Mexican trainee English teachers say chatbots gave near-daily speaking practice but felt stiff
Study title: Generative artificial intelligence and the development of English oral skills: perceptions of pre-service English language teachers
Researchers at the Escuela Normal de Atlacomulco in Mexico ran a structured intervention in the 2025–2026 academic year with 21 B1-level students in an English teaching degree. Students used ChatGPT and Gemini Flash for guided conversations, role-plays and pronunciation practice, and 8 were interviewed about their experience. They described instant, personalised feedback and more speaking practice outside class.
Read the study abstract
This study explored the perceptions of future English teachers regarding the use of generative artificial intelligence [GAI], using ChatGPT and Gemini Flash as tools to support the development of oral English skills. The objective was to document and analyze the students’ experiences with an interaction that incorporated guided conversations, role-playing, and pronunciation training. An exploratory-descriptive qualitative approach was adopted, involving 21 B1-level students from the Escuela Normal de Atla…
"student_ipad_school - 232" by flickingerbrad, CC BY 2.0.A fixed formula sets both scores from the evidence and its sources. Reader votes don’t move them.
G 46G is the gain side, out of 100. It goes up when the gain is specific, when it measurably helped people and when the sources behind it are strong.In brief
The perceptions matter because speaking is often the least prioritised language skill, with limited class time and few chances to talk outside lessons. About half of participants spent three to five days a week on 15 to 20 minutes of chatbot practice beyond what the teacher set, and none was classed as a low user.
The results are self-reported and come from one small group. Students also flagged uneven device capability, poor connectivity for a couple of them, and chatbot talk that felt mechanical, formal and unlike native speech. The authors say no pre- and post-intervention measure of oral skill was taken, so improvement in speaking is not shown.
Main points
- The study involved 21 B1-level students in an English teaching programme, with 8 purposively selected for semi-structured interviews lasting 20 to 30 minutes.
- The intervention ran six hours of interaction per week over three days a week, using three prompt types: guided conversations, role-plays and pronunciation training.
- Roughly half of participants were high users, spending three to five days a week for 15 to 20 minutes on a chatbot; the other half used it one or two days a week when asked.
- Role-plays and guided conversations were the most frequent activities, while pronunciation training was the least frequent.
- Before the study, no student had used generative AI to practise oral English, though all had used at least one of the two tools.
- Reported benefits included instant personalised feedback, higher oral fluency, greater motivation and less embarrassment about making mistakes.
- Limitations included uneven mobile device capabilities and poor internet for some; more than half of students had mobile data plans, and a couple reported very weak signal at home.
- All but one participant wanted to use such tools in future teaching; one doubted their use in rural or marginalised contexts.
The gain
Interviewed trainee English teachers at one Mexican teacher college said ChatGPT and Gemini Flash gave them instant, personalised feedback and let them practise speaking nearly every day on their phones instead of only in class.
The problem
Interviewed trainee English teachers at one Mexican teacher college said unequal device capabilities and weak home internet limited use, and that chatbot conversation felt flat, formal and unlike how native speakers actually talk.
The full story
The study used an exploratory-descriptive qualitative design at one teacher-training college. The teacher-researcher wrote prompts so the chatbots would converse at a similar level, correct errors with brief explanations, play roles such as employer, waiter or travel agent, and list words for pronunciation practice. ChatGPT and Gemini Flash were chosen because most students had a Google account and could install ChatGPT easily.
Interviews covered five dimensions: previous experience, use for oral skills, perceived benefits, limitations and teacher training. Participants were chosen by usage level, gender and academic performance, and were recorded, transcribed and coded P1 to P8. One participant said she usually scores low in the speaking section of exams and could now practise nearly every day.
Students reported being less embarrassed to make mistakes, and one shy participant said the chatbot helped her let go of inhibitions. They also said the AI stayed rigidly on its instructions, used formal English, rarely used informal expressions and did not link sounds like native speakers.
The authors conclude chatbots are a promising complement but cannot replace teachers or human interaction, and recommend contextualised, structured use. They recommend mixed-methods designs with before and after measures of oral expression in future research.
What this doesn’t fix
The study is a small qualitative sample from one college based on self-reported perceptions, and the authors say it has limitations that make findings difficult to generalise and that no before and after measures of oral expression were used.
Sources
- Peer-reviewedRevista de Tecnología y Educación2026-12-30
The AI Reality Brief
One email every Monday with the week’s AI gains and harms, each linked to the study or report behind it. It’s free.
We’ll email you a link to confirm. Unsubscribe whenever you like. What’s in it
ace
The debate