speech production over time.
Perceived fluency
Across multiple studies, the use of AI tools such as chatbots, pronunciation coaches, and
interactive avatars created environments that enhanced learners’ self-perception of fluency,
primarily by reducing anxiety and increasing willingness to communicate. AI tools foster low-
pressure, non-judgmental settings that allow learners to take oral risks without fear of
embarrassment or negative evaluation.
This was particularly evident in the findings of Kang (2022), where low-level ESL
learners reported feeling more confident and less anxious when interacting with AI avatars than
with human interlocutors. The visual and voice-based interface of these avatars enabled learners
to speak more freely and with greater comfort, leading to more spontaneous and sustained speech.
Similarly, Zhang et al. (2023) observed that users of Lora demonstrated increased willingness to
communicate (WTC) and foreign language enjoyment (FLE), alongside reductions in foreign
Utterance fluency
The use of AI tools fosters smoother speech delivery and reduced disfluencies by offering
structured repetition, real-time corrective feedback, and interactive dialogue formats that
promoted sustained verbal output. In Azizimajd’s (2023) study, learners who practiced with
Replika, demonstrated statistically significant improvements in both speech rate and pause
reduction. The repetitive and dialogic structure of this tool encouraged learners to extend their
spoken responses while minimizing hesitation markers, enabling longer and more fluent speech
runs. Similarly, Mudawy’s (2025) participants noted their improved capacity to “talk and think
more quickly,” supported by observable reductions in filler words and self-corrections.
Additional support for these findings is provided by Zou et al. (2023) and Junaidi et al.
(2020), whose studies reported that users of Lyra and Liulishuo experienced improvements in
rhythm, intonation, and grammatical control. The tools’ built-in feedback mechanisms helped
learners adjust the pacing of their speech and refine stress placement, contributing to more natural
and cohesive oral production. Lyra’s systematic voice-based corrections, in particular, enabled
learners to internalize prosodic patterns that are essential for producing fluid and comprehensible
speech.
EnglishBot, examined in Ruan et al. (2021), also demonstrated a notable impact on
utterance fluency. Learners engaged in semi-structured script-based dialogues achieved higher
scores in lexical resource and coherence during oral tasks, suggesting increased fluency and
planning efficiency. The platform’s non-threatening environment and structured format allowed
learners to focus on organizing their ideas without frequent interruptions or reformulations. Also,
learners using iLEAP showed improved articulation and reduced reliance on fillers or pauses after
repeated oral interactions. This suggests that the corrective functions of AI tools support more
seamless speech production over time.
Perceived fluency
Across multiple studies, the use of AI tools such as chatbots, pronunciation coaches, and
interactive avatars created environments that enhanced learners’ self-perception of fluency,
primarily by reducing anxiety and increasing willingness to communicate. AI tools foster low-
pressure, non-judgmental settings that allow learners to take oral risks without fear of
embarrassment or negative evaluation.
This was particularly evident in the findings of Kang (2022), where low-level ESL
learners reported feeling more confident and less anxious when interacting with AI avatars than
with human interlocutors. The visual and voice-based interface of these avatars enabled learners
to speak more freely and with greater comfort, leading to more spontaneous and sustained speech.
Similarly, Zhang et al. (2023) observed that users of Lora demonstrated increased willingness to
communicate (WTC) and foreign language enjoyment (FLE), alongside reductions in foreign
Vol. 13/ Núm. 1 2026 pág. 872
language anxiety (FLA), contributing to an enhanced perception of fluency from both the
speaker’s and the listener’s standpoint.
The affective dimension of fluency was further reinforced in the studies by Mudawy
(2025) and Rouabhia (2025). Learners consistently reported that repeated interactions with
chatbots helped them feel more at ease with spoken English, particularly in formal presentations.
These interactions appeared to help desensitize learners to performance pressure, creating a
psychologically safe space for experimentation and gradual oral improvement. In Rouabhia’s
findings, learners exhibited more confidence during live tasks after sustained chatbot practice,
suggesting that perceived fluency can be significantly shaped by prior, low-stakes interaction.
In addition to the affective benefits, Speeko and Fluent provided structured feedback and
motivational reinforcement, which contributed to learners’ positive self-assessments. These
platforms encouraged spontaneous speech through gamified elements and reward systems,
reinforcing the perception of communicative success and progress. As a result, even learners with
limited objective improvements in fluency metrics began to perceive themselves as more
competent speakers, which in turn increased their participation and verbal risk-taking in real-
world scenarios.
Other studies confirmed that perceived fluency is often rooted in metacognitive growth.
For example, the iLEAP tool, with its emoji-based visual cues and color-coded pronunciation
feedback, helped learners—especially children—associate correction with playful, constructive
learning. This reframing of error as an opportunity rather than a failure encouraged consistent
practice and a more positive self-image. The tool also allowed learners to monitor their
improvement over time, reinforcing the belief that they were becoming more fluent speakers, even
if their utterance metrics advanced incrementally.
METHODOLOGY
This study employed an action research design organized around an intervention with Al-
based speaking tasks, using Chat GPT to simulate short, guided conversations and to provide
immediate, text-based feedback on learner output. The implementation was conducted in a regular
A2-level English course with 15 students recruited by convenience sampling; participation was
voluntary and based on informed consent. Classes met four times per week, and the intervention
spanned two consecutive weeks within regular instructional time.
Across eight sessions, students completed concise speaking tasks designed to target
vocabulary expansion, pronunciation and rhythm, spontaneous speech, and speaking confidence.
In each session, the teacher introduced a communicative scenario, Students recorded short
speaking tasks and received Al-generated feedback which was copied to a learning log and briefly
discussed in class. After week one, the prompts, pacing, and feedback focus were adjusted in light
of observations and student comments.
language anxiety (FLA), contributing to an enhanced perception of fluency from both the
speaker’s and the listener’s standpoint.
