Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 209

EVALUATING AI CHATBOTS FOR GRAMMAR AND VOCABULARY ENHANCEMENT

IN EFL ONLINE CLASS: GEMINI AS MINI-TUTOR

EVALUACIÓN DE CHATBOTS DE IA PARA LA MEJORA DE GRAMÁTICA Y

VOCABULARIO EN CLASES DE EFL EN LÍNEA: GEMINI COMO MINI-TUTOR

Autores: ¹Erika Mora Herrera, ²David Gortaire Díaz, ³Gabriela Almache Granda y

Roddy Real

Roby.

¹ORCID ID: https://orcid.org/0000-0002-8156-0557

²ORCID ID: https://orcid.org/0000-0001-7364-7305

²ORCID ID: https://orcid.org/0000-0003-1858-7121

ORCID ID:

https://orcid.org/0000-0003-1474-9349

¹E-mail de contacto: emorah@utb.edu.ec

²E-mail de contacto: dgortaire@utb.edu.ec

³E-mail de contacto: galmache@utb.edu.ec

E-mail de contacto:

rreal@utb.edu.ec

Afiliación:

1*2*3*4*

Universidad Técnica de Babahoyo, (Ecuador).

Artículo recibido: 15 de Febrero del 2026

Artículo revisado: 18 de Febrerodel 2026

Artículo aprobado: 23 de Febrero del 2026

¹Licenciada en Lengua y Lingüística Inglesa, titulada en la Universidad de Filosofía Ciencia y Letras de la Educación-Escuela de Lengua,

(Ecuador). Maestría en Pedagogía del Idioma Inglés como Lengua Extranjera de la Universidad Bolivariana del Ecuador, (Ecuador).

²Ingeniero en Negocios Internacionales, Escuela Superior Politécnica del Litoral, (Ecuador), 12 años de experiencia laboral. Master en

Desarrollo Rural de la Escuela Superior Politécnica del Litoral, (Ecuador), y Master en Pedagogía del Inglés como Lengua Extranjera de

la Universidad Bolivariana del Ecuador, (Ecuador).

Ingeniera en Ciencias Empresariales, mención en Dirección y Planificación Comercial, Universidad de Especialidades Espíritu Santo,

(Ecuador). Magíster en Pedagogía de los Idiomas Nacionales y Extranjeros, mención en Enseñanza de Inglés, Universidad Casa Grande,

(Ecuador). Magíster en Educación con mención en Innovaciones Pedagógicas, Universidad Casa Grande, (Ecuador). Doctorante en

Educación e Innovación, Universidad de Investigación e Innovación de México, (México).

Ingeniero en Negocios Internacionales, titulado por la Escuela Superior Politécnica del Litoral (ESPOL), (Ecuador). Máster Universitario

en Dirección de Empresas con mención en Negocios Internacionales por la Universidad de Palermo, (Argentina). Además, posee una

Maestría en Pedagogía del Idioma Inglés como Lengua Extranjera de la Universidad Bolivariana del Ecuador, (Ecuador).

Resumen

Esta investigación examina la efectividad de

Google Gemini como tutor virtual impulsado

por inteligencia artificial para la mejora de

gramática y vocabulario en estudiantes de

inglés como lengua extranjera (EFL) en

entornos de aprendizaje en línea. El estudio

empleó un diseño de estudio de caso con

métodos mixtos, recopilando datos de

percepción de 300 estudiantes de nivel A2.1 del

Centro de Idiomas de la Universidad Técnica

de Babahoyo, Ecuador, durante un período de

intervención de tres meses (octubre-diciembre

2025). Los participantes utilizaron Gemini

tanto dentro como fuera de las clases

sincrónicas en línea mediante prompts

estructurados diseñados para práctica

gramatical, desarrollo de vocabulario y práctica

conversacional. Los datos se recopilaron a

través de encuestas que incluían escalas Likert

y preguntas abiertas, analizadas mediante

estadística descriptiva y análisis temático. Los

resultados revelan que los estudiantes

percibieron a Gemini como una herramienta

útil, accesible y generalmente efectiva, con

puntuaciones medias de percepción en el rango

de "de acuerdo" para apoyo al aprendizaje

gramatical (M = 4.01), desarrollo de

vocabulario (M = 4.02), facilidad de uso (M =

4.06) y satisfacción general (M = 4.00). Los

estudiantes valoraron particularmente la

disponibilidad 24/7, retroalimentación

inmediata y el entorno de práctica sin ansiedad.

Sin embargo, se identificaron desafíos

incluyendo respuestas ocasionalmente

inexactas (29.7%), dificultades técnicas

(25.3%) y niveles modestos de compromiso

afectivo (M = 3.82). El análisis comparativo

reveló que los usuarios frecuentes y estudiantes

con menos experiencia previa en inglés

reportaron percepciones significativamente

más positivas. Los hallazgos sugieren que los

chatbots de IA funcionan mejor como

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 210

herramientas complementarias que extienden

la instrucción en aula cuando se implementan

con andamiaje pedagógico apropiado, prompts

estructurados y apoyo continuo del instructor.

Palabras clave: Chatbots de inteligencia

artificial, Aprendizaje de Inglés como

lengua extranjera, Google Gemini.

Abstract

E This research examines the effectiveness of

Google Gemini as an AI-powered virtual tutor

for grammar and vocabulary enhancement

among English as a Foreign Language (EFL)

learners in online learning environments. The

study employed a mixed-methods case study

design, collecting perception data from 300

A2.1-level students at the Language Center of

Universidad Técnica de Babahoyo, Ecuador,

during a three-month intervention period

(October-December 2025). Participants used

Gemini both within and outside synchronous

online classes through structured prompts

designed for grammar practice, vocabulary

development, and conversational practice. Data

were collected through surveys including

Likert scales and open-ended questions,

analyzed using descriptive statistics and

thematic analysis. Results reveal that students

perceived Gemini as a useful, accessible, and

generally effective tool, with mean perception

scores in the "agree" range for grammar

learning support (M = 4.01), vocabulary

development (M = 4.02), ease of use (M =

4.06), and overall satisfaction (M = 4.00).

