Could Every Student Have Their Own AI Revision Course?
Why should two students be given exactly the same revision sheet when their weaknesses are completely different?
Two students sit the same mathematics test. Both score 50%.
On paper, they appear to need roughly the same help. Give them another revision worksheet, explain the difficult questions and encourage them to practise.
But look more closely.
One student understands the mathematics but struggles to translate written questions into equations. The other can set up the equations perfectly well but makes mistakes when rearranging them.
Their marks are identical. Their learning needs are very different.
This is where AI could become particularly useful: helping a teacher build a revision course that changes as the student learns.
Imagine a course that keeps track of topics covered, mistakes made, hints needed and confidence reported. Each new task would be chosen using that evidence.
That is a much more interesting possibility than simply asking AI to produce another twenty questions.
From a Revision Folder to a Changing Learning Plan
A conventional revision course usually has a fixed structure:
Read the notes. Watch the explanation. Answer the questions. Check the answers. Move to the next topic.
There is nothing inherently wrong with that. A clear sequence can be extremely helpful.
However, students rarely develop at exactly the pace assumed by the course. Some need to revisit earlier knowledge. Others are ready for harder applications. Some can complete a question immediately after an explanation but struggle when it returns a week later.
An individual AI-assisted course could respond to these differences.
It might suggest:
- a short diagnostic task when the cause of a mistake is unclear;
- a worked example when a method needs explaining;
- an independent question to check whether the explanation helped;
- a harder application when the student is ready;
- a later review to check whether the learning has lasted.
The useful feature is the decision about what comes next.
Generating resources is only one part of the job.
The Student Profile: More Useful Than a Single Percentage
For this approach to work, the course needs a useful record of learning.
That record need not be an enormous database. A carefully maintained table could provide a starting point.
Here is an illustrative profile for a fictional GCSE mathematics student:
| Area | Evidence from recent work | Confidence | Possible next step |
|---|---|---|---|
| Solving linear equations | Solves simple equations independently; errors with brackets | Fairly confident | Check expansion before harder equations |
| Simultaneous equations | Solves a prepared pair; struggles to form equations from a situation | Low | Practise translating statements into equations |
| Fractions | Adds simple fractions; inconsistent with mixed numbers | Low | Short diagnostic task |
| Graphs | Plots accurately; confuses gradient with intercept | High | Explain and compare the two |
| Circle calculations | Uses the formula correctly once radius is identified | Moderate | Practise identifying radius and diameter |
The confidence column is particularly revealing.
A student who is confidently wrong needs different feedback from one who is correct but convinced they cannot do the subject.
However, confidence is self-reported information, not proof of understanding. It should sit alongside actual work.
I would also want to record whether an answer was produced independently, after a small hint, or while following a complete example.
Those are three different levels of support. A course should recognise the difference.
A Practical Example: What Is Going Wrong With Simultaneous Equations?
Suppose a student struggles with this question:
Two adult tickets and one child ticket cost £23. One adult ticket and two child tickets cost £19. Find the price of each ticket.
There are several possible obstacles.
The student might not understand what the letters represent. They might struggle to form the equations. They might know the method but make a subtraction error. They might find the wording overwhelming.
Giving them ten more questions of the same type could repeat the difficulty without explaining it.
A better next step is to investigate.
Step 1: Separate the Algebra From the Interpretation
Present the equations directly:
2a + c = 23
a + 2c = 19
Can the student solve them?
If they can, there is evidence that the main difficulty may lie in translating the situation into mathematics. That is a working hypothesis, rather than a final diagnosis.
If they cannot, check the underlying algebra.
Can they solve:
3x = 12?
Can they rearrange:
x + y = 9
to make y the subject?
Can they subtract one equation from another without changing signs incorrectly?
The next task should help distinguish between these possibilities.
Step 2: Give Practice That Addresses the Evidence
If interpretation appears to be the obstacle, ask:
An adult ticket costs a pounds and a child ticket costs c pounds. Write an expression for the cost of two adult tickets.