The affective dimension of fluency was further reinforced in the studies by Mudawy
(2025) and Rouabhia (2025). Learners consistently reported that repeated interactions with
chatbots helped them feel more at ease with spoken English, particularly in formal presentations.
These interactions appeared to help desensitize learners to performance pressure, creating a
psychologically safe space for experimentation and gradual oral improvement. In Rouabhia’s
findings, learners exhibited more confidence during live tasks after sustained chatbot practice,
suggesting that perceived fluency can be significantly shaped by prior, low-stakes interaction.
In addition to the affective benefits, Speeko and Fluent provided structured feedback and
motivational reinforcement, which contributed to learners’ positive self-assessments. These
platforms encouraged spontaneous speech through gamified elements and reward systems,
reinforcing the perception of communicative success and progress. As a result, even learners with
limited objective improvements in fluency metrics began to perceive themselves as more
competent speakers, which in turn increased their participation and verbal risk-taking in real-
world scenarios.
Other studies confirmed that perceived fluency is often rooted in metacognitive growth.
For example, the iLEAP tool, with its emoji-based visual cues and color-coded pronunciation
feedback, helped learners—especially children—associate correction with playful, constructive
learning. This reframing of error as an opportunity rather than a failure encouraged consistent
practice and a more positive self-image. The tool also allowed learners to monitor their
improvement over time, reinforcing the belief that they were becoming more fluent speakers, even
if their utterance metrics advanced incrementally.
METHODOLOGY
This study employed an action research design organized around an intervention with Al-
based speaking tasks, using Chat GPT to simulate short, guided conversations and to provide
immediate, text-based feedback on learner output. The implementation was conducted in a regular
A2-level English course with 15 students recruited by convenience sampling; participation was
voluntary and based on informed consent. Classes met four times per week, and the intervention
spanned two consecutive weeks within regular instructional time.
Across eight sessions, students completed concise speaking tasks designed to target
vocabulary expansion, pronunciation and rhythm, spontaneous speech, and speaking confidence.
In each session, the teacher introduced a communicative scenario, Students recorded short
speaking tasks and received Al-generated feedback which was copied to a learning log and briefly
discussed in class. After week one, the prompts, pacing, and feedback focus were adjusted in light
of observations and student comments.
Vol. 13/ Núm. 1 2026 pág. 873
Speaking performance was evaluated with a pretest-posttest intervention design using the
Cambridge A2 Speaking Assessment descriptors, focusing on: grammar and vocabulary,
pronunciation, and interactive communication. On Day 1, students completed a pre-test consisting
of a two-minute monologue and a brief interactive segment aligned with A2 demands;
performances were audio-recorded and rated using the Cambridge descriptors.
Sessions 1–4 delivered the initial sequence of AI-mediated speaking tasks. At the end of
Week 1, the teacher conducted a structured reflection to identify necessary adjustments. Sessions
5–8 implemented the revised sequence and on Day 8, students completed a parallel post-test
following the same format as the pre-test. Quantitative analyses summarized the data with
descriptive statistics and tested pre–post changes for each rubric criterion using paired-samples
Student’s t tests (two-tailed, α = .05), reporting 95% confidence intervals and Cohen’s d.
RESULTS
This section reports the findings of the action research intervention integrating ChatGPT-
supported speaking tasks in an A2 EFL class. The pretest results show that, before the
intervention, students had only moderate performance in the three speaking dimensions.
Interactive communication obtained the highest mean (M = 3.4), which suggests that students
were relatively more capable of maintaining a basic interaction in English than of producing
accurate language. Grammar and vocabulary (M = 2.8) and Pronunciation (M = 2.7) were clearly
lower, indicating weaknesses in linguistic resources and in the clarity of oral production.
Table 2
Pretest scores
Dimension Mean Standard
deviation
Minimum Maximum
Grammar and vocabulary 2.8 1.505 1 4
Pronunciation 2.7 1.162 1 3
Interactive communication 3.4 1.404 1 4
Note: Author’s elaboration, 2025
In all three dimensions the standard deviations were relatively high, which means the
group was quite heterogeneous: some students were at very low levels, while others reached
higher scores. This pattern reflects an initial group with uneven oral English skills, with more
variation in interaction and in grammar and vocabulary than in pronunciation.
Speaking performance was evaluated with a pretest-posttest intervention design using the
Cambridge A2 Speaking Assessment descriptors, focusing on: grammar and vocabulary,
pronunciation, and interactive communication. On Day 1, students completed a pre-test consisting
of a two-minute monologue and a brief interactive segment aligned with A2 demands;
performances were audio-recorded and rated using the Cambridge descriptors.
Sessions 1–4 delivered the initial sequence of AI-mediated speaking tasks. At the end of
Week 1, the teacher conducted a structured reflection to identify necessary adjustments. Sessions
5–8 implemented the revised sequence and on Day 8, students completed a parallel post-test
following the same format as the pre-test. Quantitative analyses summarized the data with
descriptive statistics and tested pre–post changes for each rubric criterion using paired-samples
Student’s t tests (two-tailed, α = .05), reporting 95% confidence intervals and Cohen’s d.
RESULTS
This section reports the findings of the action research intervention integrating ChatGPT-
supported speaking tasks in an A2 EFL class. The pretest results show that, before the
intervention, students had only moderate performance in the three speaking dimensions.
Interactive communication obtained the highest mean (M = 3.4), which suggests that students
were relatively more capable of maintaining a basic interaction in English than of producing
accurate language. Grammar and vocabulary (M = 2.8) and Pronunciation (M = 2.7) were clearly
lower, indicating weaknesses in linguistic resources and in the clarity of oral production.