Students particularly valued 24/7 availability,

immediate feedback, and the anxiety-free

practice environment. However, challenges

were identified including occasionally

inaccurate responses (29.7%), technical

difficulties (25.3%), and modest levels of

affective engagement (M = 3.82). Comparative

analysis revealed that frequent users and

students with less prior English learning

experience reported significantly more positive

perceptions. Findings suggest that AI chatbots

function best as complementary tools

extending classroom instruction when

implemented with appropriate pedagogical

scaffolding, structured prompts, and ongoing

instructor support. The study contributes

empirical evidence to the emerging field of AI-

mediated language learning while highlighting

implementation considerations essential for

maximizing educational benefits and

addressing current technological limitations.

Keywords: Artificial intelligence chatbots,

English as a foreign language learning,

Google Gemini.

Sumário

Esta pesquisa examina a eficácia do Google

Gemini como um tutor virtual com inteligência

artificial para aprimorar a gramática e o

vocabulário de alunos de inglês como língua

estrangeira (EFL) em ambientes de

aprendizagem online. O estudo empregou um

delineamento de estudo de caso com métodos

mistos, coletando dados de percepção de 300

alunos de nível A2.1 do Centro de Línguas da

Universidade Técnica de Babahoyo, Equador,

durante um período de intervenção de três

meses (outubro a dezembro de 2025). Os

participantes utilizaram o Gemini tanto durante

quanto fora das aulas online síncronas, por

meio de instruções estruturadas elaboradas

para prática gramatical, desenvolvimento de

vocabulário e prática de conversação. Os dados

foram coletados por meio de questionários que

incluíam escalas Likert e questões abertas, e

analisados utilizando estatística descritiva e

análise temática. Os resultados revelam que os

alunos perceberam o Gemini como uma

ferramenta útil, acessível e, em geral, eficaz,

com pontuações médias de percepção na faixa

de "concordo" para suporte à aprendizagem

gramatical (M = 4,01), desenvolvimento de

vocabulário (M = 4,02), facilidade de uso (M =

4,06) e satisfação geral (M = 4,00). Os alunos

valorizaram particularmente a disponibilidade

24 horas por dia, 7 dias por semana, o feedback

imediato e o ambiente de prática livre de

ansiedade. No entanto, foram identificados

desafios, incluindo respostas ocasionalmente

imprecisas (29,7%), dificuldades técnicas

(25,3%) e níveis modestos de engajamento

afetivo (M = 3,82). A análise comparativa

revelou que usuários frequentes e alunos com

menos experiência prévia em inglês relataram

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 211

percepções significativamente mais positivas.

Os resultados sugerem que os chatbots de IA

funcionam melhor como ferramentas

complementares que ampliam o ensino em sala

de aula quando implementados com suporte

pedagógico adequado, instruções estruturadas

e acompanhamento contínuo pelo instrutor.

Palavras-chave: Chatbots de inteligência

artificial, Aprendizagem de inglês como

língua estrangeira, Google Gemini.

Introduction

The integration of artificial intelligence (AI) in

language education has emerged as a

transformative force in contemporary

pedagogical practices, particularly in the

context of English as a Foreign Language (EFL)

instruction (Qassrawi et al., 2024). As

educational institutions increasingly adopt

digital and hybrid learning modalities, the

demand for innovative, personalized, and

accessible learning tools has intensified

(Wulandari & Purnamaningwulan, 2024). AI-

powered chatbots represent a promising

technological advancement that addresses

critical challenges in EFL education, including

limited opportunities for authentic interaction,

insufficient personalized feedback, and the

scalability constraints inherent in traditional

classroom settings (Lara et al., 2023a). These

intelligent conversational agents have

demonstrated potential to serve as virtual tutors,

providing learners with immediate,

individualized support for developing

fundamental language competencies such as

grammar accuracy and vocabulary acquisition

(Flores, 2024).

Grammar and vocabulary constitute

foundational pillars of language proficiency,

serving as essential building blocks for effective

communication in English (Walker et al.,

2020a). For EFL learners, mastering these

linguistic components presents considerable

challenges, as they must internalize complex

grammatical structures and extensive lexical

repertoires while navigating limited exposure to

authentic language use outside the classroom

environment (Walker et al., 2020b). Traditional

instructional approaches often struggle to

provide the intensive, individualized practice

necessary for sustained improvement,

particularly in large-enrollment online courses

where instructor-student ratios constrain

opportunities for personalized intervention

(Puspitasari et al., 2023). The asynchronous

nature of many online EFL programs further

exacerbates these limitations, as learners may

experience delayed feedback and reduced

opportunities for spontaneous language practice

(Sholikhah & Ningsih, 2023). Consequently,

educational researchers and practitioners have

increasingly explored technological solutions

capable of supplementing human instruction

with scalable, responsive learning support

systems.

