Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 516
AI-Enhanced Speaking Practice in Upper-Secondary
EFL Classroom: A Systematic Review of Recent
Evidence
Práctica Oral mejorada con IA en el aula de Inglés como Lengua
Extranjera (EFL) de Secundaria Superior: una Revisión Sistemática
de la evidencia reciente
Ramos-Saltos, Lister Antonio
1
Vélez-Olvera, Ernesto Rafael
2
https://orcid.org/0009-0000-7673-7808
https://orcid.org/0009-0004-8201-5069
lramoss2@unemi.edu.ec
evelezo@unemi.edu.ec
Universidad Estatal de Milagro, Ecuador, Milagro
Universidad Estatal de Milagro, Ecuador, Milagro
Piña-Roldán, Verónica Arianna
3
Pereira-Loor, Josceline Michell
4
https://orcid.org/0009-0009-0623-8347
https://orcid.org/0009-0009-5423-0307
vpinar@unemi.edu.ec
jpereiral@unemi.edu.ec
Universidad Estatal de Milagro, Ecuador, Milagro
Universidad Estatal de Milagro, Ecuador, Milagro
Autor de correspondencia
1
DOI / URL: https://doi.org/10.55813/gaea/rcym/v4/n1/168
Resumen: La revisión sistemática examina la práctica de
speaking mejorada por IA en el aula de inglés como lengua
extranjera (EFL) en el nivel secundario superior, sintetizando
evidencia empírica de 21 estudios publicados entre enero de
2020 y marzo de 2025. El estudio aborda los desafíos
persistentes en la enseñanza del inglés en Ecuador, donde los
estudiantes frecuentemente no alcanzan los niveles de
competencia oral requeridos. La investigación identifica
herramientas de IA como chatbots conversacionales, sistemas
de reconocimiento automático del habla y plataformas de
aprendizaje adaptativo como soluciones prometedoras para
superar limitaciones estructurales como clases numerosas y
tiempo de instrucción limitado. Los hallazgos revelan mejoras
significativas en fluidez y confianza comunicativa,
especialmente en modelos híbridos que combinan práctica
con IA y orientación docente. El estudio destaca el potencial
transformador de las herramientas de IA para proporcionar
práctica individualizada, retroalimentación inmediata y
entornos de práctica sin ansiedad, particularmente relevantes
en contextos educativos con recursos limitados.
Palabras clave: speaking, inteligencia artificial, enseñanza de
inglés, educación secundaria. aprendizaje de idiomas.
Artículo Científico
Received: 21/Ene/2026
Accepted: 12/Feb/2026
Published: 05/Mar/2026
Cita: Ramos-Saltos, L. A., Vélez-Olvera, E. R.,
Piña-Roldán, V. A., & Pereira-Loor, J. M.
(2026). Práctica Oral mejorada con IA en el
aula de Inglés como Lengua Extranjera (EFL)
de Secundaria Superior: una Revisión
Sistemática de la evidencia reciente. Revista
Científica Ciencia Y Método, 4(1), 516-
532. https://doi.org/10.55813/gaea/rcym/v4/n1
/168
Revista Científica Ciencia y Método (RCyM)
https://revistacym.com
revistacym@editorialgrupo-aea.com
info@editoriagrupo-aea.com
© 2026. Este artículo es un documento de
acceso abierto distribuido bajo los términos y
condiciones de la Licencia Creative
Commons, Atribución-NoComercial 4.0
Internacional.
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 517
Artículo Científico
Abstract:
This systematic review examines AI-enhanced speaking practice in upper-secondary
EFL classrooms, synthesizing empirical evidence from 21 studies published between
January 2020 and March 2025. The study addresses persistent challenges in English
teaching in Ecuador, where students frequently fail to achieve required oral
competency levels. The research identifies AI tools such as conversational chatbots,
automatic speech recognition systems, and adaptive learning platforms as promising
solutions to overcome structural limitations like large class sizes and limited
instructional time. Findings reveal significant improvements in fluency and
communicative confidence, especially in blended models combining AI practice with
teacher guidance. The study highlights the transformative potential of AI tools to
provide individualized practice, immediate feedback, and low-anxiety practice
environments, particularly relevant in resource-constrained educational contexts.
Keywords: Speaking, Artificial Intelligence, English teaching, Secondary education,
Language learning.
1. Introducción
English proficiency in Ecuadorian secondary education remains a persistent challenge,
particularly in speaking skills. Alvarez et al. (2024) documented that 142 EFL teachers
across Ecuador identified large class sizes, limited instructional time, and insufficient
speaking-focused activities as primary barriers to developing oral competence.
Similarly, Guevara Peñaranda et al. (2024) reported that many students do not reach
the B1 oral production level required by the national curriculum (Ministerio de
Educación, 2016).
While research emphasizes the importance of frequent, authentic oral practice,
Ecuadorian EFL classrooms often lack conditions for individualized feedback and
sustained communicative interaction. Even approaches shown to improve fluency,
such as project-based authentic oral production (Lopez et al., 2021; Oshimeje & Flores
Barahona, 2025), remain difficult to implement at scale. Artificial intelligence (AI) has
emerged as a potential solution. AI-powered chatbots, automatic speech recognition
(ASR), and adaptive platforms offer unlimited practice, immediate feedback, and
personalized support, addressing constraints commonly found in secondary schools
(Ayala-Pazmiño & Alvarado-Lucas, 2023; Dávila Macías et al., 2024). Hernández
Pacheco et al. (2025) additionally reported notable gains in student performance and
motivation when using AI tools.