Table 2
Pretest scores
Dimension Mean Standard
deviation
Minimum Maximum
Grammar and vocabulary 2.8 1.505 1 4
Pronunciation 2.7 1.162 1 3
Interactive communication 3.4 1.404 1 4
Note: Author’s elaboration, 2025
In all three dimensions the standard deviations were relatively high, which means the
group was quite heterogeneous: some students were at very low levels, while others reached
higher scores. This pattern reflects an initial group with uneven oral English skills, with more
variation in interaction and in grammar and vocabulary than in pronunciation.
Vol. 13/ Núm. 1 2026 pág. 874
Table 3
Posttest scores
Dimension Mean Standard
deviation
Minimum Maximum
Grammar and vocabulary 4 0.755 3 5
Pronunciation 3.5 1.516 3 4
Interactive communication 4.2 0.861 2 5
Note: Author’s elaboration, 2025
The posttest results indicate that students performed best in Interactive communication
(M = 4.2) and in Grammar and vocabulary (M = 4.0), which means that after the intervention they
were generally able to use the language properly and interact orally at a solid level. In both of
these dimensions the variability was low, so most students scored close to the group average, and
the group was relatively homogeneous. Pronunciation had the lowest mean score and the highest
variability, which shows that students’ performance in pronunciation was more uneven: some
reached good levels, while others remained at more basic levels.
In Grammar and vocabulary, the mean increased from 2.8 in the pretest to 4.0 in the
posttest scores (+1.2); the SD dropped from 1.51 to 0.76, and scores shifted from a 1–4 range to
3–5, suggesting strong, fairly uniform gains where most students finished at a “good/very good”
level. In pronunciation, mean went from 2.7 to 3.5 (+0.8). The SD increased (1.16 → 1.52), with
a reported range of 3–4 in the posttest; this shows an average improvement, but gains were uneven
as some students advanced more than others.
Meanwhile, in interactive communication the mean rose from 3.4 to 4.2 (+0.8), the SD
decreased (1.40 → 0.86), and the range improved from 1–4 to 2–5; this shows a clear
improvement with greater consistency with many students reaching high scores. Overall pattern
shows that the group improved in all three speaking dimensions: means moved upward, the
minimum scores rose (no one stayed at the very bottom), and dispersion shrank in two of the three
areas, so performance not only improved but also became more consistent for most skills.
Table 4
T test results
Dimension p value IC 95%
inferior
IC 95%
superior
Cohen´s d
Grammar and vocabulary 0.0009 0.45 1.72 0.5
Pronunciation 0.0053 0.27 1.32 0.4
Interactive communication 0.0125 0.20 1.39 0.5
Note: Author’s elaboration, 2025
The t test results confirm that the differences between pretest and posttest scores are
statistically significant in the three speaking dimensions. For grammar and vocabulary, the p value
is well below the α = .05 threshold, so the posttest scores were significantly higher than the pretest
Table 3
Posttest scores
Dimension Mean Standard
deviation
Minimum Maximum
Grammar and vocabulary 4 0.755 3 5
Pronunciation 3.5 1.516 3 4
Interactive communication 4.2 0.861 2 5
Note: Author’s elaboration, 2025
The posttest results indicate that students performed best in Interactive communication
(M = 4.2) and in Grammar and vocabulary (M = 4.0), which means that after the intervention they
were generally able to use the language properly and interact orally at a solid level. In both of
these dimensions the variability was low, so most students scored close to the group average, and
the group was relatively homogeneous. Pronunciation had the lowest mean score and the highest
variability, which shows that students’ performance in pronunciation was more uneven: some
reached good levels, while others remained at more basic levels.
In Grammar and vocabulary, the mean increased from 2.8 in the pretest to 4.0 in the
posttest scores (+1.2); the SD dropped from 1.51 to 0.76, and scores shifted from a 1–4 range to
3–5, suggesting strong, fairly uniform gains where most students finished at a “good/very good”
level. In pronunciation, mean went from 2.7 to 3.5 (+0.8). The SD increased (1.16 → 1.52), with
a reported range of 3–4 in the posttest; this shows an average improvement, but gains were uneven
as some students advanced more than others.
Meanwhile, in interactive communication the mean rose from 3.4 to 4.2 (+0.8), the SD
decreased (1.40 → 0.86), and the range improved from 1–4 to 2–5; this shows a clear
improvement with greater consistency with many students reaching high scores. Overall pattern
shows that the group improved in all three speaking dimensions: means moved upward, the
minimum scores rose (no one stayed at the very bottom), and dispersion shrank in two of the three
areas, so performance not only improved but also became more consistent for most skills.
Table 4
T test results
Dimension p value IC 95%
inferior
IC 95%
superior
Cohen´s d
Grammar and vocabulary 0.0009 0.45 1.72 0.5
Pronunciation 0.0053 0.27 1.32 0.4
Interactive communication 0.0125 0.20 1.39 0.5
Note: Author’s elaboration, 2025
The t test results confirm that the differences between pretest and posttest scores are
statistically significant in the three speaking dimensions. For grammar and vocabulary, the p value
is well below the α = .05 threshold, so the posttest scores were significantly higher than the pretest
Vol. 13/ Núm. 1 2026 pág. 875
scores; the reported Cohen’s d corresponds to a moderate effect size and means the intervention
produced a meaningful change.
In pronunciation, the p value shows the improvement is also statistically significant; the
Cohen’s d suggests a small-to-moderate effect; this is coherent with what was seen in the
descriptives analysis, where pronunciation improved but with more variability among students.
For interactive communication, the p value also shows a significant difference; the Cohen’s d
points to a moderate effect size, indicating that the intervention helped students participate and
interact more effectively. Taken together, these results show that the intervention improved
students’ performance in all three dimensions, with effects that are statistically significant and of
practical relevance.