Among the proliferating landscape of

generative AI technologies, Google's Gemini

has emerged as a sophisticated large language

model with multimodal capabilities and

advanced natural language processing functions

(Sun et al., 2021). Unlike earlier generation

chatbots that relied primarily on rule-based or

retrieval-based architectures, Gemini leverages

transformer-based neural networks trained on

extensive corpora, enabling more nuanced

understanding of linguistic contexts, more

accurate error detection, and more

pedagogically appropriate feedback generation

(Zhai y Wibowo, 2023). The potential

application of Gemini as a "mini-tutor" in EFL

contexts warrants systematic investigation, as

its capacity to engage in extended dialogues,

provide explanatory feedback, generate

contextualized examples, and adapt to

individual learner needs may address critical

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 212

gaps in online language instruction (Hwang &

Nurtantyana, 2022). However, despite the

growing enthusiasm surrounding AI chatbots in

educational contexts, rigorous empirical

evaluation of their effectiveness specifically for

grammar and vocabulary enhancement in EFL

online environments remains limited (Pikhart,

2020).

The purpose of this study is to evaluate the

effectiveness of Gemini as an AI-powered mini-

tutor for enhancing grammar accuracy and

vocabulary development among EFL learners in

online classroom settings. This investigation

addresses a critical gap in the literature by

systematically examining both the pedagogical

efficacy and the practical implementation

considerations of integrating Gemini into

structured EFL curricula. Specifically, this

research explores how Gemini's conversational

capabilities, error correction mechanisms, and

adaptive feedback functions contribute to

measurable improvements in learners'

grammatical competence and lexical

knowledge. Additionally, the study investigates

learner perceptions, engagement patterns, and

potential challenges associated with AI-

mediated language learning, providing insights

essential for educators contemplating the

integration of such technologies into their

instructional designs. By examining these

dimensions through a rigorous methodological

framework, this research contributes to the

evolving discourse on AI in education while

offering practical guidance for EFL

practitioners navigating the digital

transformation of language teaching.

Materials and Methods

This study employed a case study design with a

mixed-methods approach to evaluate the

effectiveness of Google Gemini as an AI-

powered mini-tutor for grammar and

vocabulary enhancement among EFL learners

in online classes. Case study methodology was

selected as the most appropriate research design

due to its capacity to provide in-depth,

contextualized examination of contemporary

phenomena within real-life settings,

particularly when boundaries between

phenomenon and context are not clearly evident

(Dela y Dela, 2026). The study population

comprised 300 students enrolled at Level 3

(A2.1 according to the Common European

Framework of Reference for Languages) at the

Language Center of Universidad Técnica de

Babahoyo in Ecuador during the October-

December 2025 academic term. Participants

were selected through convenience sampling, a

non-probabilistic sampling technique wherein

participants are chosen based on their

accessibility and availability to the researcher

(Lu, 2019). In the context of this study, the

selection of Level 3 students was purposeful

and theoretically motivated, as A2.1 learners

represent an intermediate-low proficiency level

where students possess foundational

grammatical knowledge and vocabulary but

require extensive practice and reinforcement to

consolidate emerging competencies (Conde et

al., 2022). The demographic profile of

participants reflected the typical composition of

university-level EFL programs in Ecuador, with

students primarily between 18 and 25 years of

age enrolled in various undergraduate programs

who were fulfilling institutional English

language requirements. The Language Center at

Universidad Técnica de Babahoyo serves

students across multiple academic faculties,

creating a heterogeneous participant pool with

diverse disciplinary backgrounds.

The intervention period extended across three

months (October through December 2025),

encompassing a full academic semester during

which participants engaged with Google

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 213

Gemini both within synchronous online class

sessions and independently for asynchronous

practice outside scheduled class time. The

integration of Gemini into the curriculum

followed principles of blended learning design,

wherein technology-mediated activities

complemented rather than replaced instructor-

led instruction, creating a pedagogical ecology

that leveraged affordances of both human

teaching and AI-powered practice opportunities

(Lara et al., 2023b). During synchronous online

class sessions conducted via video conferencing

platforms, the instructor introduced specific

grammatical structures or vocabulary sets

through explicit instruction, modeling, and

guided practice, then assigned targeted chatbot

interactions as follow-up activities for

independent practice and consolidation (Celik,

2023).

To facilitate productive interactions between

students and Gemini, researchers developed a

structured collection of prompts specifically

designed to elicit grammar and vocabulary

practice aligned with the Level 3 curriculum

objectives. These prompts were carefully

crafted following best practices in AI prompt

engineering for educational purposes,

incorporating clear task specifications,

appropriate scaffolding, and explicit guidance

for the types of linguistic output expected from

students. Students were provided with

comprehensive orientation sessions at the

beginning of the intervention period that

explained the educational rationale for using AI

chatbots, demonstrated effective interaction

strategies, addressed common concerns about

data privacy and appropriate use, and

established clear expectations for required

chatbot engagement both during and outside

class time. Data collection centered primarily

on student perceptions of the chatbot activities

and their engagement with Gemini, assessed

through a comprehensive survey instrument

administered at the conclusion of the

intervention period in December 2025. The

survey was designed following established

principles of questionnaire construction in

applied linguistics research, incorporating both

closed-ended Likert-scale items for quantitative

analysis and open-ended questions for

qualitative insights.

The survey instrument was developed through

an iterative process involving initial item

generation based on relevant literature on

technology acceptance and language learning

attitudes, expert review by experienced EFL

instructors and educational technology

specialists, and pilot testing with a small group

of students not included in the main study to

identify problematic items or unclear wording

(Sun et al., 2021). The survey was administered

electronically through a secure online platform

during the final two weeks of the semester,

allowing students to complete it at their

convenience while ensuring all responses were

collected before course completion. Data

analysis proceeded through multiple phases

employing descriptive statistical techniques

appropriate for survey research and exploratory

case study investigation. Quantitative data from

Likert-scale survey items were analyzed using

SPSS (Statistical Package for the Social

Sciences) or equivalent statistical software to

calculate measures of central tendency (means,

medians) and dispersion (standard deviations,

ranges) for each survey item and constructed

scale. Qualitative data from open-ended survey

questions were analyzed using thematic

analysis procedures, a flexible method for

identifying, analyzing, and reporting patterns

within qualitative data.