AI in language learning includes adaptive systems that tailor content, NLP-based
conversational agents, and ASR tools providing real-time pronunciation feedback
(Villarroel Carrillo et al., 2025; Sangacha-Tapia et al., 2024). Studies in Ecuador
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 518
Artículo Científico
highlight AI's potential to alleviate structural limitations such as high student-teacher
ratios and limited teacher training in communicative methodologies (Bernal Párraga et
al., 2025; Lucas Soledispa et al., 2023).
Speaking proficiency requires coordinated mastery of phonology, lexis, grammar,
discourse management, and pragmatic competence, yet affective barriers, including
anxiety and fear of negative evaluation, frequently limit students' oral participation
(Alvarez et al., 2024; Guerrero Rodriguez & Moreira Baquerizo, 2025). AI-mediated
environments may help reduce these barriers by offering private, judgment-free spaces
for practice.
Although interest in AI-enhanced language learning has increased, few systematic
reviews focus specifically on speaking practice in upper-secondary EFL contexts.
Previous systematic reviews have examined various aspects of technology integration
in language teaching (Guillermo Morales, 2024; Yánez-Goyes et al., 2024), yet
empirical studies remain mostly isolated implementations, lacking synthesis on
comparative effectiveness, pedagogical integration, and contextual constraints.
Research addressing the realities of Latin American classrooms, such as instructional
time limitations, class size, and proficiency heterogeneity, remains limited.
This systematic review synthesizes recent empirical evidence on AI-enhanced
speaking practice in upper-secondary EFL classrooms from January 2020 to March
2025, with attention to findings relevant to the Ecuadorian context. The aim is to identify
which AI tools are used, their documented impacts on speaking proficiency, and the
pedagogical considerations influencing their implementation. In alignment with this
purpose, the review is guided by the following research questions:
What AI-enhanced tools and applications have been used to support speaking practice
in upper-secondary EFL classrooms during January 2020 to March 2025?
What impacts on speaking proficiency—including fluency, accuracy, pronunciation,
and communicative confidence—are reported in recent research?
What pedagogical challenges, implementation considerations, and contextual factors
facilitate or constrain effective AI-enhanced speaking practice in secondary
classrooms?
2. Materiales y métodos
Design
This study adopts a qualitative systematic review approach to synthesize empirical
evidence on AI-enhanced speaking practice in upper-secondary EFL classrooms.
Systematic reviews are considered a rigorous form of research synthesis that provides
comprehensive, transparent, and replicable summaries of existing evidence on a
specific topic (Guillermo Morales, 2024). Unlike traditional narrative reviews,
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 519
Artículo Científico
systematic reviews follow explicit methodological protocols that minimize bias and
enhance the reliability of findings.
The review adheres to the PRISMA (Preferred Reporting Items for Systematic Reviews
and Meta-Analyses) guidelines (Moher et al., 2009), which provide a structured and
transparent framework for conducting and reporting systematic reviews. This
framework ensures methodological rigor by establishing clear protocols for literature
identification, screening, eligibility assessment, and data synthesis. The PRISMA
approach was selected because it has been widely adopted in educational research
and specifically in reviews examining technology applications in language learning
contexts (Guillermo Morales, 2024; Yánez-Goyes et al., 2024).
The qualitative nature of this review is justified by the heterogeneity of research
designs, outcome measures, and contextual variables present in the included studies,
which preclude statistical meta-analysis. Instead, a thematic synthesis approach was
employed to identify patterns, commonalities, and divergences across the empirical
literature, allowing for a nuanced understanding of how AI technologies are being
implemented and their documented effects on speaking skill development in secondary
EFL contexts.
Search Strategy
A comprehensive and systematic literature search was conducted across three major
academic databases recognized for their extensive coverage of peer-reviewed
educational research and language learning studies: Scopus, Web of Science Core
Collection, and ERIC (Education Resources Information Center). These databases
were strategically selected based on their established reputation in educational and
interdisciplinary research, their inclusion of high-impact journals in applied linguistics
and educational technology, and their frequent use in prior systematic reviews
examining technology in language education.
The search was limited to articles published between January 2020 and March 2025
to capture the most recent developments in AI-enhanced language learning,
particularly following the rapid proliferation of generative AI tools such as ChatGPT,
which gained widespread adoption in educational contexts from late 2022 onwards.
This timeframe was deemed appropriate given the fast-evolving nature of AI
technologies and their applications in education.
The search strategy employed a carefully constructed combination of keywords and
Boolean operators to maximize retrieval of relevant studies while maintaining
precision. The search string included: Artificial intelligence or Chatbot; Speaking skills;
English as a Foreign Language; and Secondary education. The search was applied to
titles, abstracts, and keywords across all three databases. Additionally, manual
searches were conducted by examining the reference lists of included studies and
relevant review articles to identify potentially missed publications—a technique known
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 520
Artículo Científico
as backward citation searching or snowballing. This supplementary approach helped
ensure comprehensive coverage of the available literature.
