AI in English language teaching means using generative tools to draft, adapt and mark the material around a lesson while the teacher keeps every decision that depends on knowing the learners. Used well it removes hours of preparation each week. Used badly it produces confident, wrongly levelled English that nobody in the room can actually use.
Key takeaways
- AI is a drafting tool, not a teaching tool. It writes the materials. You still run the lesson.
- The reliable workflow is plan, prompt, inspect, adapt, teach. Skipping the inspect step is where almost every bad AI lesson comes from.
- Tools are strongest on repetitive production work: levelled texts, gap fills, rubrics, first pass feedback, slide outlines.
- They are weakest on anything needing knowledge of a specific class, cultural judgement or a decision about what to do next.
- Privacy is not optional. Student writing is personal data and it should not be pasted into a tool nobody has checked.
- Teachers who can direct and audit AI output are becoming more valuable, not less.
What AI in English language teaching actually means now
Two or three years ago this conversation was mostly speculation. It is not any more. Walk into a staffroom in Dublin, Seoul or Bogota and someone will be quietly running a reading text through a tool to get a B1 version before period three. The technology stopped being interesting and became infrastructure, which is usually the point at which it is worth writing a proper guide about it.
The useful definition is narrow. AI in English language teaching is the use of large language models and related tools to produce, adapt, level and assess the language material that surrounds a lesson. That includes lesson plans, reading passages, comprehension questions, dialogues, role cards, rubrics, vocabulary sets, feedback comments and test items. It does not include the lesson itself. A model can write you a beautiful task. It cannot notice that half the class went quiet when you set it.
That distinction sounds obvious written down. It is much easier to lose in practice, usually at about ten at night when the plan is due in the morning and the output looks good enough.
Where AI genuinely saves an English teacher time
Start with the honest list. These are the jobs where teachers consistently report real time savings rather than novelty.
Levelling and adapting texts. Taking an authentic article and producing an A2, B1 and B2 version of it is the single highest value use. It used to mean rewriting by hand or abandoning the text. Our free AI Materials Adaptor was built for exactly this job, and there is a fuller walkthrough in our guide to adapting teaching materials with AI.
First draft lesson plans. Not the finished plan. The skeleton: a warmer, a staging order, a production task. You then cut about a third of it, because models are relentlessly optimistic about how much fits into fifty minutes.
Practice material at volume. Gap fills, matching tasks, error correction exercises, additional examples of a structure your class keeps getting wrong. Tedious to write, quick to check.
First pass feedback on writing. A model is good at spotting patterns across a set of essays. It is much less good at deciding which of those patterns matters for this student this week.
Where AI is still poor at English teaching
The failure modes are consistent enough that you can plan around them.
Levelling is approximate. Ask for A2 and you will often get something closer to B1 with a few short sentences bolted on. Always read the output against the actual CEFR descriptors published by the Council of Europe rather than trusting the label the tool gives itself.
Invented facts remain a problem, and they are worse in ELT than people expect, because a reading text full of plausible but wrong information is very hard to spot when you are skimming. Anything factual in a text you hand out is your responsibility, not the model’s.
Cultural fit is weak. Models default to a broadly American classroom with broadly American reference points. If you are teaching in Kuwait, Vietnam or rural Poland, a fair amount of what comes back will need rewriting for reasons that have nothing to do with grammar.
And they flatter. Ask a model whether your lesson plan is good and it will tell you it is excellent. That is not feedback, it is politeness with a very large training budget.
“Highly recommended”
“I found the learning material easy to navigate and really high quality […]”
Stephanie Scanlan-Georgescu · August 2026 · verified review of a Level 5 course from The TEFL Institute, the accredited training provider behind tefl.ai · Read the full review · All our reviews
The workflow that actually holds up: plan, prompt, inspect, adapt, teach
Teachers who get good results are not using better tools. They are using the same tools in a fixed order.
Plan first, without the model. Decide the objective and how you will know it was met. Two minutes on paper. If you open the tool before you have done this, the model chooses your objective for you, and it will choose something generic.
Prompt with real context. Level, class size, first language, time available, what they did last week, what they keep getting wrong. The difference between a useless output and a usable one is almost entirely in how much context you gave.
Inspect before you adapt. Read for accuracy, level, cultural fit and length. Assume something is wrong, because something usually is.
Adapt in your own voice. Cut the padding, add the local example, fix the register. This is the step that turns a generated worksheet into your worksheet.
Teach, then note what happened. The feedback loop is you. If the task died in the room, that goes into the next prompt.