Results y Discussion

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 214

The analysis of survey data from 300 Level 3

EFL students at Universidad Técnica de

Babahoyo revealed generally positive

perceptions of Google Gemini as an AI-

powered mini-tutor for grammar and

vocabulary enhancement during the October-

December 2025 intervention period.

Table 1. Demographic Characteristics of

Participants

Characteristic

Category

Gender

Female

178

59.3

Male

118

39.3

Non-binary/Prefer not to

say

1.3

Age

18-20 years

162

54.0

21-23 years

32.7

24-26 years

10.7

27+ years

2.7

Academic Faculty

Engineering

27.3

Business Administration

22.7

Health Sciences

18.0

Education

15.7

Agricultural Sciences

11.7

Social Sciences

4.7

Prior English Learning

Experience

Less than 2 years

15.0

2-4 years

128

42.7

5-7 years

32.3

More than 7 years

10.0

Previous AI Chatbot

Experience

None

187

62.3

Limited (1-3 times)

25.3

Moderate (4-10 times)

9.3

Extensive (10+ times)

3.0

Daily Internet Access

Limited (less than 2 hours)

7.7

Moderate (2-4 hours)

29.7

Substantial (5-8 hours)

142

47.3

Extensive (8+ hours)

15.3

Source: Own elaboration

Table 1 presents the demographic composition

of the 300 participants who completed the

study. The sample demonstrated considerable

diversity across multiple characteristics

relevant to interpretation of findings. Gender

distribution showed a moderate female majority

(59.3%), which is consistent with enrollment

patterns in language programs at Ecuadorian

universities. The age distribution revealed that

the majority of participants were traditional

university-age students, with 86.7% falling

between 18 and 23 years old, though a smaller

proportion of mature students (13.3% aged 24

or older) provided perspectives from learners

balancing academic study with work or family

responsibilities. Academic faculty

representation spanned six major disciplinary

areas, with Engineering (27.3%) and Business

Administration (22.7%) constituting the largest

groups, reflecting the institutional profile of

Universidad Técnica de Babahoyo as a

comprehensive university with strengths in

applied and professional programs.

Table 2. Descriptive Statistics for Student

Perceptions of Gemini for Grammar Learning

Item

Median

Mode

Min

Max

Gemini helped me understand English

grammar rules better

4.12

0.78

4.00

The grammar explanations provided by

Gemini were clear and easy to understand

3.98

0.83

4.00

Gemini accurately identified grammar

errors in my writing

3.85

0.91

4.00

The feedback on grammar from Gemini

was helpful for improving my English

4.05

0.81

4.00

I could practice grammar structures more

frequently using Gemini than in regular

classes

4.23

0.76

4.00

Gemini provided examples that helped me

understand how to use grammar correctly

4.08

0.79

4.00

Using Gemini increased my confidence in

using English grammar

3.92

0.88

4.00

The grammar corrections from Gemini

were appropriate to my level (A2.1)

3.87

0.86

4.00

Overall Grammar Learning Perception

Scale

4.01

0.68

4.06

—

2.13

5.00

Source: Own elaboration

Table 2 displays descriptive statistics for eight

items assessing student perceptions of Gemini's

effectiveness for grammar learning, along with

the composite Grammar Learning Perception

Scale. The overall scale mean of 4.01 (SD =

0.68) indicates that participants generally

agreed that Gemini provided valuable support

for grammar development, with the composite

score falling solidly in the "agree" range and

demonstrating acceptable internal consistency

(α = .89). Among individual items, participants

expressed strongest agreement with the

statement "I could practice grammar structures

more frequently using Gemini than in regular

classes" (M = 4.23, SD = 0.76), highlighting a

key affordance of AI chatbots identified in

previous research: their capacity to provide

unlimited practice opportunities beyond the

temporal and logistical constraints of traditional

instruction (Pikhart, 2020). Students also

positively evaluated Gemini's explanatory

capabilities, with mean scores above 4.0 for

items related to understanding grammar rules

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 215

better, receiving helpful feedback, and

obtaining useful examples of correct usage.

Table 3. Descriptive Statistics for Student

Perceptions of Gemini for Vocabulary

Development

Item

Median

Mode

Min

Max

Gemini helped me learn new English

vocabulary words

4.18

0.74

4.00

The vocabulary explanations from

Gemini were clear and included good

examples

4.06

0.79

4.00

Gemini helped me understand how to use

vocabulary words in different contexts

3.95

0.84

4.00

I learned more vocabulary using Gemini

than I would have learned without it

4.02

0.82

4.00

Gemini provided useful synonyms and

related words that expanded my

vocabulary

4.15

0.77

4.00

The chatbot helped me remember

vocabulary words better through

conversation practice

3.88

0.89

4.00

Using Gemini increased my confidence

in using new vocabulary words

3.91

0.86

4.00

Gemini helped me understand the

differences between similar vocabulary

words

3.97

0.83

4.00

Overall Vocabulary Development

Perception Scale

4.02

0.66

4.00

—

2.25

5.00

Source: Own elaboration

Table 3 presents descriptive statistics for items

measuring student perceptions of Gemini's

contribution to vocabulary development. The

overall Vocabulary Development Perception

Scale achieved a mean of 4.02 (SD = 0.66) with

excellent internal consistency (α = .91),

indicating strong agreement that the chatbot

effectively supported lexical learning. Students

rated highest the item "Gemini helped me learn

new English vocabulary words" (M = 4.18, SD

= 0.74), suggesting that participants perceived

genuine vocabulary gains from their chatbot

interactions. Additional highly rated aspects

included Gemini's provision of useful

synonyms and related words (M = 4.15, SD =

0.77) and clear explanations with good

examples (M = 4.06, SD = 0.79) Overall, these

results indicate that Gemini was perceived as an

effective tool for vocabulary expansion and

initial learning, though its contribution to

deeper lexical processing and long-term

retention may have been more variable across

students.