Then:
Write an expression for two adult tickets and one child ticket.
Then:
Their total cost is £23. Turn your expression into an equation.
The student is practising the part that caused the difficulty.
If algebra is the obstacle, work with simple prepared equations first, then return to the ticket problem.
Step 3: Return to the Complete Question
The original problem gives:
Adult ticket = £9
Child ticket = £5
The student should check both conditions:
2 × £9 + £5 = £23
£9 + 2 × £5 = £19
Then try a fresh question independently.
Successfully following an explanation is encouraging. Solving a new problem provides stronger evidence that the method is understood.
Step 4: Check Again Later
Revisit the skill in a later session, with different wording and numbers.
If the student can still use it independently, the profile can record stronger evidence of progress.
The course should respond to what the student can do today, while checking what they can still do later.
Personalisation Should Sometimes Make Work Harder
It is easy to imagine personalised learning as a system that keeps making questions easier.
That would be a serious limitation.
A useful course should also recognise when a student needs greater challenge.
For simultaneous equations, that might mean progressing towards:
- equations requiring multiplication before elimination;
- negative or fractional solutions;
- unfamiliar contexts;
- questions where the student must choose the method;
- explaining why someone else’s solution is incorrect.
The aim is to make success increasingly independent.
A student who has mastered the basics should not spend an entire revision session repeating them simply because those questions are easy to generate.
Equally, a student who struggles should not be permanently confined to elementary work. The course needs a route back towards the demands of the examination.
The Same Approach Could Help in Science
Consider a biology student who writes:
Antibiotics kill viruses.
A useful response should do more than replace that sentence with the correct one.
The next task could ask the student to distinguish between a bacterial disease and a viral disease. A further question could check whether they understand why antibiotics are unsuitable for treating a viral infection.
Later, the course could revisit the idea within a question about antibiotic resistance.
In physics, a wrong calculation might lead to separate checks on:
- selecting the equation;
- rearranging it;
- converting units;
- substituting values;
- interpreting the answer.
In chemistry, an incorrect concentration calculation might reveal a weakness in unit conversion rather than chemical understanding.
The same principle applies across subjects:
A wrong answer is the beginning of an investigation.
It does not automatically reveal its own cause.
Confidence, Independence and Knowing When You Need Help
One of the most valuable questions a revision course could ask is:
Before you check, how confident are you that your answer is correct?
Followed by:
What makes you think that?
Over time, students could compare their judgement with their performance.
Some might discover that they know more than they think. Others might realise that recognising a familiar question is different from being able to solve it.
This connects with metacognition: learning to plan, monitor and evaluate your own learning. The Education Endowment Foundation’s guidance emphasises explicitly teaching and modelling these processes within subject teaching. That supports the educational principle behind this approach; it does not establish that any particular AI course will deliver it successfully. EEF
An AI-assisted course could prompt that reflection.
A teacher still needs to help students make sense of it.
My Perspective as a Tutor: The Record Must Leave Room for the Person
After around forty years of teaching, I am very aware that a student’s performance cannot always be explained by their subject knowledge.
A student may understand something during a conversation but struggle to express it on paper. They may rush, misread a word, lose concentration or arrive tired.
A record saying “three incorrect answers” cannot explain all of that.
In my own teaching, I would want AI to help organise the evidence and prepare useful follow-up work. I would still want to talk to the student, look at their working and ask why they chose a particular approach.
Sometimes a question needs simplifying.
Sometimes it needs drawing.
Sometimes we need to put the worksheet down and demonstrate the idea.
That matters especially in science. A student may answer questions about an experiment yet still need practice setting up apparatus, making measurements or explaining an unexpected result.
An individual revision course should be able to recommend practical work, discussion and handwritten practice alongside screen-based activities.
The Course Must Remain Anchored to the Examination
Personalisation needs a destination.
For examination students, the course must take account of the qualification, specification, tier where relevant, and the skills assessed.