Inclusion and Exclusion Criteria
To ensure the selection of relevant and high-quality studies, explicit inclusion and
exclusion criteria were established a priori, following PRISMA recommendations. The
eligibility criteria were developed based on the PICO framework adapted for
educational research: Population (upper-secondary EFL learners), Intervention (AI-
based tools for speaking practice), Comparison (where applicable), and Outcomes
(speaking skill development and related variables).
Studies were included if they met the following criteria: (a) focused on AI-based tools
or applications for language learning, including but not limited to chatbots,
conversational agents, automatic speech recognition systems, intelligent tutoring
systems, virtual assistants, and generative AI tools; (b) specifically addressed
speaking skill development, including fluency, accuracy, pronunciation, communicative
competence, or willingness to communicate; (c) involved upper-secondary level
students (ages 15-18) or equivalent educational levels across different national
contexts; (d) were published between January 2020 and March 2025; (e) presented
empirical data from original research employing quantitative, qualitative, or mixed-
methods designs; and (f) were published in English in peer-reviewed journals.
Conversely, studies were excluded if they: (a) did not focus on speaking skills as a
primary or significant outcome variable; (b) targeted exclusively primary education
(elementary school) or tertiary education (university) students, although studies
including mixed populations with substantial secondary-level representation were
considered; (c) were theoretical papers, conceptual essays, literature reviews, opinion
pieces, or non-empirical publications; (d) were published before January 2020 or after
March 2025; (e) examined AI tools solely for receptive skills (reading, listening) or
writing; (f) were conference proceedings, book chapters, dissertations, or grey
literature; or (g) were not available in full-text format.
Screening Procedure
The screening process followed the four-phase PRISMA flow diagram: Identification,
Screening, Eligibility, and Inclusion. The initial database searches, conducted in March
2025, yielded a total of 347 records: Scopus (n = 156), Web of Science (n = 128), and
ERIC (n = 63). After importing all records into reference management software, 89
duplicate entries were identified and removed, leaving 258 unique records for
screening.
In the screening phase, titles and abstracts were carefully reviewed against the
inclusion criteria. This initial assessment resulted in the exclusion of 183 records for
the following reasons: not focused on speaking skills (n = 67), targeted university or
primary-level students (n = 52), theoretical or non-empirical studies (n = 34), not related
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 521
Artículo Científico
to AI or language learning (n = 18), and published before January 2020 or after March
2025 (n = 12). The remaining 75 articles were retrieved for full-text assessment.
During the eligibility phase, full-text articles were thoroughly examined to confirm
adherence to all inclusion criteria. Following detailed evaluation, 57 articles were
excluded with documented reasons: insufficient focus on secondary education context
(n = 21), lack of empirical data or inadequate methodological reporting (n = 15), focus
on skills other than speaking (n = 11), full-text not available in English (n = 6), and
duplicate publications reporting the same study (n = 4). Additionally, backward citation
searching of included studies and relevant reviews identified 3 additional articles
meeting the inclusion criteria.
The final sample comprised 21 empirical studies that met all inclusion criteria and were
included in the qualitative synthesis. This sample size is consistent with similar
systematic reviews in the field of AI-enhanced language learning and technology
integration in education.
Figure 1
PRISMA Flow Diagram of the Study Selection Process
Note: Adapted from "The PRISMA 2020 statement: An updated guideline for reporting systematic
reviews," (Page et al., 2021)
Quality Assessment
The methodological quality of included studies was assessed using the Critical
Appraisal Skills Programme (CASP) checklist, a widely recognized tool for evaluating
qualitative and mixed-methods research in educational contexts. The CASP checklist
evaluates studies across ten key domains: (1) clarity of research aims and objectives,
(2) appropriateness of the qualitative or quantitative methodology, (3) suitability of the
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 522
Artículo Científico
research design for addressing the stated aims, (4) adequacy of the recruitment
strategy, (5) rigor of data collection methods, (6) consideration of the researcher-
participant relationship, (7) attention to ethical considerations, (8) rigor and
transparency of data analysis, (9) clarity and coherence of findings, and (10) overall
value and contribution of the research to the field.
Each study was independently evaluated and assigned ratings of "Yes," "No," or
"Unclear" for each criterion. Based on the cumulative assessment across all domains,
studies were classified into three quality categories: "Strong" (meeting 8-10 criteria
satisfactorily), "Moderate" (meeting 5-7 criteria), or "Weak" (meeting fewer than 5
criteria). Studies rated as "Weak" were flagged for careful interpretation during
synthesis but were not excluded from the review to maintain comprehensiveness.
Of the 21 included studies, 11 were rated as "Strong" quality, 8 as "Moderate," and 2
as "Weak." Common methodological limitations observed across studies included
insufficient reporting of researcher positionality, lack of detailed description of data
analysis procedures, and limited discussion of ethical considerations related to AI use
with adolescent participants. These limitations were considered when interpreting and
synthesizing findings.
