AI for Differentiated Instruction in ESL: How to Teach Mixed-Level Classes

Walk into any ESL classroom and you find a problem that has plagued language teachers for decades: learners arrive with wildly different levels, learning styles, motivations, and goals. One student needs the past simple explained for the third time. Another finished the grammar drill in two minutes and is bored. Differentiated instruction is the well-documented answer — but until recently, doing it well meant prepping three versions of every worksheet, every reading, every speaking task. That math never worked for a teacher with five classes a day.

AI tools have changed the math. A workflow that used to take ninety minutes of evening prep now takes ninety seconds in the staff room. This guide walks through practical, teacher-tested ways to use AI for differentiated instruction in ESL, with real workflows, sample prompts, and a lesson plan you can adapt this week.

What Differentiated Instruction Actually Means in ESL

Carol Ann Tomlinson, who built the modern framework, defines differentiated instruction as adapting four classroom elements — content, process, product, and learning environment — to three student variables: readiness, interest, and learning profile. In an ESL context that translates to obvious daily decisions. The B1 student and the A2 student in the same class should not read identical texts. The reflective writer and the impulsive speaker should not perform the same output task. The learner who needs Korean cognates and the learner who needs Brazilian Portuguese cognates need different vocabulary scaffolds.

Differentiation is not the same as personalization, and it is not lowering standards for weaker students. It is changing the path each learner takes to the same target — a target you set as the teacher, every lesson, with a clear language objective.

Why AI Changes the Differentiation Game

For thirty years, the real obstacle to classroom differentiation was time. Producing three versions of one reading passage took most teachers longer than producing the original. Most of us compromised: we differentiated grouping and questioning live in class, but we used one text and one worksheet because that is what the photocopier and the planning period allowed.

Large language models flip that constraint. They are fast text generators that can rewrite, simplify, expand, translate, gloss, quiz, and grade. Differentiation now costs minutes, not hours. The teacher’s job shifts from producing materials to judging them, sequencing them, and deciding which learner gets what. That is a much better use of professional time.

The Three Dimensions AI Can Differentiate

Content

Same topic, different input difficulty. Ask AI to rewrite a B2 news article at A2, B1, and B2+ levels with controlled vocabulary and sentence length. Ask it to add a Korean or Spanish glossary for specific learners. Ask it to insert pre-teaching boxes for the bottom third of the class and remove them for fast finishers.

Process

Same goal, different activity. The visual learner gets a labeled diagram task. The auditory learner gets a transcript of a recorded version with comprehension stops. The kinesthetic learner gets an information-gap pair task. AI can produce all three from the same source text in one prompt.

Product

Same target language, different output. Weaker learners produce a guided dialogue with sentence frames. Mid learners produce an opinion paragraph. Advanced learners produce a short debate position with a counterargument. AI can generate the scaffolds and the rubrics for each in seconds.

Seven AI Workflows for Mixed-Level ESL Classes

1. Triple-leveled reading passages

Paste a B2 article. Prompt: “Rewrite this at CEFR A2, B1, and B2. Keep the same paragraph structure and key facts. A2 should use only present and past simple, the 1000 most common words, sentences under twelve words.” You now have three differentiated reading texts in under a minute.

2. Layered comprehension questions

Same text, three question sets. Bloom’s taxonomy maps neatly: A2 gets remembering and understanding questions (find, name, list). B1 gets applying and analyzing (compare, explain, give examples). B2 gets evaluating and creating (argue, defend, predict). Ask AI to write five questions at each level from your reading.

3. Individualized vocabulary lists

Paste your unit’s word list and a one-line learner profile for each student (“Yuki, JP, A2, struggles with phrasal verbs” / “Mateo, BR, B1, advanced reader but weak spelling”). Ask AI to produce a personal list of eight target words per learner with example sentences and an L1 cognate note where one exists.

4. Speaking prompts by readiness

A single discussion topic (“a memorable journey”) becomes three speaking tasks: a sentence-frame dialogue for A2 (“I went to ___. I felt ___. The best part was ___.”), an open question pair for B1, and a one-minute monologue with rebuttal for B2. AI generates all three formats and the frames in one prompt.

5. Auto-scaffolded grammar practice

For a single grammar point, ask AI for a controlled-to-free progression: ten gap-fills, then five sentence transformations, then a short guided writing. Tell it to mark the cut-points where you might branch — the bottom group stops after the gap-fills, the middle group goes to transformations, the top group writes.

6. Fast-finisher extension cards

Mixed-level classes break when the strongest learners finish first and disengage. Generate a “challenge card” file at the start of each term — five extension tasks pegged to your unit theme, each ten minutes long, each requiring extended writing or research. Print and laminate. AI gives you ten weeks of cards in twenty minutes.