DISCUSSION
The study’s findings are consistent with previous reports that show AI-powered
conversational tools can foster improvements in oral performance by creating low-anxiety
practice conditions and delivering immediate, individualized feedback. Studies such as those by
Zou et al. (2023), Qiao and Zhao (2023), Shivakumar et al. (2021) and, more recently, Celik et
al. (2025) have documented that AI chatbots and virtual agents increase students’ willingness to
speak, promote more frequent oral practice and support self-correction through rapid feedback
loops.
The integration of AI into EFL instruction has introduced a shift in how speaking skills are
developed, with a strong emphasis on learner autonomy, adaptive feedback, and low-anxiety
environments. One of the most prominent pedagogical advantages is AI’s capacity to simulate
interactive and responsive communicative settings through chatbots, virtual assistants and
advanced speech recognition platforms. These tools provide learners with extended opportunities
to practice oral language without fear of negative evaluation, which is crucial for lowering the
affective filter and promoting willingness to communicate.
AI technologies support formative and individualized learning by delivering real-time,
data-driven feedback on pronunciation, fluency, and syntactic accuracy (Zou et al., 2023). This
feedback loop facilitates self-regulation and metacognitive awareness, enabling learners to
monitor their progress and make targeted improvements. In studies using Duolingo and iLEAP,
learners demonstrated increased engagement and improved lexical retrieval, aided by bite-sized,
personalized tasks and performance tracking (Qiao & Zhao, 2023; Shivakumar et al., 2021).
From a curriculum design standpoint, AI tools align with constructivist and socio-cultural
pedagogies, offering simulated real-life conversations that foster discourse competence and social
interaction. Tools like ChatGPT and Replika encourage learners to rehearse extended responses,
develop conceptual clarity, and engage in exploratory talk, thus supporting deeper levels of
scores; the reported Cohen’s d corresponds to a moderate effect size and means the intervention
produced a meaningful change.
In pronunciation, the p value shows the improvement is also statistically significant; the
Cohen’s d suggests a small-to-moderate effect; this is coherent with what was seen in the
descriptives analysis, where pronunciation improved but with more variability among students.
For interactive communication, the p value also shows a significant difference; the Cohen’s d
points to a moderate effect size, indicating that the intervention helped students participate and
interact more effectively. Taken together, these results show that the intervention improved
students’ performance in all three dimensions, with effects that are statistically significant and of
practical relevance.
DISCUSSION
The study’s findings are consistent with previous reports that show AI-powered
conversational tools can foster improvements in oral performance by creating low-anxiety
practice conditions and delivering immediate, individualized feedback. Studies such as those by
Zou et al. (2023), Qiao and Zhao (2023), Shivakumar et al. (2021) and, more recently, Celik et
al. (2025) have documented that AI chatbots and virtual agents increase students’ willingness to
speak, promote more frequent oral practice and support self-correction through rapid feedback
loops.
The integration of AI into EFL instruction has introduced a shift in how speaking skills are
developed, with a strong emphasis on learner autonomy, adaptive feedback, and low-anxiety
environments. One of the most prominent pedagogical advantages is AI’s capacity to simulate
interactive and responsive communicative settings through chatbots, virtual assistants and
advanced speech recognition platforms. These tools provide learners with extended opportunities
to practice oral language without fear of negative evaluation, which is crucial for lowering the
affective filter and promoting willingness to communicate.
AI technologies support formative and individualized learning by delivering real-time,
data-driven feedback on pronunciation, fluency, and syntactic accuracy (Zou et al., 2023). This
feedback loop facilitates self-regulation and metacognitive awareness, enabling learners to
monitor their progress and make targeted improvements. In studies using Duolingo and iLEAP,
learners demonstrated increased engagement and improved lexical retrieval, aided by bite-sized,
personalized tasks and performance tracking (Qiao & Zhao, 2023; Shivakumar et al., 2021).
From a curriculum design standpoint, AI tools align with constructivist and socio-cultural
pedagogies, offering simulated real-life conversations that foster discourse competence and social
interaction. Tools like ChatGPT and Replika encourage learners to rehearse extended responses,
develop conceptual clarity, and engage in exploratory talk, thus supporting deeper levels of
Vol. 13/ Núm. 1 2026 pág. 876
fluency (Celik et al., 2025). They also facilitate differentiated instruction, accommodating
learners of varying proficiency levels through adjustable difficulty settings and feedback types.
Despite the pedagogical promise, the use of AI in speaking instruction presents several
critical limitations that affect its overall efficacy and sustainability. While AI tools can accurately
detect mispronunciations or syntactic errors, they often fail to provide discourse-level corrections,
cultural insights, or pragmatic appropriateness (Mudawy, 2025; Dávila Macías et al., 2024). For
example, learners in Kang’s (2022) study noted that AI-generated dialogues felt artificial and
repetitive, limiting the richness and authenticity of the interaction. Also, the novelty effect of AI
tools tends to wane over time, particularly when they are not embedded in a coherent pedagogical
framework. Without continuous innovation in task design and teacher mediation, learners may
lose interest, resulting in superficial engagement and limited language growth.
Technological reliability is another major concern. Errors in voice recognition especially
when dealing with regional accents, background noise, or low-quality microphones can frustrate
learners and compromise the accuracy of feedback (Azizimajd, 2023; Junaidi et al., 2020). In
lower-resourced environments, inconsistent internet access, outdated devices, or limited technical
infrastructure further constrain the use of these tools (Mudawy, 2025; Avazova & Ilkhomova,
2025).
Moreover, the ethical and equity concerns are increasingly pressing. The use of AI in
language learning often involves the collection of personal speech data, raising questions about
privacy, data ownership, and algorithmic bias. This is especially problematic when AI models are
trained on culturally narrow datasets, which may marginalize non-standard accents or dialects.
