Can AI Create Better Revision Materials?
Why the most useful question may be the one chosen after a mistake
Revision becomes much more useful when the questions change according to what the student gets wrong.
A student completes a revision worksheet, checks the answers and discovers that several are incorrect. What happens next?
Quite often, they move on to another worksheet.
But those mistakes contain valuable information. Perhaps the student understands probability but forgets that eating a sweet changes what remains in the bag. Perhaps they can use standard form but struggle with the arithmetic inside the calculation. Perhaps they know a scientific definition but cannot apply it to an unfamiliar situation.
AI can help turn that information into focused practice. It can draft graded questions, flashcards, quizzes, summaries and revision checklists. Its value depends on whether those materials address what the student actually needs to learn.
After around forty years in teaching, my interest is in how this technology can support an essential part of tuition: finding the point at which understanding breaks down, then helping the student move forward.
Three Requests, Three Different Kinds of Revision
Consider these three instructions:
| Request | What it offers | What still needs checking |
|---|---|---|
| “Give me ten GCSE probability questions.” | General practice across a topic. | Relevance, difficulty and coverage. |
| “Create original probability questions aligned with this exam-board specification and tier.” | Practice directed towards the course being studied. | Whether the questions genuinely match the supplied requirements. |
| “Create questions for a student who keeps denominators unchanged when objects are removed.” | Practice targeting an observed misconception. | Whether that misconception explains the student’s mistakes. |
Each adds useful information. The strongest approach combines the course requirements with evidence from the student’s work.
Naming an exam board alone does not establish alignment. Supply the relevant current specification extract and suitable official examples, then check the result.
AI-generated questions should be labelled as original practice, with suggested marks where useful. They should never be presented as official exam-board questions.
The Sweet That Changes the Next Question
Imagine a bag containing four red sweets and three yellow sweets. A student takes one sweet, eats it, then takes another.
The question asks for the probability that both are red.
The student writes:
4/7 × 4/7 = 16/49
There is useful understanding here: the student has multiplied probabilities along a branch. The mistake concerns what happens after the first sweet is eaten.
A helpful follow-up sequence would be:
- After a red sweet is eaten, how many red sweets remain?
- How many sweets remain altogether?
- What is the probability that the second sweet is red, given that the first was red?
- Now calculate the probability that both sweets are red.
- How would the calculation change if the first sweet were replaced before the second selection?
The answers are three red sweets, six sweets altogether, and a second probability of 3/6.
Therefore, without replacement:
4/7 × 3/6 = 2/7
With replacement:
4/7 × 4/7 = 16/49
The contrast helps expose the underlying idea. We can then use a different bag or context to check whether the student can apply it independently.
This is where AI can help: drafting carefully varied follow-up questions once a teacher has identified a likely difficulty.
One wrong answer should prompt investigation. It is not enough to establish exactly what a student does or does not understand.
Graded Questions Need Purposeful Steps
“Easy, medium and hard” is a start, but good progression requires more thought.
For surds, a sequence might move through:
- Simplifying a single surd.
- Combining surds after simplification.
- Expanding brackets containing surds.
- Rationalising a denominator.
- Solving a problem that combines these skills.
The harder question should demand an additional decision or connection. Simply increasing the numbers may add arithmetic without improving the learning.
A useful prompt would be:
Create eight original GCSE Higher Maths surd questions, progressing from simplification to a problem combining expansion and rationalisation. Identify the skill each question assesses. Include two short checks for common misconceptions. Put worked solutions in a separate section and check each calculation.
I would still work through the solutions myself before giving the sheet to a student. Asking AI to check its work can help, but it does not provide an independent guarantee.
Once a student gains confidence, mixed practice also matters. Can they choose the right method when the heading no longer tells them which technique to use?
Flashcards Should Make the Learner Retrieve Something
AI can quickly turn notes into flashcards. The challenge is choosing useful questions.
