AI Differentiation in the Classroom: A Teacher's Guide

Every class is a mixed-ability class. Even when pupils are setted by prior attainment, the spread of reading ages, processing speeds, prior knowledge and motivation within a single classroom demands constant adaptation. AI tools now make some of that adaptation faster, but they do not remove the need for a teacher who understands why one pupil needs a sentence starter while another needs a constraint removed.

Research on differentiation consistently shows that adapting instruction to pupils' current understanding improves outcomes (Tomlinson, 2001). The problem has never been the principle but the workload. Creating three versions of a resource, modifying questioning for different groups, and tracking which pupils need which scaffold takes hours that most teachers do not have. AI tools address the production bottleneck without solving the pedagogical challenge underneath.

‍

What AI Differentiation Looks Like

AI differentiation is not a separate pedagogical approach. It is the use of AI tools to accelerate the resource creation and adaptation that good differentiation already requires. The decisions remain with the teacher; the production shifts to the machine.

In practice, this means a Year 6 teacher preparing a reading comprehension lesson can paste the text into an AI tool and request three versions: one with simplified vocabulary and sentence starters for below-expected pupils, one at the expected standard with open questions, and one with inferential and evaluative questions for greater-depth pupils. The AI produces all three in under two minutes. The teacher then reviews each version, adjusts where needed, and decides which pupils receive which resource.

The critical distinction: the AI creates the resources. The teacher makes the assessment judgements about who needs what. These are different skills, and conflating them leads to poor differentiation regardless of the tool.

‍

Three Models of AI Differentiation

There are three practical ways to use AI for differentiation in a typical classroom, each with different demands on teacher time and different levels of AI capability.

Model	How It Works	Teacher Time	Best For
Resource tiering	AI generates 3 versions of the same activity at different levels	5 min to prompt + 10 min to review	Worksheets, reading tasks, homework
Scaffold generation	AI creates graduated support: sentence starters, worked examples, prompt cards	5 min to prompt + 5 min to review	Writing tasks, problem-solving, extended responses
Adaptive questioning	AI generates question sets at varying Bloom's levels from the same content	3 min to prompt + 5 min to review	Plenaries, retrieval practice, formative checks

Most teachers start with resource tiering because it maps directly onto existing practice. If you already create a "must, should, could" worksheet, AI simply does it faster. The scaffold and questioning models require more prompt-engineering skill but offer greater pedagogical return.

‍

Prompting AI for Differentiation

The quality of AI-differentiated resources depends entirely on the quality of your prompt. A generic instruction like "make this easier" produces resources that are shorter but not genuinely adapted. Effective differentiation prompts include five specific elements.

Element	Why It Matters	Example
Year group and subject	Calibrates vocabulary and complexity	"Year 8 history"
Learning objective	Keeps all versions focused on the same outcome	"Explain the causes of the English Civil War"
Number of tiers	Defines how many versions you need	"Three versions: support, core, extension"
What changes between tiers	Prevents AI from just making text shorter	"Support: sentence starters and word bank. Core: open questions. Extension: source evaluation."
Format requirements	Ensures output is classroom-ready	"Fit on one A4 page per tier. Use a table for the word bank."

‍

A Worked Example: KS3 Science

Here is a complete prompt for differentiating a Year 9 biology task:

"You are a KS3 science teacher in a UK state school. Create three versions of a worksheet on 'How the heart pumps blood around the body' for Year 9 pupils.

Support version: Use a diagram with labels for pupils to match. Include a word bank (atria, ventricles, valves, oxygenated, deoxygenated). Sentences use no more than 15 words. Four questions testing recall only.

Core version: Pupils describe the journey of blood through the heart in their own words. Include a blank diagram for labelling. Six questions mixing recall and explanation.

Extension version: Pupils explain what happens when a heart valve fails and predict the consequences. Include a data table of resting vs exercising heart rates for analysis. Four questions requiring evaluation."

This prompt produces three genuinely different resources in under two minutes. The key is specifying what changes between tiers, not just the difficulty label.

‍

Differentiation by Subject

The application of AI differentiation varies across the curriculum because different subjects demand different types of adaptation. A Year 5 maths lesson and a Year 10 English lesson require fundamentally different approaches to scaffolding.

‍

English

AI is particularly effective at generating graduated writing scaffolds. For a Year 7 creative writing task, you can prompt the AI to produce: a story map with sentence starters for lower-attaining pupils, a structure strip with paragraph prompts for middle-attaining pupils, and a constraints-based challenge (write the story in exactly 250 words, or from a secondary character's perspective) for higher-attaining pupils. The same narrative content, three different entry points.

For reading comprehension, AI can generate questions at different depths from the same text. Lower-attaining pupils get retrieval questions ("Find two words that describe the forest"). Middle-attaining pupils get inference questions ("Why does the character hesitate?"). Higher-attaining pupils get evaluative questions ("How does the writer create a sense of unease? Support your answer with evidence from the text").