Table 4. Descriptive Statistics for Ease of Use,

Engagement, and Satisfaction with Gemini (N =

300)

Dimension/Item

Median

Mode

Ease of Use

Gemini was easy to access and use

4.32

0.72

4.00

I did not experience technical problems when using

Gemini

3.67

1.04

4.00

The chatbot interface was user-friendly and intuitive

4.08

0.81

4.00

I could easily understand how to interact with Gemini

4.15

0.76

4.00

Ease of Use Scale

4.06

0.69

4.00

—

Engagement

Interacting with Gemini was interesting and engaging

3.89

0.87

4.00

I enjoyed practicing English with the chatbot

3.76

0.94

4.00

Using Gemini motivated me to practice English more

frequently

3.82

0.91

4.00

I felt more comfortable making mistakes with Gemini

than with my teacher or classmates

4.21

0.79

4.00

I looked forward to chatbot practice activities

3.54

0.98

4.00

The chatbot kept my attention during practice sessions

3.68

0.93

4.00

Engagement Scale

3.82

0.74

3.83

—

Overall Satisfaction

Overall, I am satisfied with using Gemini for learning

English

3.94

0.84

4.00

I would recommend using Gemini to other English

students

3.98

0.82

4.00

I plan to continue using AI chatbots for language learning

in the future

4.05

0.80

4.00

Using Gemini was a valuable addition to my English

course

4.02

0.79

4.00

Overall Satisfaction Scale

4.00

0.72

4.00

—

Source: Own elaboration

Table 4 presents three conceptually distinct but

related dimensions of student perceptions: ease

of use, engagement, and overall satisfaction

with Gemini. The Ease of Use scale achieved a

mean of 4.06 (SD = 0.69, α = .84), indicating

that students generally found Gemini accessible

and user-friendly. The highest-rated ease of use

item was "Gemini was easy to access and use"

(M = 4.32, SD = 0.72), suggesting that the

technical implementation was successful in

minimizing barriers to student participation.

However, the item "I did not experience

technical problems when using Gemini"

received notably lower ratings (M = 3.67, SD =

1.04) and exhibited the highest standard

deviation in the entire survey, indicating

substantial variability in technical experiences,

suggesting that despite modest engagement

levels, the perceived benefits were sufficient to

motivate continued adoption beyond the

required intervention period.

Table 5. Frequency of Gemini Usage Patterns

During the Intervention Period

Usage Pattern

Frequency of Use (per week)

Rarely (1-2 times)

16.0

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 216

Occasionally (3-4 times)

127

42.3

Frequently (5-6 times)

29.7

Very frequently (7+ times)

12.0

Average Duration per Session

Short (less than 10 minutes)

20.7

Moderate (10-20 minutes)

152

50.7

Extended (21-30 minutes)

21.3

Very extended (more than 30

minutes)

7.3

Primary Context of Use

Only during class activities

18.7

Mostly during class,

occasionally outside

32.7

Equally during and outside class

29.0

Mostly outside class,

occasionally during

15.7

Only outside class for

independent practice

4.0

Most Common Activity Types

(multiple selections allowed)

Grammar error correction

practice

234

78.0

Vocabulary expansion and

definitions

267

89.0

Conversational practice on

assigned topics

189

63.0

Requesting explanations of

grammar rules

212

70.7

Getting examples of vocabulary

in context

245

81.7

Practicing writing

sentences/paragraphs

156

52.0

Asking questions about English

language

198

66.0

Device Used for Access

Smartphone

203

67.7

Laptop computer

23.7

Desktop computer

6.0

Tablet

2.7

Source: Own elaboration

Table 5 provides detailed information about

how students actually engaged with Gemini

during the three-month intervention period,

revealing considerable variability in usage

patterns across the participant sample.

Regarding frequency of use, the modal pattern

was occasional use (3-4 times per week,

42.3%), with substantial minorities engaging

frequently (5-6 times weekly, 29.7%) or very

frequently (7+ times weekly, 12.0%), while

16.0% used the chatbot rarely (1-2 times

weekly). Regarding context of use, the data

revealed relatively balanced distribution

between primarily in-class use (51.4% when

combining the first two categories) and

substantial out-of-class independent practice

(48.7% when combining the last three

categories), suggesting successful

implementation of the blended learning design

where chatbot activities bridged formal

instruction and autonomous learning. The

activity type data provides valuable insights

into how students actually utilized Gemini, with

vocabulary expansion and definitions being the

most common activity (89.0%), followed

closely by getting examples of vocabulary in

context (81.7%) and grammar error correction

practice (78.0%).