A course could become very good at practising a narrow weakness while overlooking an entire section of the syllabus.
I would therefore want two views of progress:
The close view: What does this student need next?
The whole-course view: What remains to be covered and revisited?
Students also need mixed questions where the topic and method are not announced in advance.
A worksheet headed “Simultaneous Equations” provides a substantial clue. An examination may expect the student to recognise the need for simultaneous equations themselves.
Eventually, practice must include choosing the approach as well as carrying it out.
AI Can Produce Convincing Mistakes
A personalised course is only useful if its questions, explanations and feedback are dependable.
AI can produce an incorrect answer, an ambiguous question or an explanation that sounds persuasive but does not address the student’s error.
It may also infer too much from too little evidence.
One incorrect answer should not become a permanent label in a student profile.
For that reason, I would want teacher checks, verified solutions and a clear way to challenge the feedback. Uncertain conclusions should remain marked as uncertain.
The Department for Education’s guidance similarly stresses that generative AI cannot replace expert subject knowledge and professional judgement. GOV.UK
There is another practical limitation: a normal chatbot conversation should not be assumed to maintain a complete, accurate learning record indefinitely.
A continuing course needs a deliberately maintained record that can be reviewed and corrected.
A Learning Profile Should Contain Only What Is Needed
The educational value lies in information such as:
Confuses radius and diameter; correctly identifies both after a diagram prompt.
It rarely requires a student’s full name, home address, family circumstances or detailed personal history.
A sensible starting point is a minimal learning record, using a student code where appropriate, in a setting approved for the intended use.
Before introducing a tool, check its suitability for the student’s age and how it handles the information entered.
The profile should also be written respectfully.
“Needs further practice interpreting ratio questions” is useful.
“Bad at maths” is a judgement that offers no next step.
What Could a Twenty-Minute Session Look Like?
Here is one possible structure:
| Time | Activity | Purpose |
|---|---|---|
| 3 minutes | Recall questions from earlier work | Check what remains accessible |
| 4 minutes | Short diagnostic task | Investigate a current difficulty |
| 6 minutes | Explanation and targeted practice | Address the identified gap |
| 5 minutes | Fresh independent application | Check whether support can be withdrawn |
| 2 minutes | Reflection and record update | Plan the next session |
The timings would change with the learner.
A student who needs more thinking time should receive it. Another may be ready to move quickly to a demanding application.
The record might end with:
Formed two equations independently. Needed one hint when subtracting negative terms. Next session: brief sign check, followed by a fresh word problem.
That is much more useful than simply recording “completed worksheet”.
A Small Starting Point Is Enough
Teachers and tutors do not need to begin with an elaborate automated platform.
A manageable trial could involve one topic, a small learning record and a teacher-reviewed set of follow-up tasks.
A useful instruction to AI might be:
Using the anonymised learning record below, propose a short revision session on simultaneous equations. Start with two diagnostic questions that help distinguish difficulty forming equations from difficulty solving them. Suggest different next steps depending on the answers. Include one independent application and a later review question. Provide checked solutions separately. Mark uncertain interpretations as hypotheses for the teacher to review.
This would produce a proposal for the teacher to assess.
The important test is whether the resulting tasks help the student learn more effectively—and whether the time spent maintaining the system is worthwhile.
Could Every Student Have Their Own Course?
Potentially, yes.
AI could make it more practical to prepare different explanations, tasks and review sequences for different learners.
But giving every student access to a chatbot would not, by itself, create an individual revision course. Access to suitable devices, clear guidance, dependable resources and appropriate support would still matter.
The strongest version would combine a carefully maintained learning record with teacher judgement and regular evidence of independent progress.
Two students might be working towards the same examination while following different routes through their revision.
One might need help understanding the question.
Another might need help selecting the method.
A third might be ready to explain why the method works.
They all deserve a suitable next step.
The most exciting possibility is that AI could help us act more consistently on something good teachers already know: the next question matters, and it should be chosen with the learner in mind.
No comments:
Post a Comment