AI for language teachers: what to hand over to the machine, and what to keep
AI for language teachers: what to delegate (lesson prep, exercise variants, student practice) and what to keep (goals, explanations, marking).
The rule that works for AI for language teachers is short: hand the machine whatever is repetitive and easy to check, such as a first draft of an activity, variants of an exercise, a text adapted to the group's level and the learner's speaking practice between lessons. Keep what decides whether the learner actually learns: each learner's goal, the explanation when they get stuck, the mark and the relationship that keeps them coming back. AI suggests; the teacher decides and checks.
It is Sunday evening and Tuesday's lesson is not ready. The listening task is missing, the new learner who joined mid-term needs an easier version of the reading text, and last week's essays are still in a pile. A colleague swears that AI can do all of it. It can do some of it. The real question is which part, and what you lose when you hand over the wrong one.
This article draws that line from what the research shows, with no list of apps. For the wider picture, see our guide to how AI is changing language learning.
AI for language teachers: where it helps and where it gets in the way
Use is already widespread, though uneven. In TALIS 2024, the OECD's international survey of teachers, 36% of teachers on average across OECD countries say they have used AI in their work. The figure is 56% in Brazil, according to the OECD country note on Brazil, and 14% in France, according to the country note on France. In both countries the least frequent use is assessing or marking student work, at 36% in Brazil and 24% in France.
One caveat before drawing conclusions: TALIS surveys teachers in mainstream schools, not language teachers specifically, and not tutors or language school staff. For English teaching, the closest picture comes from a British Council survey of 1,348 English language teachers in 118 countries and territories (British Council, 2024). The tasks they used AI for appear in this order: creating materials, helping students practise, creating lesson plans, correcting students' English, grading or assessing, and finally administrative tasks. The survey does not say how the teachers were selected, so the result is an indication rather than a picture of all teachers; even so, the pattern matches the OECD's.
So teachers have already drawn a line on their own: AI goes far more into preparation than into marking. It gets in the way when a teacher starts signing off work they have not read, or when a tool settles a question that requires knowing the learner.
What to hand over to AI
The best evidence on time comes from a randomised controlled trial by the Education Endowment Foundation (EEF) in England, independently evaluated by the NFER. 259 teachers in 68 secondary schools took part, and 129 of them used ChatGPT with an implementation guide to prepare Year 7 and 8 science lessons, according to the EEF's announcement. The result was a 31% cut in planning time: 25.3 minutes saved a week, bringing planning down to 56.2 minutes against 81.5 in the group asked not to use generative AI, the same announcement reports.
Two details matter. These were science teachers: as far as we could verify, there is no equivalent trial with language teachers. And the gain came from specific tasks, such as creating questions and quizzes, generating activity ideas and tailoring existing materials to particular groups of pupils, most often for one activity per lesson.
Quality is the open question. An independent panel of teachers saw no noticeable difference between the two groups' materials, but the EEF itself urges caution, because the sample of materials reviewed was limited. In other words, AI speeds up the draft, and quality still depends on someone checking it.
The first draft of your preparation
This is where AI tools for language teachers give back the most time: the first draft of an activity, a few variants of the same exercise for different levels, a text rewritten for another level.
An illustrative example. You have a B2 article on remote working that goes down well with your morning group, and you want to use it with an A2 learner. AI rewrites it with shorter sentences, more frequent vocabulary and only the present and past simple. Your job is to check three things: that no word above the level has slipped through, that the grammar is correct and that the text still serves the aim of the lesson. The draft arrives quickly; the decision to use it is yours.
Learner practice between lessons
What happens between lessons matters as much as the lesson. A review of 43 studies carried out for the British Council found that AI tools can help learners practise English outside class and can lessen their fear of speaking. The same review is candid about the limits: more research is needed to see whether these benefits last without continued use of the tool.
That changes your role. Instead of setting a list of exercises and hoping they get done, you decide what the learner should work on (meeting vocabulary, a job interview, the present perfect) and you check the result in the next lesson. With Alyx, for example, learners practise with their own teacher's AI Teacher, created with the teacher's face and voice from a photo and a short audio recording. It holds a conversation with the learner and corrects pronunciation, grammar and vocabulary in real time, without interrupting the conversation.
The task that probably weighs most was left out of the trial
Marking probably weighs more than lesson planning, and it was left out of the trial. Ben Styles of the NFER points out in the EEF's announcement that AI was reserved for specific activities within lesson planning, and that marking and administration probably represent a greater burden. We found no equivalent trial on marking. If AI helps you mark, use it to flag errors and suggest comments, never to decide the grade.
| Task | Hand it to AI? | What stays with you |
|---|---|---|
| Draft activity or quiz | Yes | Checking level, accuracy and aim |
| Variants of an exercise | Yes | Choosing which goes to which learner |
| Text adapted to a level | Yes | Checking vocabulary and grammar |
| Speaking practice between lessons | Yes, on a topic you set | Listening to the learner and adjusting the next lesson |
| Feedback on a piece of writing | Only as a suggestion | Rewriting and signing off the comment |
| Marks and level changes | No | The whole decision |
| Learner goals and the relationship | No | All of it |
What never to hand over
Four things are beyond any tool, and all of them decide whether the learner learns.
Each learner's goal
Why this learner is studying the language, by when, and where they will use it: a job interview, a family trip, a weekly call with an overseas team. AI works with whatever you describe to it; finding out the goal is your job, in conversation. UNESCO's AI competency framework for teachers sets out 15 competencies across five dimensions and starts with a human-centred mindset (UNESCO, 2024). The same text describes how AI has turned the teacher-student relationship into a teacher-AI-student dynamic, with the teacher still at one end.