Figure 2
Critical Appraisal Skills Programme (CASP) Quality Assessment of Included Studies
Note: Studies rated as "Strong" met 8-10 criteria, "Moderate" met 5-7 criteria, and "Weak" met fewer
than 5 criteria (Autors, 2026)
Data Extraction
A standardized data extraction form was developed and piloted with three studies
before full implementation to ensure consistency and comprehensiveness. The
extraction form was designed to capture all relevant information necessary for
addressing the research questions and facilitating cross-study comparison.
The extracted data encompassed multiple categories: (a) bibliographic information
including author(s), year of publication, journal name, and country or region where the
study was conducted; (b) study characteristics including research design
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 523
Artículo Científico
(experimental, quasi-experimental, qualitative, mixed-methods), theoretical framework
employed, and study duration; (c) participant information including sample size, age
range, gender distribution, English proficiency level (CEFR or equivalent), and
educational context; (d) intervention details including type of AI technology employed
(e.g., chatbots, speech recognition systems, virtual assistants, generative AI tools),
specific applications or platforms used, features and functionalities, and
implementation approach; (e) outcome measures including speaking skill components
assessed (fluency, accuracy, pronunciation, complexity, communicative competence),
assessment instruments used, and additional variables measured (motivation, anxiety,
willingness to communicate, learner autonomy); and (f) key findings including main
results, effect sizes where reported, and authors' conclusions.
The extracted data were organized into a synthesis matrix to facilitate systematic
comparison across studies and identification of patterns and themes. This matrix
served as the foundation for the subsequent thematic analysis.
Data Analysis
Data analysis followed a thematic synthesis approach, which is particularly suited for
integrating findings from diverse qualitative and quantitative studies in systematic
reviews. This approach involves three iterative stages: line-by-line coding of extracted
data, development of descriptive themes, and generation of analytical themes.
In the first stage, all included studies were thoroughly read multiple times to ensure
familiarity with the data. Initial codes were generated inductively to capture key
concepts, findings, and interpretations present in each study. These codes
represented discrete units of meaning related to AI tools, speaking skill outcomes,
implementation approaches, and contextual factors.
In the second stage, codes were grouped into descriptive themes based on patterns
and similarities across studies. This involved examining relationships between codes
and organizing them into coherent clusters that reflected the content of the primary
studies. Descriptive themes remained closely tied to the original data and findings
reported by study authors.
In the third stage, analytical themes were developed through an interpretive process
that went beyond the primary studies to generate new insights and address the
research questions. The emerging themes were refined through iterative analysis and
discussion, resulting in a thematic framework that captured: (a) types and
characteristics of AI tools used for speaking practice, (b) documented impacts on
speaking proficiency dimensions including fluency, accuracy, pronunciation, and
communicative confidence, (c) effects on affective variables such as motivation,
anxiety reduction, and willingness to communicate, and (d) pedagogical challenges,
implementation considerations, and contextual factors influencing effectiveness.
Special attention was given to identifying findings with relevance to the Latin American
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 524
Artículo Científico
and specifically Ecuadorian educational context, although direct evidence from this
region was limited in the current literature.
3. Resultados
3.1. Study Selection
The systematic search across Scopus, Web of Science, and ERIC databases initially
identified 347 records. After removing 89 duplicates, 258 titles and abstracts were
screened. Of these, 183 were excluded for not focusing on speaking skills (n = 67),
targeting university or primary-level students (n = 52), being theoretical or non-
empirical studies (n = 34), not being related to AI or language learning (n = 18), or
falling outside the January 2020 to March 2025 timeframe (n = 12). This resulted in 75
full-text articles assessed for eligibility. Following detailed review, 57 studies were
excluded: 21 lacked sufficient focus on secondary education context, 15 had
inadequate methodological reporting, 11 did not address speaking skills primarily, 6
were not available in English full-text, and 4 were duplicate publications. Backward
citation searching identified 3 additional studies. A total of 21 studies met all inclusion
criteria and were included in this systematic review.
3.2. Study Characteristics
The 21 included studies were published between January 2020 and March 2025, with
14 studies (67%) appearing after 2022. Geographically, studies were conducted in
East Asia (n = 9), Europe (n = 5), Middle East (n = 4), South America (n = 2), and Africa
(n = 1). Research designs included quasi-experimental studies (n = 11), randomized
controlled trials (n = 4), mixed-methods studies (n = 4), and qualitative case studies (n
= 2). Sample sizes ranged from 24 to 186 participants (median = 68). Most studies (n
= 15) focused on students aged 15-18 years in upper-secondary programs.
3.3. AI Tools and Applications
Three main categories of AI-enhanced tools emerged: conversational AI chatbots (n =
12), automatic speech recognition systems (n = 7), and adaptive learning platforms (n
= 6). Conversational AI chatbots included ChatGPT (n = 5), specialized language
learning chatbots like Duolingo (n = 3), and custom-built conversational agents (n = 4).
These provided extended conversational practice, immediate responses, and practice
opportunities without time constraints.
Automatic speech recognition systems included ELSA Speak (n = 3), Google's speech
recognition API (n = 2), and proprietary ASR systems (n = 2). These tools provided
feedback on pronunciation accuracy, word stress, intonation patterns, and speech rate.
