AI for Differentiated Instruction in ESL Teaching

Teaching a mixed-proficiency ESL class means running several lessons at once. The student who has mastered reported speech sits beside a classmate still building simple sentence confidence. Differentiated instruction has long been the professional response to this reality, but the preparation time it demands—separate texts, scaffolded tasks, leveled rubrics—has made full implementation feel impractical for most classroom teachers. AI tools are beginning to change that equation by compressing what used to take hours of extra planning into minutes, giving teachers the bandwidth to actually differentiate rather than just intend to.

This guide walks through how to put AI to work at each stage of the differentiation process: content creation, task design, speaking activity adaptation, and formative assessment. It is not a survey of apps to download. It is a practical look at where AI changes the preparation load in ways that translate directly into classroom impact.

What Differentiated Instruction Means in an ESL Context

Differentiated instruction, as developed by Carol Ann Tomlinson, is not about giving easier work to struggling students. It is about adjusting how content is delivered, how students process it, and how they demonstrate understanding—so that every learner is working at a productive level of challenge. In an ESL classroom, language proficiency creates an additional layer on top of whatever cognitive variation already exists. A student may have sophisticated analytical thinking but lack the English vocabulary to express it. Another may have strong oral fluency but struggle with academic writing conventions.

The three core levers in Tomlinson’s framework are content (what students learn), process (how they engage with it), and product (how they demonstrate mastery). In language teaching, the Common European Framework of Reference provides a useful reference grid: an A1 learner and a B2 learner in the same classroom are not just at different points on a linear scale—they are doing qualitatively different cognitive work with the language, which is why a single lesson plan rarely serves both well. AI tools have the most immediate impact on the content and process dimensions, where the volume of material creation is highest.

Using AI to Differentiate Instructional Content

The most time-consuming part of differentiation is creating multiple versions of the same material. A teacher who wants beginner, intermediate, and advanced versions of a reading passage on climate change traditionally has to write all three from scratch or search for pre-existing texts that match the topic and the level—a search that rarely delivers exactly what is needed. AI changes this fundamentally by treating text-level adaptation as a generation task rather than a search task.

Generating Level-Appropriate Reading Texts

With a tool like Claude, ChatGPT, or Gemini, a teacher can paste in a source article and prompt the AI to produce simplified and extended versions simultaneously. A useful prompt looks like this: “Rewrite this article at three reading levels. The beginner version should use vocabulary from the first 1,000 most common English words, short sentences, and run to no more than 200 words. The intermediate version should introduce content-specific vocabulary with brief in-text glosses. The advanced version should retain the original complexity and add one extension paragraph with a critical thinking question.” The result is three usable drafts in under two minutes, which the teacher then refines.

That refinement step matters and should not be skipped. AI-generated reading texts frequently oversimplify in ways that strip meaningful context, or they include vocabulary mismatches at the supposed beginner level. Teachers who treat AI output as a first draft rather than a finished product consistently report better classroom results than those who use it unchanged.

Adapting Grammar Tasks and Writing Prompts

Writing prompts are another high-volume area. A single composition topic can be scaffolded across proficiency bands with relative ease if you know what to ask for. A beginner prompt might frame the task with sentence starters and a strict ceiling of 80–100 words. An intermediate prompt removes the starters but provides an outline structure. An advanced prompt asks for a thesis-driven response with a counterargument. AI can generate all three versions from a brief description of the topic and the learning objective, dramatically reducing the time spent building writing scaffolds from scratch.

The same logic applies to grammar exercises. Rather than hunting for three different workbooks that happen to cover the same target structure at different levels, a teacher can ask an AI to generate a controlled practice exercise (fill-in-the-blank with a word bank), a guided exercise (sentence transformation with an example), and a free-production exercise (a short writing task that elicits the same structure naturally). All three can share the same thematic topic, which maintains content coherence across proficiency tiers.

scrabble, scrabble pieces, lettering, letters, wood, scrabble tiles, white background, words, type, typography, design, layou

Adapting Speaking Activities Across Proficiency Levels

Speaking differentiation is harder to systematize than reading or writing because it unfolds in real time. The practical AI application here is in preparation: using AI to generate role-play scenarios, discussion question banks, and speaking frame templates before the lesson rather than improvising differentiation on the spot—a strategy that typically benefits only the students who speak up fastest.

