Image: Mohamed Hassan (CC0), via Wikimedia Commons
Every teacher I know is asking some version of the same question right now: Should I let my students use AI, and if so, how?
For my Master's thesis at the Universidad de Salamanca, I turned that worry into a research question: ¿Sigue siendo comunicativa la enseñanza comunicativa? In other words, is communicative language teaching still communicative once AI enters the classroom?
This post shares what I explored, what I've seen work, and what I'd avoid. No hype, no panic. Just what happens when real students write in a second language with an AI tool at their side.
Essential question: When a student writes with AI, whose voice ends up on the page?
The real risk isn't cheating
Most conversations about AI in schools start with academic dishonesty. That matters, but in a language classroom there's a deeper issue: authorship.
Roz Ivanič (1998) reminds us that writing is never just producing correct sentences; it's building an identity as a writer. When AI "fixes" a student's text, the result may be more accurate, but is it still theirs? Did they make choices, or did they accept them?
And Anna Camps (2003) shows that real language learning in writing happens through metalinguistic reflection: thinking and talking about language while you use it. If AI does the thinking, that reflection disappears, and so does much of the learning.
Image: DALL-E (Public domain), via Wikimedia Commons
My study in a nutshell
I worked with 22 seventh-grade Spanish students and followed their writing through four moments:
- Initial draft: students wrote on their own, with no AI.
- AI interaction: they asked an AI tool for feedback on their text.
- Final version: they decided what to keep, change, or reject.
- Metalinguistic reflection: they explained why they made those decisions.
The students were B1–B2 learners, including heritage and native speakers, and 20 of them completed all four moments. Here is what I found.
Finding 1: A polished final text is not proof of learning. This was the most important result. Some of the best final texts came from students who reflected deeply on every decision. Others, just as polished, came from students who accepted everything the AI suggested and could not explain a single change. When asked what they learned, one student answered simply: "No, no aprendí nada nuevo." Looking only at the final product, you would never know the difference.
Finding 2: Authorship is a process, not a product. Comparing first drafts and final versions, I identified four authorship profiles: consolidated authorship (35%), developing authorship (35%), transformed authorship, where the AI reshaped the student's voice (20%), and substituted authorship, where the AI replaced it (10%). One student signed her final email "[Tu Nombre]", the placeholder the AI had left. Another signed hers "Por, Inteligencia artificial." On the other end, a student rejected a suggestion because "quiero que mi carta suene como yo."
Finding 3: The prompt predicts the outcome. The way students asked the AI for help was the most reliable predictor of whether their voice survived. Students who asked for alternatives while protecting their content kept control of their text. Students who typed the topic or asked the AI to "make it better" usually ended up with a text that was no longer theirs.
Finding 4: Reflection was uneven. Only 20% of students showed critical metalinguistic reflection (naming specific language features and justifying their choices). 40% reflected at a procedural level, and 40% gave brief or empty answers.
My conclusion: communicative teaching is still possible with AI, but it requires us to update how we assess, make metalinguistic reflection a central piece of evidence, and teach students explicitly how to write prompts.
The sequence itself turned out to be the most transferable part of the project, and you can use it tomorrow.
What works
1. AI after the draft, never before. Students need to produce language first. When the first attempt is theirs, AI feedback becomes something to evaluate, not something to copy.
2. AI as a feedback partner, not a ghostwriter. Ask the tool to point out problems instead of rewriting the text. The student does the fixing.
3. AI as a rehearsal partner for speaking. A low-stakes conversation with a chatbot can help anxious students practice before talking with a classmate. Rehearsal, not replacement.
4. AI for the teacher's prep. Generating leveled texts, extra examples of a structure, or differentiated versions of a reading saves hours, as long as you review everything before it reaches students.
5. Reflection as the non-negotiable step. If students can't explain why they accepted or rejected a suggestion, the learning didn't happen yet.
6. Teach prompting explicitly. Model the difference between "Make my text better" and "Give me two alternative ways to express this idea in a formal tone, without changing the content." The second prompt keeps the student as the author.
Image: Alabama Extension (CC0), via Wikimedia Commons
What doesn't work
- AI before students write. It anchors them to the machine's ideas and vocabulary, and their own voice never shows up.
- Accepting the AI version as the final version. Polished text with no student decisions behind it tells you nothing about what they can do.
- Replacing peer interaction. A chatbot can't negotiate meaning the way two students can. Communication is still a human act.
- Trusting AI blindly. AI tools make mistakes, mix regional varieties, and sometimes "correct" things that were already right. Students need to learn to question it.
- Using AI scores as grades. Automated feedback can support assessment, but the professional judgment stays with the teacher.
A 4-step routine you can use tomorrow
Draft → Ask → Decide → Reflect
- Draft: students write on paper or with AI turned off (10–15 minutes).
- Ask: they paste their text into the AI with a feedback-only prompt (see below).
- Decide: they highlight what they changed in one color and what they rejected in another.
- Reflect: they complete reflection frames in Spanish.
Reflection frames:
- Cambié ______ por ______ porque ______.
- La IA me sugirió ______, pero yo decidí ______ porque ______.
- Hoy aprendí que en español ______.
Image: Heidy Garcia, CC BY 4.0, via Wikimedia Commons
Prompts that keep students in charge
Student-facing prompt (novice/intermediate):
"Soy estudiante de español de nivel principiante. No reescribas mi texto. Dime solo tres errores que debo revisar y explícame cada uno en una frase simple."
Student-facing prompt (advanced/AP):
"Analiza mi texto. Señala dos lugares donde puedo usar un vocabulario más preciso y un lugar donde la conexión entre ideas no es clara. No me des la respuesta; hazme una pregunta que me ayude a corregirlo."
Teacher prompt:
"Create three versions of this short text for Spanish learners at Novice High, Intermediate Low, and Intermediate Mid levels. Keep the same story and cultural content, and list the key structures used in each version."
Differentiation
- Support: give students a checklist of what to look for in the AI feedback (verb endings, agreement, accents) and let them work in pairs during the Decide step.
- Challenge: students compare feedback from two different AI tools and argue which one was more useful.
- Heritage learner lane: focus on register and regional variety. Ask them to identify when the AI "corrects" an expression that is perfectly valid in their family's Spanish, and to explain the difference between standard and regional forms.
Quick check for teachers
Multiple choice: According to this approach, when should students first use AI in a writing task?
a) Before they start writing, to get ideas
b) After they write a first draft on their own
c) Only when the final version is ready to submit
(Answer: b)
Fill in the gap: The step that turns AI feedback into real learning is ______ reflection. (metalinguistic)
Final thought
AI isn't going away, and pretending it doesn't exist won't protect our students. But communicative teaching can stay communicative if we keep one principle in place: the student makes the decisions, and the student explains them.
The question isn't whether to use AI. It's whether our students are still the authors of their own language.
¿Y tú? How are you handling AI in your language classroom? Share your experience in the comments. 💬