‍

Mathematics

Maths differentiation through AI works best at the task level. You can paste a set of core problems and ask the AI to generate a scaffolded version (with worked examples and part-completed solutions), a core version (the original problems), and an extension version (applying the same concept to unfamiliar contexts or multi-step problems). Tools like Diffit and TeacherMatic can do this automatically from a topic name.

The trap in maths differentiation is reducing the cognitive demand rather than providing access to it. A scaffolded version should still require pupils to think; it should remove unnecessary barriers (reading load, working memory demands) while preserving the mathematical reasoning. AI tools default to making problems easier. Prompt them to make problems more accessible instead.

‍

Science

Science differentiation benefits from AI's ability to generate multiple representations of the same concept. For a Year 8 lesson on photosynthesis, AI can produce: a labelled diagram with fill-in-the-gaps for lower-attaining pupils, a written explanation task with key vocabulary highlighted for middle-attaining pupils, and an experimental design challenge ("How would you test whether light intensity affects the rate of photosynthesis?") for higher-attaining pupils. All three address the same concept but through different cognitive demands.

‍

Humanities

In history, geography and RE, differentiation often centres on source material and question complexity. AI can rewrite a primary source at a lower reading age while preserving its key content, generate guided analysis questions for less confident pupils, and create evaluation frameworks for more able pupils. For a Year 9 geography lesson on climate change, the same data set can be accompanied by three different task sheets: one asking pupils to read the data (support), one asking them to identify trends (core), and one asking them to evaluate the reliability of the data source (extension).

‍

Primary

In primary classrooms, AI differentiation often focuses on reading level adaptation. A Year 3 class reading about the Great Fire of London can receive the same historical content at three reading ages. The AI adjusts sentence length, vocabulary complexity and the number of ideas per paragraph without changing the core historical facts. This allows mixed-ability guided reading where every pupil engages with the same topic at an appropriate level.

For maths, primary teachers use AI to generate concrete-pictorial-abstract progressions. The AI produces visual representations (pictorial stage) alongside abstract number sentences, allowing pupils at different stages of mathematical development to work on the same learning objective.

‍

AI Differentiation for SEND

Pupils with special educational needs and disabilities benefit significantly from AI-generated adapted resources, but the adaptation must be specific to the need. "Making it easier" is not differentiation for SEND; creating resources that account for specific processing requirements is.

SEND Need	AI Adaptation	Prompt Instruction
Dyslexia	Simplified sentence structure, larger text prompts, sans-serif font specification	"Rewrite using sentences under 12 words. Avoid dense paragraphs. Use bullet points where possible."
ADHD	Chunked tasks with clear checkpoints, visual progress markers	"Break this into 5 short tasks of 3-4 minutes each. Add a checkbox before each task."
Autism (ASD)	Explicit instructions, reduced ambiguity, predictable structure	"Use numbered steps. Avoid figurative language. State exactly what a successful answer looks like."
Speech and language	Visual supports, keyword highlighting, colourful semantics integration	"Add a visual cue beside each key term. Use colour coding: who (orange), what doing (yellow), what (green)."
Dyscalculia	Concrete examples before abstract, number line scaffolds, reduced number of problems	"Start each problem with a real-world context. Include a number line. Reduce to 5 problems instead of 10."
EAL	Bilingual glossaries, visual vocabulary, simplified instruction language	"Create a glossary of 10 key terms with simple English definitions and space for home language translation."

The 2024 DfE survey found that 57% of SEND teachers already use AI tools for creating adapted resources, making this the fastest-growing area of AI adoption in UK schools (DfE, 2024). The advantage is clear: a teaching assistant who previously spent 30 minutes adapting a single worksheet can now generate SEND-appropriate versions in minutes, freeing time for direct pupil support.

For a deeper look at AI tools matched to specific SEND needs, see our guide to AI in special education.

‍

Adaptive Learning Platforms

Beyond teacher-prompted differentiation, several platforms offer real-time adaptive learning where the AI adjusts difficulty based on pupil responses. These work differently from manual differentiation because the AI makes the adaptation decisions, not the teacher.

Platform	Subject Focus	How It Differentiates	Teacher Control
Century Tech	Maths, English, science	AI selects next question based on prior answers	Can set topic focus; AI controls difficulty
Seneca Learning	All GCSE/A-Level subjects	Spaced repetition with difficulty scaling	Assigns topics; AI schedules review timing
Sparx Maths	Mathematics	Personalised homework paths based on class teaching	Teacher sets topic; AI adjusts question difficulty
Tassomai	Science (primarily)	Daily quiz with algorithm-driven focus areas	Limited; AI determines revision priorities
Diffit	Cross-curricular reading	Adapts text reading level; generates tiered activities	Teacher chooses text and target levels

These platforms are useful for homework and independent practice where real-time teacher adjustment is not possible. During lessons, teacher-led differentiation using AI-generated resources remains more responsive because you can read the room in ways an algorithm cannot. A pupil's body language, their willingness to ask questions, and their interaction with peers all inform differentiation decisions that no platform can make.