Table 6. Comparison of Perception Scores

Across Student Characteristics

Characteristic

Grammar

Learning

M (SD)

Vocabulary

Development

M (SD)

Engagement

M (SD)

Overall

Satisfaction

M (SD)

Gender

Female (n = 178)

4.08 (0.65)

4.09 (0.62)

3.87 (0.71)

4.05 (0.69)

Male (n = 118)

3.91 (0.72)

3.92 (0.71)

3.74 (0.78)

3.92 (0.76)

Age Group

18-20 years (n = 162)

4.05 (0.67)

4.06 (0.65)

3.91 (0.72)

4.08 (0.70)

21-23 years (n = 98)

3.98 (0.69)

3.99 (0.68)

3.76 (0.75)

3.94 (0.73)

24+ years (n = 40)

3.93 (0.71)

3.94 (0.67)

3.65 (0.79)

3.82 (0.78)

Prior English Learning

Experience

Less than 4 years (n =

173)

4.09 (0.66)

4.11 (0.63)

3.89 (0.71)

4.07 (0.69)

5+ years (n = 127)

3.90 (0.70)

3.89 (0.69)

3.71 (0.77)

3.90 (0.75)

Previous AI Chatbot Experience

None (n = 187)

3.96 (0.70)

3.98 (0.68)

3.76 (0.76)

3.94 (0.74)

Some experience (n =

113)

4.11 (0.63)

4.09 (0.62)

3.93 (0.69)

4.10 (0.67)

Frequency of Gemini

Use

Rarely/Occasionally

3.87 (0.72)

3.89 (0.71)

3.65 (0.79)

3.82 (0.77)

Frequently/Very

frequently

4.21 (0.58)

4.21 (0.56)

4.06 (0.62)

4.24 (0.61)

Academic Faculty

STEM (Engineering,

Health, Agriculture) (n

= 171)

3.97 (0.69)

3.99 (0.67)

3.79 (0.75)

3.96 (0.73)

Non-STEM (Business,

Education, Social

Sciences) (n = 129)

4.07 (0.66)

4.06 (0.65)

3.85 (0.72)

4.05 (0.70)

Overall Sample (N =

300)

4.01 (0.68)

4.02 (0.66)

3.82 (0.74)

4.00 (0.72)

Source: Own elaboration

Table 7 presents comparative descriptive

statistics examining whether perception scores

varied systematically across different student

characteristics, providing insights into which

learner populations may benefit most from AI

chatbot integration and which may require

additional support or modified approaches.

Gender comparisons revealed that female

students reported slightly higher mean scores

across all perception dimensions, with the

largest difference observed for grammar

learning perceptions (Female M = 4.08 vs. Male

M = 3.91). While these differences appear

modest in absolute terms, they may reflect

documented gender differences in technology

acceptance, learning strategy preferences, or

willingness to engage with novel educational

technologies, though interpretations must be

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 217

cautious given the descriptive nature of these

comparisons (Tarhini et al., 2017).

Age group comparisons showed a pattern

wherein younger students (18-20 years)

consistently reported more positive perceptions

than older students (24+ years), with the largest

difference again appearing in engagement

scores (18-20 years M = 3.91 vs. 24+ years M =

3.65). This finding might reflect greater digital

nativity and comfort with AI technologies

among younger learners, or alternatively, could

indicate that mature students with established

learning preferences found the chatbot less

compatible with their preferred approaches to

language study. The comparison by prior

English learning experience revealed an

unexpected pattern: students with less

experience (under 4 years) reported more

positive perceptions across all dimensions than

those with more extensive backgrounds (5+

years).

The findings from this case study provide

empirical evidence that Google Gemini can

serve as an effective AI-powered mini-tutor for

grammar and vocabulary enhancement in EFL

online classes, while also revealing important

nuances, limitations, and implementation

considerations that warrant careful attention

from educators and researchers. The overall

pattern of results indicates that students at the

A2.1 proficiency level perceived Gemini as a

useful, accessible, and generally effective tool

for language learning, with mean perception

scores consistently falling in the "agree" range

across dimensions of grammar learning support,

vocabulary development, ease of use, and

overall satisfaction.

These quantitative findings align with emerging

research suggesting that contemporary large

language models possess sufficient linguistic

sophistication and pedagogical capabilities to

provide meaningful support for language

learners, particularly in domains requiring

extensive practice, immediate feedback, and

personalized explanations (Wang, 2019; Wang,

2022). However, the qualitative data and

variability observed across perception items

reveal a more complex picture wherein the

effectiveness of chatbot-mediated learning

depended critically on factors such as

implementation design, student engagement

patterns, individual learning preferences, and

the specific linguistic features being practiced.

Students' positive perceptions of Gemini's

contributions to grammar learning (M = 4.01,

SD = 0.68) and vocabulary development (M =

4.02, SD = 0.66) suggest that the chatbot

successfully addressed key challenges inherent

in EFL instruction, particularly the provision of

extensive practice opportunities and immediate,

personalized feedback that are difficult to

deliver consistently in traditional classroom

settings (Astrini et al., 2024). The finding that

students most strongly endorsed the statement

about practicing grammar structures more

frequently with Gemini than in regular classes

(M = 4.23) empirically validates theoretical

claims that AI chatbots' primary pedagogical

value lies in their capacity to supplement rather

than replace human instruction by providing

unlimited, on-demand practice that extends

learning beyond temporal and spatial

constraints of formal lessons (Bibauw et al.,

2019; Huang et al., 2022).

For vocabulary development, the strong ratings

for Gemini's provision of synonyms, related

words, and contextual examples (M = 4.06-4.18

across relevant items) suggest alignment with

research-supported vocabulary teaching

principles emphasizing semantic elaboration,

network building, and encounter with words in

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 218

varied authentic contexts (Boers, 2021; Lindner

et al., 2019). However, the relatively lower

rating for the item assessing retention through

conversation practice (M = 3.88, SD = 0.89) and

the qualitative observation that vocabulary

gains may have been concentrated in receptive

knowledge raises questions about the depth of

processing actually occurring during chatbot

interactions. This interpretation suggests that

effectiveness of chatbots for vocabulary

learning may depend critically on task design,

with more open-ended communicative

activities potentially yielding deeper processing

than definitional queries or structured exercises,

though this hypothesis requires empirical

investigation through comparative studies

examining different interaction types.