Our step by step guide to ESL lesson plans with AI walks the same loop with a worked example, and the free Lesson Plan Generator is a reasonable place to try it.
Choosing tools without ending up with nine subscriptions
The market is noisy and most of it overlaps. A sensible setup is one general assistant plus one or two specialist tools for jobs the general assistant does badly. Everything else can wait until you have a real problem it solves.
| Teaching job | What AI does well | What you must still check | Typical time saved per week |
|---|---|---|---|
| Levelling a reading text | Produces several CEFR versions from one source in under a minute | Actual level, invented facts, vocabulary load | Approximately 1 to 2 hours |
| Drafting a lesson plan | Gives a staging skeleton and task ideas fast | Timing, objective, whether the class can do it | Approximately 1 hour |
| Building practice exercises | Endless gap fills, matching tasks and extra examples | Answer keys, ambiguity, repetition | Approximately 45 minutes |
| Marking writing | Flags patterns across a set and drafts comments | Band judgement, tone, which errors to prioritise | Approximately 1 to 3 hours |
| Writing rubrics | Turns an objective into criteria and descriptors | Whether the criteria match what you taught | Approximately 30 minutes |
Time savings are typical ranges reported by teachers rather than measured averages. Your mileage will depend heavily on how much checking your context demands.
If you want a longer comparison, we keep one in best AI tools for English teachers, and every free tool we run is listed on the AI tool overview page.
What about students using AI?
They already are, and pretending otherwise makes you the last person in the room to find out. The productive position is not a ban. It is a clear statement of what is allowed, what is not, and what you will do about it.
Most teachers land somewhere around this: AI is fine for brainstorming, for checking a sentence you already wrote, for generating extra practice, and for looking things up. It is not fine for producing text you then submit as evidence of your own English. Learners accept that framing surprisingly readily, because it matches how they already think about calculators and dictionaries.
The harder question is what to do when you suspect a submission was generated. Detection tools are not reliable enough to accuse anyone, and they misfire most often on the exact group you least want to accuse, which is second language writers.
The rules, the privacy and the paperwork
If you teach in the EU or handle EU learners, this part is not optional. The EU AI Act, Regulation (EU) 2024/1689, has been phasing in since August 2024, and the AI literacy duty on organisations that deploy AI systems has applied since 2 February 2025. Education is listed as a high risk area in Annex III, though the substantive high risk obligations have since been deferred to December 2027.
Underneath that sits the older and more immediately relevant point: student writing is personal data. Under GDPR, pasting a named learner’s essay into a consumer chatbot that you have not checked is a processing decision, and it is one your school may not have authorised. The Irish Data Protection Commission publishes plain guidance worth ten minutes of your time.
The practical version is short. Strip names. Use tools your institution has approved. Do not upload anything about a child that you would not read aloud in the staffroom.
How are English teachers actually using AI?
We are running a survey of English teachers on what AI has changed in their work, what it has not, and where it gets things wrong. Twelve questions, about four minutes, no email address required.
The results will be published free on tefl.ai. There is very little independent data on this, so what teachers tell us here is what the report will say.
The skills that make a teacher harder to replace
The teachers doing best with this are not the most technical. They are the ones who are good at three things: giving precise context, spotting a wrong answer quickly, and knowing what they actually wanted before they asked for it. All three are teaching skills that existed long before any of this.
The classroom half of the job has, if anything, gone up in value. Reading a room, sequencing, correcting at the right moment, knowing when to leave an error alone: none of that is going anywhere. What has changed is that the preparation half no longer needs to eat your evenings.
Where to go next
This guide is the overview. The supporting pieces go deeper on individual jobs: grading English writing against CEFR levels, differentiated instruction with AI, using AI as a speaking partner, and the question of whether AI lesson planning counts as cheating.
Two of those supporting guides deal with the classroom rules rather than the tools. If you need a written rule your learners can actually follow, start with writing an AI policy for your English classroom. If the question is what to do when a submission does not look like a learner’s own English, read spotting AI written work in an English class, which covers why detection software is the wrong tool for the job.
Start with one job, not the whole workflow
Pick the single task that costs you the most time this week and try the loop above on that alone. Levelling a text is usually the best first choice, because the time saving is obvious and the checking is quick. Our free AI CEFR Writing Grader and the rest of the toolkit are free to use, and if you want a credential that says you can do this properly, the AI-Skilled Teacher Certificate covers the same ground with assessment attached.