Additionally, the digital divide excludes learners who lack access to reliable internet, modern
devices, or the digital literacy needed to engage effectively with these platforms.
CONCLUSION
This action research confirmed that short Al-based speaking tasks can effectively improve
students' fluency and confidence in EFL contexts. Although the intervention was small-scale, it
demonstrates the pedagogical potential of Al as a complementary tool for oral practice. In
practical terms, Chat GPT worked as a low-risk rehearsal space where learners could try out
language, receive immediate feedback and self-adjust before speaking in front of peers or the
teacher.
Pedagogically, AI tools foster learner autonomy, reduce foreign language anxiety, and
create interactive, low-risk environments that promote spontaneous oral production. By providing
immediate, individualized feedback, they help learners internalize language structures, develop
planning strategies, and improve self-regulation. Their adaptability to various proficiency levels
also makes them well-suited for differentiated instruction and curriculum integration, especially
in blended or technology-rich classrooms.
fluency (Celik et al., 2025). They also facilitate differentiated instruction, accommodating
learners of varying proficiency levels through adjustable difficulty settings and feedback types.
Despite the pedagogical promise, the use of AI in speaking instruction presents several
critical limitations that affect its overall efficacy and sustainability. While AI tools can accurately
detect mispronunciations or syntactic errors, they often fail to provide discourse-level corrections,
cultural insights, or pragmatic appropriateness (Mudawy, 2025; Dávila Macías et al., 2024). For
example, learners in Kang’s (2022) study noted that AI-generated dialogues felt artificial and
repetitive, limiting the richness and authenticity of the interaction. Also, the novelty effect of AI
tools tends to wane over time, particularly when they are not embedded in a coherent pedagogical
framework. Without continuous innovation in task design and teacher mediation, learners may
lose interest, resulting in superficial engagement and limited language growth.
Technological reliability is another major concern. Errors in voice recognition especially
when dealing with regional accents, background noise, or low-quality microphones can frustrate
learners and compromise the accuracy of feedback (Azizimajd, 2023; Junaidi et al., 2020). In
lower-resourced environments, inconsistent internet access, outdated devices, or limited technical
infrastructure further constrain the use of these tools (Mudawy, 2025; Avazova & Ilkhomova,
2025).
Moreover, the ethical and equity concerns are increasingly pressing. The use of AI in
language learning often involves the collection of personal speech data, raising questions about
privacy, data ownership, and algorithmic bias. This is especially problematic when AI models are
trained on culturally narrow datasets, which may marginalize non-standard accents or dialects.
Additionally, the digital divide excludes learners who lack access to reliable internet, modern
devices, or the digital literacy needed to engage effectively with these platforms.
CONCLUSION
This action research confirmed that short Al-based speaking tasks can effectively improve
students' fluency and confidence in EFL contexts. Although the intervention was small-scale, it
demonstrates the pedagogical potential of Al as a complementary tool for oral practice. In
practical terms, Chat GPT worked as a low-risk rehearsal space where learners could try out
language, receive immediate feedback and self-adjust before speaking in front of peers or the
teacher.
Pedagogically, AI tools foster learner autonomy, reduce foreign language anxiety, and
create interactive, low-risk environments that promote spontaneous oral production. By providing
immediate, individualized feedback, they help learners internalize language structures, develop
planning strategies, and improve self-regulation. Their adaptability to various proficiency levels
also makes them well-suited for differentiated instruction and curriculum integration, especially
in blended or technology-rich classrooms.
Vol. 13/ Núm. 1 2026 pág. 877
From a fluency perspective, AI tools support cognitive fluency by enhancing learners’
ability to process language efficiently through repeated exposure and scaffolded rehearsal;
utterance fluency by promoting smoother speech production, reducing filler words and
hesitations, and improving intonation and pacing; and perceived fluency by increasing learners’
confidence and willingness to communicate through low-anxiety interactional formats and
gamified reinforcement.
However, the study highlights critical limitations that must be addressed. AI tools currently
offer limited discourse-level and pragmatic feedback, often relying on scripted or superficial
interactions that may hinder deeper communicative competence. Technical challenges such as
voice recognition errors, low connectivity, and inconsistent device performance further restrict
the accessibility and reliability of AI-based learning. Moreover, ethical concerns regarding data
privacy, user surveillance, and algorithmic bias present ongoing risks, particularly in under-
resourced or vulnerable educational contexts.
Based on the findings of the study, several recommendations emerge to enhance the
effective integration of AI tools in promoting speaking fluency in EFL contexts. It is essential to
integrate AI technologies within a structured pedagogical framework. Rather than using these
tools in isolation, they should be incorporated into a broader curriculum that includes teacher
guidance, peer interaction, and reflective learning activities.
The selection of AI tools should be guided by learners’ proficiency levels, needs, and
preferences. Tools like Andy or iLEAP may be more suitable for beginners due to their structured
and supportive nature, while more advanced learners may benefit from interactive platforms like
ChatGPT or Replika that promote spontaneous dialogue and conceptual fluency. Also, the use of
multimodal feedback such as visual, audio, and textual cues should be prioritized, as it caters to
diverse learning styles and strengthens learners’ metacognitive awareness of their oral production.
Applications like iLEAP and Speeko exemplify this approach by offering detailed and engaging
feedback on articulation and prosody.
Finally, it is crucial to foster learners’ critical understanding of AI’s capabilities and
limitations. They should be aware that while AI tools are valuable for practice and immediate
feedback, they are not substitutes for rich, human-mediated instruction and conversation. This
awareness will help learners use AI supportively, rather than dependently, in their language
development journey.
In conclusion, while AI holds transformative potential for EFL speaking instruction, its
impact will depend on thoughtful implementation, teacher mediation, and ongoing innovation.