The variability observed across student

experiences and the identification of both

significant advantages and challenges point to

the critical importance of thoughtful

pedagogical design and implementation support

rather than simply deploying technology and

expecting positive outcomes (Chen et al., 2020;

Wang, 2022). The TPACK framework

emphasizes that effective technology

integration requires intersection of

technological knowledge, pedagogical

knowledge, and content knowledge, and

findings from this study reinforce that providing

students with chatbot access alone is

insufficient—success requires structured

guidance on how to interact productively,

explicit training in prompt formulation,

integration with curricular objectives and

classroom instruction, and ongoing instructor

support to troubleshoot problems and mediate

learning (Celik, 2023; Reyes et al., 2017).

The instructor's role in chatbot-enhanced

courses emerges as fundamentally important

but qualitatively different from traditional

teaching roles. Rather than serving as the

primary source of input, explanation, and

feedback, instructors in chatbot-integrated

contexts function more as designers of learning

sequences, curators of productive prompts and

activities, monitors of student engagement

patterns, interveners when students encounter

difficulties or confusion, and facilitators of

metalinguistic reflection that helps students

extract generalizable knowledge from their

chatbot interactions (Lindner et al., 2019;

Wulandari y Purnamaningwulan, 2024). This

study's findings contribute to evolving

theoretical understanding of how AI chatbots

function within language learning ecologies and

what mechanisms account for their pedagogical

effects (Hernández & Rodríguez, 2024; Lu,

2019). The findings also have implications for

understanding the role of feedback in language

learning. The immediate, individualized

feedback provided by Gemini represents a form

of automated written corrective feedback that

has been extensively studied in CALL research,

but with important distinctions from traditional

automated feedback systems (Divekar* et al.,

2022). The chatbot context may influence

uptake in complex ways: reduced anxiety might

increase receptiveness to feedback, but the non-

human source might reduce credibility or

attention compared to feedback from respected

human instructors (Lameras & Arnab, 2021).

Conclusions

This case study investigated the effectiveness of

Google Gemini as an AI-powered mini-tutor for

grammar and vocabulary enhancement among

300 A2.1-level EFL students at Universidad

Técnica de Babahoyo during a three-month

online course intervention. The findings reveal

that students generally perceived Gemini as a

valuable, accessible, and effective learning tool,

with mean perception scores consistently in the

"agree" range across dimensions of grammar

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 219

learning support (M = 4.01), vocabulary

development (M = 4.02), ease of use (M = 4.06),

and overall satisfaction (M = 4.00). Qualitative

analysis identified key affordances that students

valued most highly: 24/7 availability enabling

flexible practice schedules, immediate

personalized feedback addressing individual

learning needs, and a low-anxiety environment

facilitating risk-taking and experimentation

without fear of social judgment. These

advantages address persistent challenges in EFL

education, particularly the provision of

extensive, individualized practice opportunities

that exceed what is feasible in traditional

classroom settings constrained by time,

resources, and instructor-student ratios (An et

al., 2023).

However, the study also revealed important

limitations and implementation challenges that

temper uncritical enthusiasm about AI chatbots

in language education. Students reported

occasional inaccurate or confusing responses

from Gemini (29.7%), technical connectivity

difficulties (25.3%), and concerns about

knowing when to trust chatbot feedback

(28.0%), highlighting that current LLM

technology, while sophisticated, remains

imperfect and requires critical user engagement

rather than passive acceptance. The modest

engagement scores (M = 3.82) relative to

perceived usefulness suggest that while students

recognized the instrumental value of chatbot

practice, many did not find interactions

intrinsically motivating or enjoyable, with

usage patterns showing considerable variability

and only 12% of students achieving daily

practice. These findings underscore that

successful chatbot integration requires

thoughtful pedagogical design including

structured prompts, explicit training in effective

usage, integration with course requirements and

assessments, ongoing instructor support, and

deliberate strategies to sustain engagement

beyond initial novelty (Godwin, 2022; Pérez,

2021).

The theoretical and practical contributions of

this research include empirical documentation

of student perceptions across multiple

dimensions of chatbot-mediated learning in an

authentic educational context, identification of

specific affordances and limitations of

contemporary LLMs for grammar and

vocabulary instruction, development and

validation of structured prompt frameworks

aligned with curricular objectives, and

illumination of implementation factors that

influence effectiveness including usage

frequency, student characteristics, and

pedagogical scaffolding. The finding that high-

frequency users reported substantially more

positive perceptions than occasional users

points to the critical importance of promoting

sustained, consistent engagement rather than

sporadic use. The comparative analyses

suggesting that less experienced learners and

students with prior chatbot exposure reported

more favorable perceptions indicate that

effectiveness may vary across learner

populations and that targeted support may be

needed for skeptical or technologically

inexperienced students.

Important limitations constrain the conclusions

that can be drawn from this study, most notably

the reliance on self-reported perception data

without direct measures of learning outcomes,

the case study design precluding causal

inference, and the convenience sampling

limiting generalizability beyond the specific

Ecuadorian university context. Future research

should address these limitations through

experimental or quasi-experimental designs

with comparison groups, incorporation of

validated pre-post assessments measuring

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 220

actual grammar and vocabulary gains, analysis

of chatbot interaction transcripts to understand

productive versus unproductive engagement

patterns, and longitudinal investigations

examining whether benefits accumulate and

engagement sustains over extended timeframes

(Mackey & Gass, 2016; Plonsky & Oswald,

2014). Additionally, research is needed on

optimal integration strategies including ideal

balance between chatbot and human instruction,

most effective prompt designs for different

learning objectives and proficiency levels, and

interventions to support sustained motivation

and self-regulated practice.

Referencias Bibliográficas

An, X., Chai, C., Li, Y., Zhou, Y., Shen, X.,

Zheng, C., & Chen, M. (2023). Modeling

English teachers’ behavioral intention to use

artificial intelligence in middle schools.