‍

Avoiding Common AI Differentiation Mistakes

The speed of AI resource generation creates new risks that did not exist when differentiation was entirely manual.

1. Differentiating down, not across. The default AI behaviour when asked to "simplify" is to remove content. This produces a support version that covers less curriculum rather than providing a different route to the same learning. Always specify that all tiers must address the same learning objective. The support version should scaffold access to the learning, not reduce the learning itself.

2. Over-differentiating. Producing five tiers of a worksheet when two would suffice creates management overhead without pedagogical benefit. In most lessons, three versions (support, core, extension) are sufficient. More tiers means more decisions about who receives what, more resources to monitor, and more complexity to manage during the lesson.

3. Static grouping. Using AI to generate differentiated resources makes it tempting to assign pupils permanently to a tier. "Jamie always gets the support sheet." This contradicts what we know about the zone of proximal development (Vygotsky, 1978). Pupils should move between tiers based on their current understanding of the specific topic, not a fixed label. Use AI-generated resources flexibly, not as a tracking system.

4. Neglecting the extension group. AI tools are much better at simplifying than extending. When you prompt for an "extension" version, the AI often produces more of the same rather than something qualitatively different. For genuine extension, specify what higher-order thinking looks like: analysis, evaluation, creation, transfer to new contexts. Higher-order thinking requires deliberate prompt design.

5. Skipping the review. AI-generated differentiated resources are drafts, not finished products. A support version may use vocabulary that is still too complex. An extension version may include content beyond the curriculum. Five minutes of teacher review catches errors that would take 20 minutes to unpick during the lesson.

‍

AI Differentiation and Rosenshine's Principles

AI-assisted differentiation maps directly onto several of Rosenshine's principles of instruction (Rosenshine, 2012). Understanding these connections helps you use AI tools more effectively rather than treating them as separate from your existing pedagogy.

Principle 1 (Daily review): AI generates differentiated retrieval practice questions at three levels. Lower-attaining pupils recall key facts. Middle-attaining pupils explain connections. Higher-attaining pupils apply knowledge to new scenarios. All three groups engage with the same prior learning, but the cognitive demand scales appropriately.

Principle 6 (Check for understanding): AI creates tiered exit tickets that assess the same learning objective at different depths. This gives you diagnostic data on every pupil, not just the pupils who put their hands up.

Principle 10 (Weekly and monthly review): AI generates personalised revision materials based on which topics each pupil found most difficult. Rather than a generic revision sheet, each pupil receives a set of questions targeting their specific gaps. Platforms like Seneca and Carousel Learning do this automatically.

The principle is simple: AI handles the production of differentiated materials. Rosenshine's framework tells you when and how to deploy them. The combination of research-informed timing and AI-powered resource creation is more effective than either approach alone.

‍

A Practical Weekly Workflow

Here is a workflow that integrates AI differentiation into weekly planning without adding hours to your preparation time.

When	AI Task	Teacher Task
Sunday evening (15 min)	Generate tiered resources for 3 key lessons	Review outputs, adjust tier assignments based on last week's data
Before each lesson (5 min)	Generate differentiated starter questions	Decide which pupils need which tier today
During the lesson	N/A (resources already prepared)	Circulate, adjust scaffolding live, move pupils between tiers
After the lesson (5 min)	Auto-mark exit tickets	Note which pupils need tier adjustments for next lesson
Friday (10 min)	Generate personalised revision based on the week's data	Review revision materials, send home as weekend homework

Total additional AI time: approximately 40 minutes per week. Total time saved on manual resource creation: approximately 2-3 hours per week. The net gain is not just time but quality. Resources that would have been a single worksheet for the whole class become three purposeful versions that genuinely meet different needs.

‍

Getting Started: Your First Lesson

Choose one lesson with a class you know well. Select a task you would normally give to everyone as the same worksheet. Open ChatGPT, Claude, or Diffit and use this template:

"I am teaching [subject] to [year group]. The learning objective is [objective]. Create three versions of a [task type]:

Support: [specific scaffold type, e.g., sentence starters, word bank, part-completed answers]

Core: [the standard task with open questions]

Extension: [higher-order challenge, e.g., evaluate, compare, apply to new context]

All three versions must address the same learning objective. Format each to fit on one A4 page."

Review the output. Print the three versions. Distribute based on your assessment of where each pupil is with this specific topic. After the lesson, note what worked and what needed adjusting. That single lesson gives you enough information to decide whether AI differentiation is worth scaling across your teaching.

For a broader guide to AI tools in the classroom, see our hub article on AI for teachers. For assessment-focused differentiation, see AI marking and feedback. And for specific approaches to managing mixed-ability groups, see our complete guide to differentiation strategies.

‍

Further Reading