The explanation when a learner gets stuck
Many English teachers will recognise this one. For the third time, a German-speaking learner writes "I have seen the film yesterday". AI corrects it to "I saw the film yesterday", and corrects it again the following week. The error comes back. You ask what they meant and find the source: they are mapping the German perfect ("Ich habe den Film gestern gesehen") onto the English present perfect, without knowing that a finished time in the past calls for the past simple in English. A short explanation of the difference between "I have seen that film" (experience, no time given) and "I saw the film yesterday" (a specific time) settles what the automatic corrections never did.
AI corrects the sentence. The teacher works out why the learner got it wrong.
Marks and decisions about the learner
Giving a mark, moving a learner up a level, saying whether they are ready for the exam: that responsibility is yours. The EU AI Act classes as high-risk the AI systems intended to evaluate learning outcomes in education and vocational training institutions (Annex III), and requires high-risk systems to be designed so that they can be effectively overseen by people while in use (Article 14), as set out in Regulation (EU) 2024/1689. Outside the EU, it is still a useful benchmark.
In practice, AI can suggest a correction or point out the errors that keep recurring in a piece of writing. The mark, and the comment the learner will read, go through you.
The relationship
It is the teacher who notices that a learner arrived quiet, has missed two weeks in a row or lost heart after a bad test, and who adjusts the lesson to bring them back. That work is what keeps a learner studying when motivation dips.
A simple method: AI suggests, the teacher checks
Many English language teachers feel they have not had enough training in how to use AI, according to the same British Council survey. This is where AI literacy for language teachers starts in practice. Until proper training arrives, a short checking routine does part of the job. Before anything AI produces reaches a learner, check:
- Level. Do the vocabulary and structures match the learner's CEFR level? One C1 word in an A2 text is enough to stall the reading.
- Accuracy. Are the grammar and spelling right? AI makes mistakes too, and makes them confidently.
- Aim. Does the activity train what this learner needs now, or does it just look like a good activity?
- Learner data. Have you pasted a name, a recording or an identifiable piece of work into an open tool? Don't. Use initials or anonymised texts.
- Decision. What goes to the learner, what you rewrite, what you drop. That choice is yours, and it makes clear who answers for the material.
Writing good requests to AI helps as well, and prompts deserve a guide of their own; the checking applies whatever you ask.
For schools: agree the rule as a team
In a language school, the same line needs to be written down, so that each teacher does not invent their own. Three decisions are enough to start:
- Who uses what. Which tasks staff may use AI for (preparation, variants, learner practice) and with which tools.
- What never goes out unchecked. Marks, level changes, feedback on written work and any message to a learner, a parent or a client company.
- Where the data goes. Which learner data may enter a tool, where it is hosted and how long it is kept.
The third question is the one most often forgotten. At Alyx, for example, data is hosted and processed in France in line with the GDPR, speaking session recordings are deleted automatically after 30 days, and teachers see the performance report, not the learner's audio (details on our AI and data page). Ask any supplier the same question before you sign.
Where to start
Pick a single task this week, ideally the one you repeat most: variants of the same exercise or adapting a text. Ask AI for the draft, run the checklist and note how long it took, checking included. After a few weeks you will know, from your own figures, where artificial intelligence saves you time and where it merely moves the work around.
Keep out of the machine what brings a learner back the following week: their goal, your explanation, your marking, your attention. To see how practice between lessons can follow your own method, take a look at the Alyx platform for language schools and teachers.
Frequently asked questions
Will AI replace language teachers?
The experts consulted by the British Council think not: the majority view is that AI will not replace the need for human teachers any time soon, and may never. What changes is the division of labour. AI takes on drafts and practice; the teacher keeps the learner's goals, the explanations, the marking and the relationship.
Where does AI help language teachers most?
In lesson preparation and in learner practice. In a trial in England, teachers used it to create questions and quizzes, generate activity ideas and tailor materials to groups of pupils. A British Council review of 43 studies found that AI tools help learners practise English outside class and ease their fear of speaking, though it is not yet known whether the effect lasts.
How much time can a teacher save with AI?
The only controlled trial we found, run by the Education Endowment Foundation, measured a 31% cut in lesson planning time, about 25 minutes a week, for science teachers in England using ChatGPT with a guide. We found no equivalent measurement for language teachers, so treat unsourced promises of hours saved each week with suspicion.
Can I use AI to mark and grade students' work?
Use it to flag errors and suggest comments, but the mark and any decision about the learner should be yours. Teachers already use AI least for this: in TALIS 2024, assessing or marking student work is the least frequent use of AI in both Brazil (36%) and France (24%). Under the EU AI Act, systems that evaluate learning outcomes are high-risk and must allow effective human oversight.
Are there free AI tools for language teachers?
Free tools can handle the first draft of lesson preparation: activities, exercise variants, texts adapted to a level. The criterion that matters most is what happens to your learners' data. Never paste a name, a recording or an identifiable piece of work into an open tool, and check everything before it reaches the classroom.
How should I check AI output before using it in class?
Run it through five filters: the CEFR level of the vocabulary and structures, grammatical accuracy, fit with the learner's goal, the absence of personal data, and your final decision on what goes out, what you rewrite and what you drop. AI suggests, the teacher checks.