Education and Information Technologies,

28(5), 5187–5208.

https://doi.org/10.1007/s10639-022-11286-z

Astrini, A., Prastiwi, Y., & Sutopo, A. (2024).

Exploring students’ experience of using

Nearpod in grammar lesson in relation with

students’ increased engagement: A

descriptive qualitative approach. Journal of

International Multidisciplinary Research,

2(5), 205–214.

Boers, F. (2021). Evaluating second language

vocabulary and grammar instruction: A

synthesis of the research on teaching words,

phrases, and patterns. Routledge.

Celik, I. (2023). Towards Intelligent-TPACK:

An empirical study on teachers’ professional

knowledge to ethically integrate artificial

intelligence-based tools into education.

Computers in Human Behavior, 138,

107468.

https://doi.org/10.1016/j.chb.2022.107468

Chen, L., Chen, P., & Lin, Z. (2020). Artificial

intelligence in education: A review. IEEE

Access, 8, 75264–75278.

https://doi.org/10.1109/ACCESS.2020.2988

510

Conde, L., Cueva, G., Chamba, L., & Ureña, M.

(2022). Impact of artificial intelligence in

basic general education in Ecuador. 2022

17th Iberian Conference on Information

Systems and Technologies (CISTI), 1–7.

Dela, C., & Dela, L. (2026). Leveraging

artificial intelligence for intelligent student

support: An AI-enabled SRM framework for

higher education. Multidisciplinary Science

Journal, 8(3).

https://doi.org/10.31893/multiscience.20261

Divekar, R., Drozdal, J., Chabot, S., Zhou, Y.,

Su, H., Chen, Y., Zhu, H., Hendler, J., &

Braasch, J. (2022). Foreign language

acquisition via artificial intelligence and

extended reality: Design and evaluation.

Computer Assisted Language Learning,

35(9), 2332–2360.

https://doi.org/10.1080/09588221.2021.187

9162

Flores, C. (2024). La evaluación educativa en la

era de la inteligencia artificial: Cambios de

paradigmas. LATAM Revista

Latinoamericana de Ciencias Sociales y

Humanidades, 5(1).

https://doi.org/10.56712/latam.v5i1.1694

Hernández, N., & Rodríguez, M. (2024).

Inteligencia artificial aplicada a la educación

y la evaluación educativa en la universidad:

Introducción de sistemas de tutorización

inteligentes y otras tendencias futuras.

Revista de Educación a Distancia, 24(78).

https://doi.org/10.6018/red.594651

Hwang, W., & Nurtantyana, R. (2022). The

integration of multiple recognition

technologies and artificial intelligence to

facilitate EFL writing in authentic contexts.

International Conference on Information

Technology (InCIT), 379–383.

https://doi.org/10.1109/InCIT56086.2022.1

0067490

Lameras, P., & Arnab, S. (2021). Power to the

teachers: An exploratory review on artificial

intelligence in education. Information, 13(1),

14.

Lara, R., Criollo, L., Calderón, C., & Matamba,

B. (2023). La inteligencia artificial: Análisis

del presente y futuro en la educación

Ciencia y Educación

(L-ISSN: 2790-8402 E-ISSN: 2707-3378)

Vol. 7 No. 2.2

Edición Especial II 2026

Página 221

superior. Revista Científica Multidisciplinar

G-Nerando, 4(1).

Pikhart, M. (2020). Intelligent information

processing for language education: The use

of artificial intelligence in language learning

apps. Procedia Computer Science, 176,

1412–1419.

https://doi.org/10.1016/j.procs.2020.09.151

Puspitasari, I., Pureka, M., & Azizah, N. (2023).

Integrating ICT: EFL students challenge in

learning grammar. Proceeding of

Conference on English Language Teaching

(CELTI 2023), 3, 433–441.

Qassrawi, R., ElMashharawi, A., Itmeizeh, M.,

& Tamimi, M. (2024). AI-powered

applications for improving EFL students’

speaking proficiency in higher education.

Forum for Linguistic Studies, 6(5), 535–549.

https://doi.org/10.30564/fls.v6i5.6966

Sholikhah, N., & Ningsih, F. (2023). Interactive

pathways: Exploring students’ acceptance of

using Nearpod for English grammar

proficiency. Jurnal Bahasa Lingua Scientia,

15(2), 415–439.

Sun, Z., Anbarasan, M., & Kumar, D. (2021).

Design of online intelligent English teaching

platform based on artificial intelligence

techniques. Computational Intelligence,

37(3), 1166–1180.

Walker, N., Monaghan, P., Schoetensack, C., &

Rebuschat, P. (2020). Distinctions in the

acquisition of vocabulary and grammar: An

individual differences approach. Language

Learning, 70(S2), 221–254.

Wang, Z. (2022). Computer-assisted EFL

writing and evaluations based on artificial

intelligence: A case from a college reading

and writing course. Library Hi Tech, 40(1),

80–97.

Wulandari, M., & Purnamaningwulan, R.

(2024). Exploring Indonesian EFL pre-

service teachers’ experiences in AI-assisted

teaching practicum: Benefits and drawbacks.

LLT Journal: A Journal on Language and

Language Teaching, 27(2), 878–894.

Zhai, C., & Wibowo, S. (2023). A systematic

review on artificial intelligence dialogue

systems for enhancing English as foreign

language students’ interactional competence

in the university. Computers and Education:

Artificial Intelligence, 4, 100134.

https://doi.org/10.1016/j.caeai.2023.100134

Esta obra está bajo una licencia de

Creative Commons Reconocimiento-No Comercial

David Gortaire Díaz, Gabriela Almache Granda y

Roddy Real Roby.