Educators and developers must prioritize pedagogical alignment, equity of access, and ethical
safeguards to ensure that AI-based tools truly enhance not replace human-centered,
communicative language learning
From a fluency perspective, AI tools support cognitive fluency by enhancing learners’
ability to process language efficiently through repeated exposure and scaffolded rehearsal;
utterance fluency by promoting smoother speech production, reducing filler words and
hesitations, and improving intonation and pacing; and perceived fluency by increasing learners’
confidence and willingness to communicate through low-anxiety interactional formats and
gamified reinforcement.
However, the study highlights critical limitations that must be addressed. AI tools currently
offer limited discourse-level and pragmatic feedback, often relying on scripted or superficial
interactions that may hinder deeper communicative competence. Technical challenges such as
voice recognition errors, low connectivity, and inconsistent device performance further restrict
the accessibility and reliability of AI-based learning. Moreover, ethical concerns regarding data
privacy, user surveillance, and algorithmic bias present ongoing risks, particularly in under-
resourced or vulnerable educational contexts.
Based on the findings of the study, several recommendations emerge to enhance the
effective integration of AI tools in promoting speaking fluency in EFL contexts. It is essential to
integrate AI technologies within a structured pedagogical framework. Rather than using these
tools in isolation, they should be incorporated into a broader curriculum that includes teacher
guidance, peer interaction, and reflective learning activities.
The selection of AI tools should be guided by learners’ proficiency levels, needs, and
preferences. Tools like Andy or iLEAP may be more suitable for beginners due to their structured
and supportive nature, while more advanced learners may benefit from interactive platforms like
ChatGPT or Replika that promote spontaneous dialogue and conceptual fluency. Also, the use of
multimodal feedback such as visual, audio, and textual cues should be prioritized, as it caters to
diverse learning styles and strengthens learners’ metacognitive awareness of their oral production.
Applications like iLEAP and Speeko exemplify this approach by offering detailed and engaging
feedback on articulation and prosody.
Finally, it is crucial to foster learners’ critical understanding of AI’s capabilities and
limitations. They should be aware that while AI tools are valuable for practice and immediate
feedback, they are not substitutes for rich, human-mediated instruction and conversation. This
awareness will help learners use AI supportively, rather than dependently, in their language
development journey.
In conclusion, while AI holds transformative potential for EFL speaking instruction, its
impact will depend on thoughtful implementation, teacher mediation, and ongoing innovation.
Educators and developers must prioritize pedagogical alignment, equity of access, and ethical
safeguards to ensure that AI-based tools truly enhance not replace human-centered,
communicative language learning
Vol. 13/ Núm. 1 2026 pág. 878
REFERENCES
Al-Raimi, M., Mudhsh, B. A., Muqaibal, M. H., & Alyafaei, Y. (2024). To what extent has
artificial intelligence impacted EFL teaching and learning? A systematic review. Jurnal
Arbitrer, 11(3), 399–412. https://doi.org/10.25077/ar.11.3.399-412.2024
Alaraj, M. M. (2017). EFL speaking acquisition: Identifying problems, suggesting learning
strategies and examining their effect on students' speaking fluency. The International
Journal of Social Sciences and Humanities Invention, 4(1), 3215–3221.
https://doi.org/10.18535/ijsshi/v4i1.05
Alghamdy, R. Z. (2023). Pedagogical and ethical implications of artificial intelligence in EFL
context: A review study. English Language Teaching, 16(10), 87–96.
https://doi.org/10.5539/elt.v16n10p87
Alshumaimeri, Y. A., & Alshememry, A. K. (2024). The extent of AI applications in EFL learning
and teaching. IEEE Transactions on Learning Technologies, 17, 653–661.
https://doi.org/10.1109/TLT.2023.3322128
Avazova, B. K. Q., & Ilkhomova, U. D. (2025). An investigation into the impact of artificial
intelligence on enhancing oral proficiency of ESL students. EduVision: Journal of
Innovations in Pedagogy and Educational Advancements, 1(3), 243–247.
Azizimajd, H. (2023). Investigating the impacts of voice-based student-chatbot interactions in the
classroom on EFL learners’ oral fluency and foreign language speaking anxiety.
Technology Assisted Language Education, 1(2), 61–83.
https://doi.org/10.22126/tale.2023.2732
Brown, P. S. (2017). What is fluency and how do we develop it? TESL Ontario. CONTACT
Magazine, Early View (Nov), 5–7.
Celik, B., Yildiz, Y., & Kara, S. (2025). Using ChatGPT as a virtual speaking tutor to boost EFL
learners’ speaking self-efficacy. Australian Journal of Applied Linguistics, 8(1), Article
102418. https://doi.org/10.29140/ajal.v8n1.102418
Darwin, D., Mukminatien, N., Suryati, N., Laksmi, E. D., & Marzuki. (2024). Critical thinking in
the AI era: An exploration of EFL students’ perceptions, benefits, and limitations. Cogent
Education, 11(1), 2290342.
Dávila Macías, A. M., Armijos Solano, D. O., Palma Perero, L. M., Roca Panimboza, J. A., &
Lucas Soledispa, C. J. (2023). The potential of artificial intelligence to improve speaking
skills in a second language (English) fluently. Ciencia Latina Revista Científica, 8(3),
3826–3842. https://doi.org/10.37811/cl_rcm.v8i3.11592
Ghasemi, A. A., & Mozaheb, M. A. (2021). Developing EFL learners’ speaking fluency: Use of
practical techniques. MEXTESOL Journal, 45(2), 1–14.