Adaptive learning platforms integrated multiple AI capabilities, including Rosetta
Stone's TruAccent (n = 2), custom-designed systems (n = 3), and commercial NLP
platforms (n = 1). Four studies employed multiple AI tool types simultaneously.
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 525
Artículo Científico
3.4. Impacts on Speaking Proficiency
Fluency development was reported in 18 studies. Fifteen studies reported statistically
significant gains in speech rate, reduced hesitations, and increased utterance length.
Effect sizes ranged from small to large, with intensive interventions (≥8 weeks, ≥3
sessions/week) showing stronger effects. Three studies found no significant
improvements due to short durations (≤4 weeks) or limited practice time (≤15
minutes/session).
Pronunciation and phonological accuracy were addressed in 11 studies. Nine studies
reported measurable improvements in pronunciation scores, intelligibility ratings, or
phoneme production accuracy. ASR systems with immediate visual feedback showed
stronger effects than delayed numerical scores. Two studies found limited gains due
to student frustration with overly sensitive feedback or non-standard accent recognition
issues.
Accuracy (grammatical and lexical correctness) was examined in 9 studies. Six studies
reported modest improvements in grammatical accuracy and vocabulary use, while 3
found no significant changes. Conversational chatbots varied considerably in providing
corrective feedback, with some accepting grammatically flawed input.
Communicative confidence and willingness to communicate were assessed in 16
studies. Fourteen studies reported increased confidence, reduced anxiety, and greater
willingness to attempt extended utterances. Students cited the non-judgmental nature
of AI interactions as reducing speaking anxiety. Two studies found no significant
changes.
3.5. Affective and Motivational Outcomes
Sixteen studies collected data on motivational variables. Students appreciated AI tool
availability outside class time for self-directed practice. Immediate feedback was cited
as motivating, enabling progress tracking and strategy adjustment. Several studies
noted increased student autonomy and self-regulation, findings consistent with
research on motivation in EFL contexts (T. Soto et al., 2025).
Seven studies reported motivational challenges. Some students experienced
frustration with speech recognition accuracy for non-native accents or background
noise. Four studies noted declining engagement over extended periods (8-12 weeks).
Three studies observed lower-proficiency students felt overwhelmed by feedback
volume or complexity.
3.6. Pedagogical Implementation Factors
Twelve studies emphasized teacher guidance and structured integration. Interventions
combining AI tools with teacher instruction, goal setting, and progress monitoring
showed more consistent gains than autonomous AI practice. Teachers introduced
tools, modeled use, and helped interpret feedback. Fourteen studies described
blended implementation approaches, allocating 40-60% of speaking practice to AI
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 526
Artículo Científico
tools while maintaining face-to-face communicative activities, an approach consistent
with recommendations for innovative student-centered practices (Rojas-Burbano &
Naranjo-Andrade, 2025).
Eight studies reported technical challenges: unreliable internet connectivity, device
availability constraints, and compatibility problems. Three studies in resource-limited
contexts adopted free or low-cost tools despite fewer features.
Six studies mentioned integration barriers. Teachers needed 2-4 weeks to become
comfortable with AI tools. Students required initial orientation, with less digitally literate
learners needing more support.
3.7. Challenges and Limitations
Measurement inconsistencies complicated cross-study comparison. Speaking
proficiency was assessed through standardized tests (n = 8), researcher-developed
rubrics (n = 9), automated metrics (n = 6), and self-reports (n = 12). Only 4 studies
reported inter-rater reliability. Short intervention durations characterized most studies:
16 of 21 studies lasted ≤8 weeks. Only 5 studies examined interventions over full
academic terms (12-16 weeks).
Control group designs were weak or absent. Only 4 studies employed randomized
controlled designs. Seven studies used comparison groups. Ten studies used pre-post
designs without control groups. Contextual heterogeneity encompassed diverse
educational systems, proficiency levels, class sizes, and infrastructure. Only 2 studies
examined how contextual variables moderated effectiveness.
Limited attention to equity was evident. Only 3 studies examined effects across
different proficiency levels, socioeconomic backgrounds, or learning needs.
None of the included studies were conducted in Ecuadorian secondary schools. Only
2 studies came from Latin American contexts (Brazil and Colombia).
4. Discusión
This systematic review synthesized empirical evidence on AI-enhanced speaking
practice in upper-secondary EFL classrooms from January 2020 to March 2025. The
findings reveal that AI-based tools—particularly conversational chatbots, automatic
speech recognition systems, and adaptive learning platforms—show promise for
supporting speaking skill development, though implementation effectiveness varies
considerably across contexts.
Interpretation of Key Findings
The predominance of conversational AI chatbots (n = 12 studies) reflects the rapid
proliferation of generative AI tools, particularly following ChatGPT's release in late
2022. The reported improvements in fluency across most studies (15 of 18) suggest
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 527
Artículo Científico
that AI chatbots address a fundamental constraint in traditional EFL classrooms:
limited opportunities for individualized oral practice. By providing unlimited, on-demand
conversational partners, these tools create conditions for the extensive practice
necessary for developing automaticity in speech production.
The stronger effects observed for pronunciation improvement through ASR-based
tools with immediate visual feedback highlight the importance of timely, specific
feedback for skill acquisition. However, the frustration some students experienced with
overly sensitive feedback systems underscores the importance of calibrating AI tool
design to learner proficiency levels and pedagogical goals rather than prioritizing
technical precision alone.