A paired role-play on negotiating a schedule, for example, can be scaffolded for beginners with a script that has blanks to fill in, for intermediates with a prompt card listing the key functional phrases they should try to use, and for advanced learners with only a situation description and an instruction to negotiate without prior preparation. AI can produce all three scaffolds from a brief description of the communicative goal, and the teacher distributes them digitally or in print before the activity begins.

Classroom logistics shape outcomes here. Teachers who group learners homogeneously for task-level activities but heterogeneously for peer-feedback activities tend to see better results from differentiation: students practice at their own productive level during the task, then share results across levels. This creates natural comprehensible input for lower-proficiency learners and genuine communicative pressure for higher-proficiency ones.

gray and white click pen on white printer paper

Differentiated Assessment and Feedback

Assessment is where differentiation most often breaks down. Teachers either hold all students to the same rubric—which can be discouraging for beginners and insufficiently challenging for advanced learners—or they attempt to create separate rubrics for each level, which is unsustainable across a full teaching load. AI offers a practical path through this tension.

Rubric Scaffolding and Focused Formative Feedback

AI makes tiered rubric creation feasible in a way it simply was not before. A prompt like “Write a speaking rubric for a 90-second self-introduction. Create three versions: one for A1–A2 learners focused on intelligibility and basic sentence completion, one for B1–B2 focused on vocabulary range and cohesion, and one for C1–C2 that adds criteria for naturalness, idiomatic use, and spontaneous elaboration” returns a working rubric set in seconds. The teacher reviews it for alignment with their specific context, adjusts the descriptors, and it is ready to use.

Formative feedback is another area where AI extends a teacher’s capacity without replacing professional judgment. After a writing task, a teacher can paste a student’s paragraph into an AI tool and prompt it for feedback pitched at a specific level: “Give feedback on this A2 learner’s paragraph. Focus only on subject-verb agreement and past tense formation. Do not address vocabulary or complex syntax yet.” Focused, level-appropriate feedback is consistently more useful for lower-proficiency learners than comprehensive correction, which can overwhelm students already working at the edge of their linguistic range.

Building a Sustainable Weekly Workflow

Teachers who sustain differentiation over a full semester tend to have a weekly preparation routine rather than trying to differentiate every lesson from scratch. A practical rhythm looks like this: at the start of a unit, use AI to generate leveled versions of core reading texts and a bank of scaffolded writing prompts. Mid-unit, run speaking activities using AI-generated prompt cards for each proficiency tier. At the end of the unit, apply tiered rubrics and use AI to generate focused, level-specific written feedback.

This approach does not mean every lesson is fully differentiated. In practice, that would be unsustainable. The goal is for structural differentiation to be built into the unit plan once—and for AI to handle the volume of material generation—so the teacher can focus on observation, facilitation, and the relational work that no tool can replace. When differentiation is planned into units rather than improvised per class, it is far more likely to persist across an entire course.

Students walk with a red flag in a park.

What AI Cannot Do—and Why That Matters

A clear-eyed picture of AI’s limits in differentiation is as important as understanding its possibilities. AI does not know your students. It produces plausible, generic output—not material calibrated to the specific gaps of the individual learner you have observed over weeks. The diagnostic work—noticing that one student consistently avoids complex syntax to mask a grammar gap, or that another understands content but shuts down under timed conditions—is irreplaceably human.

AI also cannot manage the relational dimension of differentiation. Students notice when they are given different materials, and how a teacher frames that matters enormously. Explaining the pedagogical rationale, positioning differentiated tasks as responsive rather than remedial, and reading the room when a student feels singled out—all of that requires teacher presence and the trust built in a shared classroom over time.

Used deliberately, AI for differentiated instruction in ESL classrooms is a prep-time multiplier that frees teachers to be more present and responsive with their learners. Used carelessly, it generates generic scaffolds that miss actual student needs while adding administrative overhead without improving learning outcomes. The difference lies in how seriously a teacher engages with AI output—and how much professional judgment shapes what happens next in the classroom.