7. Differentiated homework with AI tutors

Send students home with a customized practice prompt for a free AI chatbot. “Practice ordering food at a restaurant. Ask me to play the waiter. Correct me only when I make a mistake with the past simple.” Different learners get different prompts targeting their specific weakness, while the teacher gets a transcript or summary the next morning.

Five Prompts You Can Steal Today

The prompt is the lesson plan now. Better prompts produce better differentiation. Here are five that work in any class:

1. The level-set prompt: “Rewrite the following text at CEFR [A2/B1/B2]. Use only [X] sentence length. Keep all key facts. Mark new words in bold.”
2. The question-stack prompt: “Generate five comprehension questions for this text at each of these levels: literal, inferential, and evaluative.”
3. The error-list prompt: “Here are three writing samples from one student. List the three most frequent errors and produce a five-item targeted exercise for each.”
4. The role-play prompt: “Design a three-stage role play (controlled, semi-controlled, free) on the topic of [X] for a CEFR [Y] class. Include sentence frames for the first stage.”
5. The rubric prompt: “Write a single-page rubric for a [task] at CEFR [level]. Use four bands. Focus on [target language] and [communicative goal].”

A Sample Differentiated Lesson Plan

Here is a 60-minute lesson built end-to-end with AI for a mixed A2-B2 class on the topic of remote work.

• 0–5 min: Whole-class warmer. Show one photo, ask “What is this person doing? Where? Why?” Open level.
• 5–20 min: Reading. Three printed versions of the same article (A2, B1, B2). Students self-select or are assigned. Each version has a glossary box for new words.
• 20–35 min: Comprehension. Three layered question sheets. Pairs check answers within their level group.
• 35–50 min: Speaking. Three task cards. A2 pairs do a sentence-frame dialogue. B1 pairs do an opinion exchange. B2 trios run a mini-debate. The teacher rotates between groups.
• 50–60 min: Cooler. Whole-class. Each level shares one sentence they produced. Teacher gives one whole-class language note.

Total prep time with AI assistance: roughly fifteen minutes. Without AI, this lesson would have taken an evening to build, and most of us would have given up after the second rewrite.

Limits, Cautions, and Where AI Gets It Wrong

AI-generated ESL material still needs a teacher’s eye before it goes in front of learners. Three failure modes show up repeatedly:

• Vocabulary creep. When you ask for an A2 text, models often slip in a B2 word (“nevertheless,” “established,” “phenomenon”) by paragraph three. Always scan the output and replace.
• Cultural drift. Examples default to North American contexts. If you teach in Taiwan or Spain or the Gulf, regenerate with a culture-specific prompt or edit names, food, and places by hand.
• False confidence. AI invents facts. Never use AI-generated readings on a topic where factual accuracy matters (history, science, current events) without verifying details against a real source.

Treat AI like a fast junior teaching assistant: willing, productive, occasionally wrong. Your professional judgment is the quality control.

Getting Started This Week

If this is new to you, do not try to differentiate everything at once. Pick one lesson and one dimension. The easiest first win is content: take next week’s reading, paste it into ChatGPT or Claude, ask for a simpler version, and bring both versions to class. Let learners choose. Watch what happens. The students who normally check out will engage with the simpler text. The students who normally race ahead will pick the harder one. You will see, in the room, why this is worth the small extra effort.

From there, expand one workflow per month. By the end of a term you will have leveled readings, layered question banks, fast-finisher cards, and differentiated speaking tasks for every unit you teach. The prep load goes down. The learning goes up. That is the trade we have been waiting for.

Differentiation Is a Habit, Not a Tool

AI is the lever that makes daily differentiation realistic for the first time in our profession’s history. But the lever does not move on its own. The teacher still chooses what to differentiate, for whom, and toward what objective. The decisions are still yours. The drudgery is finally not.

Start small, keep the prompts you like, throw away the ones that produce sloppy output, and build a personal library over a term. By next year you will look back and wonder how anyone ever taught a mixed-level class without it.

Quellen

• Carol Ann Tomlinson’s differentiated instruction framework via ASCD.
• CEFR level descriptors and assessment guidance via Cambridge Englisch.
• ESL methodology and professional development via the TESOL International Association.
• Classroom-tested teacher resources and research via the Britischer Rat.

More Worksheets

PPP Lesson Plan: The Complete 2026 Guide for ESL TeachersESL Assessment: 12 Strategies That Work (2026 Guide)

Electric Vehicles | Elementary ESL Reading Worksheet PDF