AI for Differentiated ESL Instruction: Tiered Materials

Teachers in mixed-ability ESL classrooms often describe differentiation as the goal they can see but never fully reach. The theory is clear: adjust the content, the process, or the product to meet students where they are. The execution is brutal. Writing four versions of a reading text, generating four sets of questions at different cognitive loads, and scaffolding four different writing prompts—all before Monday’s lesson—is simply more work than most schedules allow. AI tools have begun to change that calculation in ways worth understanding clearly, not as a promise of effortless lessons, but as a genuine shift in what one teacher can build in an hour.

What Differentiation Actually Demands of a Teacher

The most common misconception about differentiated instruction is that it means writing entirely separate lessons for every student. In practice, differentiation in ESL is more surgical. A teacher working with A2, B1, and B2 learners in the same class is typically modifying two variables: the linguistic complexity of the input students receive, and the cognitive demand of the output they produce. A well-differentiated lesson might use the same thematic content—a news story about urban farming, for example—while presenting it in a simplified version for A2 students, an unmodified version for B2 students, and something in between for B1. The comprehension tasks then vary not just in vocabulary level but in the type of thinking required: literal recall at the lower end, inference and synthesis at the higher end.

That is not easy to build, but it is structured. And structured problems are exactly what large language models handle well.

How AI Changes the Materials Pipeline

Before AI writing tools were widely available, tiering a reading text was a manual process: find or write the base text, then simplify or expand it sentence by sentence, checking vocabulary against a frequency list, shortening clauses, replacing passive constructions. A teacher might spend 40 minutes on a single tiered pair of texts. With a well-prompted AI tool, the same task takes five to eight minutes—including review and light editing.

The pipeline shift is not just about speed. It is about psychological feasibility. When differentiation is a 40-minute task per text, most teachers realistically tier one or two texts per week. When it takes eight minutes, differentiation becomes a default move rather than an occasional one.

silver laptop on woman's lap — silver laptop on woman’s lap

Tiering a Reading Text

The core AI move for text differentiation is simple: provide the base text and a clear proficiency target, then instruct the model to rewrite it for that level. The quality of the output depends almost entirely on the specificity of your instructions. A prompt that says “simplify this for beginners” produces inconsistent results. A prompt that specifies CEFR level, target vocabulary frequency, sentence length caps, and clause complexity produces text you can use in class with minimal editing.

A workable prompt structure looks like this: “Rewrite the following text for a CEFR A2 learner. Use only vocabulary from the Oxford 3000. Keep sentences under 15 words. Avoid relative clauses and passive constructions. Maintain the original meaning and all key facts.” That level of instruction constrains the model enough to produce consistent output across multiple texts. Once you have a prompt that works, save it as a template and reuse it every time you tier a new text.

A diverse group of young adults collaborating in a modern office setting, discussing a project.

Creating Task Variations from a Single Core Activity

Tiering the text is only half the work. The tasks students complete on that text also need to reflect different levels of demand. AI handles this variation efficiently once you understand the axis you are moving along.

For reading comprehension, the axis runs from literal to inferential to evaluative. A prompt asking the model to “generate three comprehension questions at the A2 level for this text, focusing on directly stated information” produces questions suitable for lower-level students. A second prompt requesting “three B2-level questions that require students to infer meaning, identify the author’s purpose, or evaluate the argument” produces a different set that challenges stronger readers on the same text. The teacher then selects questions from each set and assembles differentiated task sheets in minutes rather than drafting them from scratch.