REFERENCES
Al-Raimi, M., Mudhsh, B. A., Muqaibal, M. H., & Alyafaei, Y. (2024). To what extent has
artificial intelligence impacted EFL teaching and learning? A systematic review. Jurnal
Arbitrer, 11(3), 399–412. https://doi.org/10.25077/ar.11.3.399-412.2024
Alaraj, M. M. (2017). EFL speaking acquisition: Identifying problems, suggesting learning
strategies and examining their effect on students' speaking fluency. The International
Journal of Social Sciences and Humanities Invention, 4(1), 3215–3221.
https://doi.org/10.18535/ijsshi/v4i1.05
Alghamdy, R. Z. (2023). Pedagogical and ethical implications of artificial intelligence in EFL
context: A review study. English Language Teaching, 16(10), 87–96.
https://doi.org/10.5539/elt.v16n10p87
Alshumaimeri, Y. A., & Alshememry, A. K. (2024). The extent of AI applications in EFL learning
and teaching. IEEE Transactions on Learning Technologies, 17, 653–661.
https://doi.org/10.1109/TLT.2023.3322128
Avazova, B. K. Q., & Ilkhomova, U. D. (2025). An investigation into the impact of artificial
intelligence on enhancing oral proficiency of ESL students. EduVision: Journal of
Innovations in Pedagogy and Educational Advancements, 1(3), 243–247.
Azizimajd, H. (2023). Investigating the impacts of voice-based student-chatbot interactions in the
classroom on EFL learners’ oral fluency and foreign language speaking anxiety.
Technology Assisted Language Education, 1(2), 61–83.
https://doi.org/10.22126/tale.2023.2732
Brown, P. S. (2017). What is fluency and how do we develop it? TESL Ontario. CONTACT
Magazine, Early View (Nov), 5–7.
Celik, B., Yildiz, Y., & Kara, S. (2025). Using ChatGPT as a virtual speaking tutor to boost EFL
learners’ speaking self-efficacy. Australian Journal of Applied Linguistics, 8(1), Article
102418. https://doi.org/10.29140/ajal.v8n1.102418
Darwin, D., Mukminatien, N., Suryati, N., Laksmi, E. D., & Marzuki. (2024). Critical thinking in
the AI era: An exploration of EFL students’ perceptions, benefits, and limitations. Cogent
Education, 11(1), 2290342.
Dávila Macías, A. M., Armijos Solano, D. O., Palma Perero, L. M., Roca Panimboza, J. A., &
Lucas Soledispa, C. J. (2023). The potential of artificial intelligence to improve speaking
skills in a second language (English) fluently. Ciencia Latina Revista Científica, 8(3),
3826–3842. https://doi.org/10.37811/cl_rcm.v8i3.11592
Ghasemi, A. A., & Mozaheb, M. A. (2021). Developing EFL learners’ speaking fluency: Use of
practical techniques. MEXTESOL Journal, 45(2), 1–14.
Vol. 13/ Núm. 1 2026 pág. 879
Ghai, B., & Mueller, K. (2021). Fluent: An AI augmented writing tool for people who stutter. In
Proceedings of the 23rd International ACM SIGACCESS Conference on Computers and
Accessibility (ASSETS '21) (pp. 1–8). ACM. https://doi.org/10.1145/3441852.3471211
Jiang, R. (2022). How does artificial intelligence empower EFL teaching and learning nowadays?
A review on artificial intelligence in the EFL context. Frontiers in Psychology, 13,
1049401. https://doi.org/10.3389/fpsyg.2022.1049401
Junaidi, Hamuddin, B., Julita, K., Rahman, F., & Derin, T. (2020). Artificial intelligence in EFL
context: Rising students’ speaking performance with Lyra virtual assistance.
International Journal of Advanced Science and Technology, 29(5), 6735–6741.
Kang, H. (2022). Effects of artificial intelligence (AI) and native speaker interlocutors on ESL
learners' speaking ability and affective aspects. Multimedia-Assisted Language Learning,
25(2), 9–43.
Kim, H.-S., Cha, Y., & Kim, N. Y. (2021). Effects of AI chatbots on EFL students’
communication skills. Korean Journal of English Language and Linguistics, 21, 712–
734. https://doi.org/10.15738/kjell.21..202108.712
Mishra, P. (2017). Speaking skill. In Pedagogy of English Teaching (Part I) (pp. 138–143).
University of Allahabad.
Mudawy, A. M. A. (2025). Exploring EFL learners’ perceptions on the use of AI-powered
conversational tools to improve speaking fluency: A case study at Majmaah University.
Forum for Linguistic Studies, 7(1). https://doi.org/10.30564/fls.v7i1.7774
Qiao, H., & Zhao, A. (2023). Artificial intelligence-based language learning: Illuminating the
impact on speaking skills and self-regulation in Chinese EFL context. Frontiers in
Psychology, 14, Article 1255594. https://doi.org/10.3389/fpsyg.2023.1255594
Rao, P. S. (2019). The importance of speaking skills in English classrooms. Alford Council of
International English & Literature Journal (ACIELJ), 2(2), 6–13.
Rouabhia, R. (2025). Developing Academic Speaking Fluency with AI-Powered Chatbots: A
Study of Second-Year Master's Students in the English Department at Ali Lounici
University. http://dx.doi.org/10.2139/ssrn.5141128
Ruan, S., Jiang, L., Xu, Q., Davis, G. M., Liu, Z., Brunskill, E., & Landay, J. A. (2021).
EnglishBot: An AI-powered conversational system for second language learning. In
Proceedings of the 26th International Conference on Intelligent User Interfaces (IUI '21)
(pp. 1–11). ACM. https://doi.org/10.1145/3397481.3450648
Shafiee Rad, H. (2024). Revolutionizing L2 speaking proficiency, willingness to communicate,
and perceptions through artificial intelligence: A case of Speeko application. Innovation
in Language Learning and Teaching. https://doi.org/10.1080/17501229.2024.2309539
Ghai, B., & Mueller, K. (2021). Fluent: An AI augmented writing tool for people who stutter. In
Proceedings of the 23rd International ACM SIGACCESS Conference on Computers and
Accessibility (ASSETS '21) (pp. 1–8). ACM. https://doi.org/10.1145/3441852.3471211
Jiang, R. (2022). How does artificial intelligence empower EFL teaching and learning nowadays?