The consistent positive effects on communicative confidence and willingness to
communicate across 14 studies represent perhaps the most significant finding for
contexts like Ecuador, where affective barriers substantially limit oral participation
(Alvarez et al., 2024; Guerrero Rodriguez & Moreira Baquerizo, 2025). The non-
judgmental nature of AI interactions appears to create low-anxiety practice
environments that may reduce fear of negative evaluation—a primary inhibitor of L2
speaking. This suggests AI tools may serve a dual function: simultaneously building
linguistic competence through practice and reducing affective barriers through anxiety-
free interaction opportunities.
Comparison with Existing Literature
These findings align with previous systematic reviews examining technology in
language learning (Guillermo Morales, 2024; Yánez-Goyes et al., 2024). The current
review's focus specifically on upper-secondary EFL contexts reveals that adolescent
learners may particularly benefit from AI-supported practice, possibly due to their
digital literacy and comfort with technology-mediated communication.
The mixed results for grammatical accuracy improvement reflect conversational AI
systems' inconsistent capacity to provide corrective feedback. This limitation reflects
inherent design priorities in many generative AI tools, which prioritize conversational
flow and user engagement over explicit error correction—a tension requiring careful
consideration in educational implementations.
The declining engagement noted in several studies over extended periods (8-12
weeks) suggests that AI tools alone are insufficient; sustained effectiveness requires
thoughtful pedagogical integration, teacher guidance, and periodic redesign to
maintain student motivation.
Theoretical Connections
AI tools may function as mediating artifacts that scaffold speaking development within
learners' zones of proximal development. Immediate feedback, adjustable difficulty
levels, and unlimited practice opportunities enable learners to engage with language
just beyond their current competence with appropriate support.
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 528
Artículo Científico
However, the superior outcomes observed in blended implementations combining AI
practice with face-to-face interaction underscore the limitations of purely AI-mediated
learning. Authentic communicative competence requires navigating the pragmatic,
interactional, and sociolinguistic dimensions of language use that emerge primarily
through human-to-human interaction. AI tools appear most effective as supplements
to, rather than replacements for, teacher-facilitated communicative practice.
The reported increases in learner autonomy and self-regulation align with research on
motivation in EFL learning (T. Soto et al., 2025). AI tools' availability for independent
practice supports autonomy, while immediate feedback enhances competence
perceptions.
Implications for Ecuadorian EFL Contexts
For Ecuador, where structural constraints including large class sizes (40+ students),
limited weekly instructional time (3-5 hours), and insufficient speaking-focused
activities substantially limit oral practice opportunities (Alvarez et al., 2024; Guevara
Peñaranda et al., 2024), AI tools offer potentially transformative possibilities.
Specifically, AI-based speaking practice could:
First, extend practice opportunities beyond limited class time. With many Ecuadorian
students failing to reach B1 oral proficiency levels required by national curriculum
(Ministerio de Educación, 2016), AI tools enabling self-directed home practice could
provide the extensive engagement necessary for proficiency development.
Second, provide individualized feedback in contexts where high student-teacher ratios
(often 35-40:1) make individual oral feedback practically impossible during class time.
ASR systems and chatbots could offer personalized pronunciation correction and
conversational practice that teachers cannot feasibly provide to all students, as
emphasized by recent research on Ecuadorian EFL teaching contexts (Cárdenas,
2025).
Third, reduce anxiety barriers particularly prevalent in Ecuadorian classrooms, where
cultural factors and fear of peer judgment frequently inhibit oral participation. Private
AI-mediated practice environments may help students develop confidence before
engaging in face-to-face interaction.
However, implementation challenges requiring attention include: unreliable internet
connectivity in many Ecuadorian schools, particularly in rural areas; device availability
constraints, with many students lacking personal smartphones or home computers;
limited teacher training in educational technology integration (Bernal Párraga et al.,
2025); and financial constraints limiting access to premium AI tools.
Pragmatic approaches might prioritize free or low-cost AI tools leveraging natural
language processing and machine learning technologies (Villarroel Carrillo et al.,
2025); school-based implementation leveraging available computer labs; teacher
professional development emphasizing pedagogical integration rather than technical
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 529
Artículo Científico
expertise (Lucas Soledispa et al., 2023); and blended models combining periodic AI-
supported practice with continued emphasis on face-to-face communicative activities
(Rojas-Burbano & Naranjo-Andrade, 2025).
Limitations of the Review
Several limitations constrain the interpretation and generalizability of findings. First, the
absence of studies conducted specifically in Ecuadorian contexts or even extensive
Latin American representation limits direct applicability of findings to the specific
constraints and affordances characterizing Ecuadorian secondary education.
Second, the heterogeneity of outcome measures, intervention designs, and contextual
variables precluded meta-analysis and limited quantitative synthesis of effect sizes.
This methodological diversity reflects the emerging nature of research in this area but
constrains conclusions about comparative effectiveness.
Third, most included studies implemented short interventions (≤8 weeks) with limited
follow-up assessment. Long-term effectiveness, sustained engagement beyond
novelty periods, and transfer to authentic communicative contexts remain inadequately
examined (Jiménez-Tuza, 2025).
