Personalised Learning With AI Without Lowering Expectations

A pupil may understand an idea but struggle with the language used to explain it. Another may need practice with a missing prerequisite before tackling the class task. Both need a useful response from their teacher, and those responses may look quite different.

AI can propose a simpler explanation, a new example or a different sequence of practice. Before using it, a teacher needs to decide what the change is meant to achieve. If the original task asks pupils to justify a conclusion, the adapted version should still give them that thinking to do.

When planning personalised learning with AI, decide which part of the task needs to change and what would show that the change helped.

Start with the goal and the barrier

Personalisation can mean many things. Here, it means adjusting how a pupil approaches learning in response to an observed need. It does not require a different curriculum for every pupil or an AI account for every lesson.

Begin with the knowledge, explanation or skill you want to see. Then look at what is getting in the way. A vocabulary difficulty calls for a different response from a misconception. A pupil who cannot access the material needs a different adjustment from one who can complete the task but cannot explain the reasoning.

An AI-generated ability label cannot settle those questions. The teacher needs evidence from the pupil’s work, responses or existing support arrangements. UNESCO’s guidance on generative AI in education places human agency and pedagogical validation among the conditions for educational use. In this decision, that means the tool can suggest a change while the teacher judges its purpose and suitability.

It also helps to be clear about the technology. An adaptive practice system may select activities from a sequence using pupil responses. A generative AI tool can produce new explanations or questions. Evidence about one does not establish the effectiveness of the other.

The same adjustment can have different consequences depending on the goal. Audio might help a pupil examine historical evidence, while reading the words independently may itself be the goal in a decoding task. CAST’s guidance on goals, challenge and support offers a useful starting point for considering different means of reaching a challenging goal.

Existing individual goals and necessary accommodations still apply. A shared class goal does not remove the school’s responsibility to make appropriate provision for particular pupils.

Five ways to adapt the route

Before asking AI to make a task “easier”, name the part you want to change. The comparison below offers five ways to think about that choice. It is a planning aid for this article, rather than a tested model of personalisation.

More than one adjustment may be useful. A pupil might need vocabulary support and a slower introduction to an unfamiliar concept. The combination should respond to what the teacher has observed.

Sometimes simpler work is the right preparation. If equivalent ratios are insecure, a short return to them may help a pupil approach proportional reasoning. Make the connection explicit and decide when to revisit the target problem. Repeatedly assigning single-step arithmetic while reporting the original goal as met would conceal a change in expectations.

Once you have chosen an adjustment, review the resulting AI-generated material for accuracy, accessibility and classroom fit. A sensible adaptation choice still needs usable material.

What the evidence supports

There is evidence that particular forms of technology-supported, targeted learning can help. The details of the teaching arrangement matter.

The Mindspark evaluation in Delhi found improved mathematics and Hindi attainment in a supplementary programme for pupils mainly in grades 6 to 9. The programme combined adaptive software with small-group teaching. Its results therefore concern that combination of provision; they do not show what a general-purpose chatbot would achieve on its own.

In Maine, a study of ASSISTments mathematics homework found improved attainment among grade 7 pupils. Hints, feedback and teacher reports were accompanied by teacher training. The training and classroom arrangements were part of the intervention, so the result should not be attributed to the software alone.

The Education Endowment Foundation’s review of individualised instruction describes limited evidence and substantial variation. Organising targeted work is part of the teaching challenge.

Generative AI adds another question: does better supported performance carry over when the pupil works independently? In Bastani and colleagues’ secondary mathematics study, conducted in one Turkish high school, an unguarded GPT-4 tutor improved practice performance but reduced later unassisted examination performance relative to the control group. A tutor designed with safeguards largely removed that harm, without producing a positive examination effect.

That study cannot establish what every AI tool will do in every classroom. It does make polished work or successful supported practice an insufficient basis for assuming learning. The OECD’s Digital Education Outlook 2026 similarly draws attention to the distinction between task performance and learning. A teacher’s review needs to return to the intended knowledge or reasoning.

What this could look like in class

Consider these three hypothetical situations. They illustrate the choice; they are not reported school case studies or evidence of improved attainment.

History: keep the comparison with the pupil

The class is comparing why two sources describe an event differently. One pupil finds unfamiliar vocabulary difficult to follow.

A glossary or teacher-checked audio version could give the pupil access to both sources. The pupil would still need to select evidence and explain a difference between the accounts. If AI supplies the completed comparison and the pupil matches its conclusions to the sources, much of the intended reasoning has already been done.

At the review point, listen to or read the pupil’s explanation in an accessible response format. The relevant question is whether they can use evidence to justify the difference.

Mathematics: make prerequisite practice lead somewhere

A pupil is working towards solving and justifying a proportional-reasoning problem but is unsure about equivalent ratios.

Brief prerequisite practice could address that gap before the pupil returns to the target problem. AI might propose examples, which the teacher would check and select. On returning to the problem, the pupil needs to choose and explain an appropriate relationship.

A correct answer to a familiar arithmetic question would not, by itself, show that the pupil had reached that goal.

Science: leave the causal explanation open

Pupils need to explain the effect of changing one variable. A checked visual representation and vocabulary support could help a pupil understand the situation.

The pupil should still make a prediction and explain the causal connection. If AI has written the explanation and left only the variable’s name blank, filling that space reveals little about the reasoning the teacher wanted to see.

The support can stay in place while the teacher checks which part of the explanation the pupil can supply.

Use a short learning-goal test

For one adaptation you are considering, work through five questions:

1. What is the learning? Name what the pupil needs to know, explain, justify or perform.
2. What is the barrier? Identify the evidence behind your choice of support.
3. What will change? Describe what the pupil, teacher and tool will each do.
4. Who will do the important thinking? Check whether the adapted task could be completed without the knowledge or reasoning you intended.
5. When will you review it? Decide what you will look for and when you will reconsider the support.

These questions are a proposed planning aid, not a validated assessment instrument. If the adaptation removes the target thinking, revise it. If a different goal is appropriate, make that decision explicit through the school’s academic and learning-support arrangements.

Reviewing support does not mean withdrawing every form of help. Temporary prompts may change as understanding develops; necessary access accommodations may remain. The purpose of the review is to learn whether the pupil can demonstrate the intended understanding under appropriate conditions.

Where the work leaves that unclear, choose a proportionate way to check understanding after AI-assisted work. Use what you learn to decide whether that adaptation needs to change.

Begin with one task

Take a task already in your plans. Write down the learning goal, the barrier you have observed and the adjustment you want to try. Then decide what pupil response would help you judge whether it is working.

That gives you something concrete to discuss with a colleague or learning-support team. You can explain why the work looks different, what the pupil is still expected to learn and what will inform your next decision.