A review on artificial intelligence in the EFL context. Frontiers in Psychology, 13,
1049401. https://doi.org/10.3389/fpsyg.2022.1049401
Junaidi, Hamuddin, B., Julita, K., Rahman, F., & Derin, T. (2020). Artificial intelligence in EFL
context: Rising students’ speaking performance with Lyra virtual assistance.
International Journal of Advanced Science and Technology, 29(5), 6735–6741.
Kang, H. (2022). Effects of artificial intelligence (AI) and native speaker interlocutors on ESL
learners' speaking ability and affective aspects. Multimedia-Assisted Language Learning,
25(2), 9–43.
Kim, H.-S., Cha, Y., & Kim, N. Y. (2021). Effects of AI chatbots on EFL students’
communication skills. Korean Journal of English Language and Linguistics, 21, 712–
734. https://doi.org/10.15738/kjell.21..202108.712
Mishra, P. (2017). Speaking skill. In Pedagogy of English Teaching (Part I) (pp. 138–143).
University of Allahabad.
Mudawy, A. M. A. (2025). Exploring EFL learners’ perceptions on the use of AI-powered
conversational tools to improve speaking fluency: A case study at Majmaah University.
Forum for Linguistic Studies, 7(1). https://doi.org/10.30564/fls.v7i1.7774
Qiao, H., & Zhao, A. (2023). Artificial intelligence-based language learning: Illuminating the
impact on speaking skills and self-regulation in Chinese EFL context. Frontiers in
Psychology, 14, Article 1255594. https://doi.org/10.3389/fpsyg.2023.1255594
Rao, P. S. (2019). The importance of speaking skills in English classrooms. Alford Council of
International English & Literature Journal (ACIELJ), 2(2), 6–13.
Rouabhia, R. (2025). Developing Academic Speaking Fluency with AI-Powered Chatbots: A
Study of Second-Year Master's Students in the English Department at Ali Lounici
University. http://dx.doi.org/10.2139/ssrn.5141128
Ruan, S., Jiang, L., Xu, Q., Davis, G. M., Liu, Z., Brunskill, E., & Landay, J. A. (2021).
EnglishBot: An AI-powered conversational system for second language learning. In
Proceedings of the 26th International Conference on Intelligent User Interfaces (IUI '21)
(pp. 1–11). ACM. https://doi.org/10.1145/3397481.3450648
Shafiee Rad, H. (2024). Revolutionizing L2 speaking proficiency, willingness to communicate,
and perceptions through artificial intelligence: A case of Speeko application. Innovation
in Language Learning and Teaching. https://doi.org/10.1080/17501229.2024.2309539
Vol. 13/ Núm. 1 2026 pág. 880
Shahini, G., & Shahamirian, F. (2017). Improving English speaking fluency: The role of six
factors. Advances in Language and Literary Studies, 8(6), 100–106.
https://doi.org/10.7575/aiac.alls.v.8n.6p.100
Shivakumar, A., Shukla, S., Vasoya, M., Kasrani, I. M., & Pei, Y. (2023). AI-enabled language
speaking coaching for dual language learners. IADIS International Journal on Internet,
17(1), 66–78.
Sumakul, D. T. Y. G., Hamied, F. A., & Sukyadi, D. (2022). Artificial intelligence in EFL
classrooms: Friend or foe? LEARN Journal: Language Education and Acquisition
Research Network, 15(1), 232–256.
Tavakoli, P., Nakatsuhara, F., & Hunter, A.-M. (2020). Aspects of fluency across assessed levels
of speaking proficiency. Modern Language Journal, 104(1), 169–191.
https://doi.org/10.1111/modl.12620
Zhang, C., Meng, Y., & Ma, X. (2024). Artificial intelligence in EFL speaking: Impact on
enjoyment, anxiety, and willingness to communicate. System, 121, 103259.
https://doi.org/10.1016/j.system.2024.103259
Zou, B., Du, Y., Tan, C., & Yao, Y. (2023). An investigation into artificial intelligence speech
evaluation programs with automatic feedback for developing EFL learners’ speaking
skills. SAGE Open, 13(3). https://doi.org/10.1177/21582440231193818
Shahini, G., & Shahamirian, F. (2017). Improving English speaking fluency: The role of six
factors. Advances in Language and Literary Studies, 8(6), 100–106.
https://doi.org/10.7575/aiac.alls.v.8n.6p.100
Shivakumar, A., Shukla, S., Vasoya, M., Kasrani, I. M., & Pei, Y. (2023). AI-enabled language
speaking coaching for dual language learners. IADIS International Journal on Internet,
17(1), 66–78.
Sumakul, D. T. Y. G., Hamied, F. A., & Sukyadi, D. (2022). Artificial intelligence in EFL
classrooms: Friend or foe? LEARN Journal: Language Education and Acquisition
Research Network, 15(1), 232–256.
Tavakoli, P., Nakatsuhara, F., & Hunter, A.-M. (2020). Aspects of fluency across assessed levels
of speaking proficiency. Modern Language Journal, 104(1), 169–191.
https://doi.org/10.1111/modl.12620
Zhang, C., Meng, Y., & Ma, X. (2024). Artificial intelligence in EFL speaking: Impact on
enjoyment, anxiety, and willingness to communicate. System, 121, 103259.
https://doi.org/10.1016/j.system.2024.103259
Zou, B., Du, Y., Tan, C., & Yao, Y. (2023). An investigation into artificial intelligence speech
evaluation programs with automatic feedback for developing EFL learners’ speaking
skills. SAGE Open, 13(3). https://doi.org/10.1177/21582440231193818