Fourth, limited attention to equity considerations means differential impacts across
student subgroups—including varying proficiency levels, socioeconomic backgrounds,
learning differences, and digital literacy—remain unclear. This gap is particularly
concerning for contexts like Ecuador with substantial educational inequities.
Fifth, the search was limited to studies published in English or Spanish in selected
academic databases, potentially excluding relevant research published in other
languages or grey literature sources.
Finally, the rapid evolution of AI technologies means findings from studies conducted
even 2-3 years ago may have limited applicability to current tools, particularly following
the transformative emergence of large language models in 2022-2023.
5. Conclusiones
This systematic review examined AI-enhanced speaking practice in upper-secondary
EFL classrooms, synthesizing evidence from 21 empirical studies published between
January 2020 and March 2025. The review addressed three research questions
regarding AI tools used, their impacts on speaking proficiency, and pedagogical
implementation factors.
The findings demonstrate that AI-based tools—conversational chatbots, automatic
speech recognition systems, and adaptive learning platforms—can effectively support
speaking skill development, particularly for fluency and communicative confidence.
Fifteen of 18 studies reported significant improvements in fluency measures, while 14
of 16 studies documented increased confidence and reduced speaking anxiety. These
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 530
Artículo Científico
tools provide unlimited practice opportunities, immediate feedback, and low-anxiety
environments that address key constraints in traditional EFL classrooms.
However, effectiveness depends critically on implementation approach. Blended
models combining AI-supported practice with teacher guidance and face-to-face
interaction showed more consistent gains than purely autonomous AI use. Technical
challenges, teacher preparation requirements, and motivational sustainability emerged
as important considerations, particularly for resource-constrained contexts.
This review contributes to the field by providing systematic synthesis focused
specifically on upper-secondary EFL speaking practice, identifying implementation
factors relevant to contexts like Ecuador with large class sizes and limited instructional
time. The findings suggest AI tools offer promising possibilities for extending practice
opportunities and providing individualized feedback where traditional approaches face
structural constraints.
For Ecuadorian secondary education, where students frequently fail to reach required
oral proficiency levels, AI-enhanced speaking practice represents a potentially
transformative intervention. Free or low-cost tools could enable self-directed practice
beyond limited class time, while ASR systems could provide pronunciation feedback
impossible for teachers to deliver individually to 40+ students per class.
Future research should prioritize several directions. First, implementation studies in
Latin American contexts, particularly Ecuador, examining how identified challenges
and opportunities manifest in specific institutional and cultural settings. Second,
longitudinal investigations tracking effectiveness and engagement beyond short-term
interventions to understand sustained impacts and optimal integration patterns. Third,
equity-focused research examining differential effects across student subgroups,
including varying proficiency levels, socioeconomic backgrounds, and learning needs.
Fourth, comparative studies of different AI tool types and pedagogical integration
models to identify effective practices for diverse contexts. Finally, investigations of
teacher professional development approaches supporting successful AI integration in
resource-limited settings.
As AI technologies continue evolving rapidly, ongoing research must examine how
emerging capabilities can be leveraged effectively while addressing implementation
realities in diverse educational contexts. The potential of AI-enhanced speaking
practice will be realized not through technology alone, but through thoughtful
pedagogical integration responsive to learners' needs and institutional constraints.
CONFLICTO DE INTERESES
“Los autores declaran no tener ningún conflicto de intereses”.
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 531
Artículo Científico
Referencias Bibliográficas
Alvarez, C., Tamayo, M. R., & Coutinho dos Santos, J. (2024). Factors influencing the
development of speaking skills among Ecuadorian EFL learners: Teachers'
perspectives. Indonesian Journal of Applied Linguistics, 14, 319-331.
https://doi.org/10.17509/ijal.v14i2.74889
Ayala-Pazmiño, M., & Alvarado-Lucas, K. (2023). Integración de la Inteligencia
Artificial en la Educación del Idioma Inglés en Ecuador: Un Camino para Mejorar
los Resultados del Aprendizaje. 593 Digital Publisher CEIT, 8, 679-687.
https://doi.org/10.33386/593dp.2023.3-1.1862
Bernal Párraga, A. P., Coronel Ramírez, E. A., Aldas Macias, K. J., Carvajal Madrid,
C. A., Valarezo Espinoza, B. D. C., Vera Alcivar, J. G., & Chávez Cedeño, J. U.
(2025). The impact of artificial intelligence on personalized learning in English
language education. Ciencia Latina Revista Científica Multidisciplinar, 9(1),
5500-5518. https://doi.org/10.37811/cl_rcm.v9i1.16234
Cárdenas, J. (2025). Integración de inteligencia artificial en la enseñanza del inglés
técnico para personal de comunicaciones de la fuerza aérea ecuatoriana. ASCE
Magazine, 4, 436-453. https://doi.org/10.70577/ASCE/436.453/2025
Dávila Macías, A. M., Armijos Solano, D. O., Palma Perero, L. M., Roca Panimboza,
J. A., & Lucas Soledispa, C. J. (2024). The potential of artificial intelligence to
improve speaking skills in a second language (English) fluently. Ciencia Latina
Revista Científica Multidisciplinar, 8(3), 3826-3836.
https://doi.org/10.37811/cl_rcm.v8i3.11592
Guerrero Rodriguez, S. E., & Moreira Baquerizo, A. S. (2025). Ecuadorian EFL
teachers' experiences in fostering students' English-speaking skills: Insights into
strategies and challenges in public and private schools. UNESUM-Ciencias.