For writing tasks, the axis typically runs from controlled to guided to open-ended. Lower-proficiency students benefit from sentence frames, vocabulary banks, and tightly scoped prompts. Higher-proficiency students work from open prompts with higher word count expectations and more complex rhetorical goals. AI can generate all three versions of a writing task when instructed explicitly: “Write three versions of a writing prompt based on this topic—one for A2 students using sentence frames and a 60-word minimum, one for B1 students using guiding questions and a 120-word minimum, one for B2 students as an open analytical prompt with a 200-word minimum.”

person holding on red pen while writing on book

A Prompting Framework That Delivers Consistent Results

Random prompting produces random output. Teachers who get reliable results from AI differentiation tools have usually developed a personal framework: a set of prompt templates tied to their curriculum and their students’ proficiency bands. Building that framework requires some upfront investment, but the payoff is materials that feel coherent across levels rather than like three separate lessons stapled together.

Anchoring Prompts to Proficiency Descriptors

The most reliable anchor for differentiation prompts is a recognized proficiency framework. CEFR descriptors—A1 through C2—give AI tools enough specificity to calibrate vocabulary, syntax, and task complexity consistently. The British Council and Cambridge Assessment both publish can-do descriptor documents that you can paste directly into your prompts as reference material. When an AI tool has a concrete description of what a B1 speaker can do—follow the main points in a clear, standard conversation about familiar topics; produce connected text with basic coherence and linking—it generates more usable output than when it works from a vague label like “intermediate.”

Iterating Without Starting Over

One advantage of AI-assisted materials building that teachers often underestimate is the ease of iteration. In a traditional workflow, revising a text or task set means returning to the original document and reworking it manually. With AI, iteration is a conversation. If the first draft of a simplified text is still too lexically dense for your A2 students, you instruct the model to revise specifically the vocabulary—not to regenerate the entire text. “The third and fifth sentences use too many polysyllabic words. Simplify those while keeping the rest unchanged.” This targeted revision approach preserves the parts of the output that are already working and reduces the time spent on refinement significantly.

Stand Up Tri-Fold Brochure Mockups, Download the PSD version of this mockup for free from https://mockupfree.net

Where Teachers Go Wrong With AI Differentiation

The most common failure mode is over-relying on AI output without review. AI tools occasionally introduce errors in simplified texts—deleting nuance that changes meaning, or oversimplifying a concept in a way that misrepresents the original. Every tiered text needs a read-through before it reaches students. A quick check for factual accuracy, natural phrasing, and level-appropriate vocabulary takes two to three minutes and catches most problems.

A second failure mode is ignoring the affective dimension of differentiation. AI can tier a text linguistically, but it cannot tell you whether your A2 students will find the topic engaging at their level, or whether the simplified version accidentally sounds condescending. Teacher judgment on tone, relevance, and student motivation remains essential. AI handles the structural labor; the teacher handles the human layer.

Finally, some teachers use AI differentiation as a reason to stop building shared classroom moments. Differentiation is not segregation by ability. Students at different proficiency levels benefit from shared tasks, partner work, and discussions that cut across levels—structured so that each student can contribute meaningfully. AI-differentiated materials should feed into a lesson design that brings the class back together, not fragment the group into isolated ability tracks that never intersect.

Building a Sustainable AI-Assisted Differentiation Practice

The teachers who sustain AI-assisted differentiation over time are not the ones who build new materials from scratch every lesson. They are the ones who build a library. Each time a prompt template produces good output, they save it. Each time a differentiated text works well in class, it goes into a shared folder organized by topic, level, and skill. Over a semester, this library becomes a resource that dramatically reduces future preparation time—not because the teacher stopped thinking, but because the structural work was captured and reused.

If you are starting this practice, the most useful first step is not to differentiate everything at once. Choose one text-based activity per week. Use AI to tier it across your class’s proficiency bands. Teach it, note what worked and what needed adjustment, and refine your prompt template accordingly. Within four to six weeks, you will have a prompting workflow that is both fast and reliable—and a growing library of materials your students have already tested in real conditions.

AI for differentiated instruction in ESL is not a shortcut that replaces pedagogical skill. It is a force multiplier for teachers who already understand what differentiation requires. The theory of differentiation has not changed. The logistics of building differentiated materials have changed substantially—and that difference is worth taking seriously.