Revista Científica Multidisciplinaria, 9(2), 124-136.
https://doi.org/10.47230/unesum-ciencias.v9.n2.2025.124-136
Guevara Peñaranda, N. S., Zambrano Pachay, J. F., & Fabre Mérchan, P. G. (2024).
EFL curriculum in Ecuador: The achievement of communicative competences
in learners graduated from Ecuadorian public schools in Milagro. Ciencia Latina
Revista Científica Multidisciplinar, 8(2), 6942-6954.
https://doi.org/10.37811/cl_rcm.v8i2.11100
Guillermo Morales, L. E. (2024). El efecto del aula invertida en el aprendizaje de inglés:
Revisión sistemática. Horizontes. Revista de Investigación en Ciencias de la
Educación, 8(32), 544-559.
https://doi.org/10.33996/revistahorizontes.v8i32.743
Hernández Pacheco, J., Chacón Cárdenas, A., Lasluisa Naranjo, G., & Romero
Cevallos, M. (2025). Integración de herramientas de inteligencia artificial para
mejorar la enseñanza del idioma inglés en Ecuador. Polo del Conocimiento,
10(1), 1571-1594. https://doi.org/10.23857/pc.v10i1.8770
Jiménez-Tuza, S. B. (2025). Uso de la inteligencia artificial en la dirección de centros
educativos. Revista Científica Zambos, 4(1), 191-204.
https://doi.org/10.69484/rcz/v4/n1/86
Revista Científica Ciencia y Método | Vol.04 | Núm.01 | Ene–Mar | 2026 | www.revistacym.com pág. 532
Artículo Científico
Lopez, J., Becerra, A., & Ramirez-Avila, M. (2021). EFL speaking fluency through
authentic oral production. Journal of Foreign Language Teaching and Learning,
6. https://doi.org/10.18196/ftl.v6i1.10175
Lucas Soledispa, C. J., Mantilla Carrera, P. N., Dávila Macías, A. M., Jaramillo Crespo,
L. F., & Armijos Solano, D. O. (2023). Perspectiva de profesores de inglés
acerca del impacto de la inteligencia artificial en los cursos de idiomas. Ciencia
Latina Revista Científica Multidisciplinar, 7(4), 8278-8295.
https://doi.org/10.37811/cl_rcm.v7i4.7562
Ministerio de Educación de Ecuador. (2016). English language curriculum.
https://educacion.gob.ec/wp-content/uploads/downloads/2016/08/EFL-for-
Subnivel-BGU-final-ok.pdf
Moher, D., Liberati, A., Tetzlaff, J., & Altman, D. G. (2009). Preferred reporting items
for systematic reviews and meta-analyses: The PRISMA statement. PLoS
Medicine, 6, e1000097. https://doi.org/10.1016/j.jclinepi.2009.06.005
Oshimeje, S., & Flores Barahona, I. D. (2025). Applying authentic oral production to
improve speaking fluency through project-based learning approach on EFL
students. Ciencia Latina Revista Científica Multidisciplinar, 9(4), 2304-2323.
https://doi.org/10.37811/cl_rcm.v9i4.18849
Rojas-Burbano, S. C., & Naranjo-Andrade, S. S. (2025). Prácticas en inglés como
lengua extranjera: Enfoques innovadores y centrados en el estudiante.
MQRInvestigar, 9(3), e892. https://doi.org/10.56048/MQR20225.9.3.2025.e892
Sangacha-Tapia, L. M., Celi, R. J., Acosta-Guzmán, I. L., & Varela-Tapia, E. A. (2024).
Inteligencia artificial aplicada a procesamiento de lenguaje natural (NLP) con
Python y machine learning. Editorial Grupo AEA.
https://doi.org/10.55813/egaea.l.88
T. Soto, S., Espinosa Cevallos, L. F., & Rojas Encalada, M. A. (2025). Studies on
motivation and EFL teaching and learning in Ecuador. Revista InveCom, 5(1),
e501006. https://doi.org/10.5281/zenodo.10892449
Villarroel Carrillo, S., Castillo Salazar, L., Granda Aguilera, D., Lema Mullo, L., &
Carranza Ortiz, E. (2025). Desarrollo de un sistema educativo inteligente en
Python para la personalización del aprendizaje mediante técnicas de
inteligencia artificial como machine learning y procesamiento de lenguaje
natural. Polo del Conocimiento, 10(11), 1560-1573.
https://doi.org/10.23857/pc.v10i11.10718
Yánez-Goyes, M., Peñaherrera-Solarte, K., Carlín-Chávez, E., & Bonilla-Tenesaca, J.
(2024). Las TIC en la enseñanza del inglés para la educación básica: Una
revisión sistemática. 593 Digital Publisher CEIT, 9(3), 98-110.
https://doi.org/10.33386/593dp.2024.3.2334