A card saying “Photosynthesis” on the front and an entire paragraph on the back gives the learner a large, uncertain task.
More focused cards might ask:
- What is the word equation for photosynthesis?
- Why can light intensity limit the rate of photosynthesis?
- Why might increasing light intensity eventually stop increasing the rate?
These separate recall from explanation and application.
The learner should attempt an answer before turning the card over. Reading both sides repeatedly can feel productive without showing whether the information is available from memory.
The Education Endowment Foundation emphasises planning retrieval activities around their purpose and the particular knowledge pupils need to recall. Generating a large deck is only the beginning. EEF
Ten carefully chosen cards that are used thoughtfully can be more useful than a huge collection that becomes another unfinished task.
Quizzes: Make a Wrong Answer Informative
A useful multiple-choice quiz includes plausible wrong answers.
For the sweet problem, 16/49 is informative because it may indicate that the student kept the probabilities unchanged after removing a sweet. A random fraction tells us less.
Ask AI to explain the possible misconception behind each incorrect option. Then ask the student to explain their choice.
Even a correct answer needs interpretation. Was it a confident calculation, a guess, or recognition of a familiar pattern?
For some learners, I would begin with three short questions and discuss the reasoning behind each. A small quiz can provide enough information to decide what to teach next.
Summaries Should Lead to Independent Explanation
AI can draft concise summaries from checked notes, but polished wording can hide shallow understanding.
A better sequence is:
- Study a short, checked explanation.
- Close it.
- Explain the idea aloud or write it from memory.
- Compare the explanation with the original.
- Correct the gaps and answer an application question.
For Biology, a summary of diffusion should retain the idea of net movement down a concentration gradient. Making the language simpler must preserve the science.
I would also ask for a “commonly confused with” section. Diffusion, osmosis and active transport become clearer when students explain their differences.
A useful question after reading a summary is:
“Could you explain this to someone else without looking?”
Revision Checklists: Replace “Covered” with Evidence
A tick beside “circle theorems” tells us little about what a student can do.
A stronger checklist uses observable statements:
- I can identify the relevant theorem in an unfamiliar diagram.
- I can calculate a missing angle.
- I can give the reason for each step.
- I can connect several theorems in one problem.
- I can solve a fresh question without hints.
Useful columns would be:
| Skill | Needs teaching | Can do with help | Can do independently | Checked again later |
|---|
Confidence matters, but independent performance provides stronger evidence of readiness.
A student saying “I understand it now” is encouraging. Solving a different question a week later tells us more.
A Practical Routine: Attempt, Inspect, Adapt, Revisit
Here is a manageable way to use AI-assisted revision.
Start with a short independent attempt.
Keep answers and hints out of sight.
Look at the working.
Does the difficulty concern knowledge, interpretation, method, arithmetic or explaining a reason?
Generate focused follow-up practice.
Ask for a short explanation, a supported example and a fresh independent question.
Check the material.
Verify the answer, wording, syllabus relevance and any diagrams.
Return later.
Try a new question after a gap, then include the skill in mixed practice. The EEF’s evidence review considers both spaced learning and retrieval practice; neither depends on AI. d2tic4wvo1iusb.cloudfront.net
For a student who is easily overwhelmed, this might mean one question at a time, plenty of space and brief instructions. We can make the task easier to access while preserving the intended intellectual challenge.
A Reusable Prompt for Better Revision
Create a short revision activity for [qualification, subject, exam board, specification code and tier where applicable].
Use the specification extract and checked notes below. Flag any missing information rather than inventing requirements.
The student can already [skills]. Their recent work suggests difficulty with [specific error and an anonymised example].
Include two diagnostic questions, one brief worked example, four progressively challenging questions and one transfer question in a different context.
Include one question checking earlier learning. Use clear UK English and put suggested answers and worked solutions in a separate section.
Explain the purpose of each question. Identify uncertainties and anything requiring teacher verification. Label all questions as original practice.
[Insert relevant source material.]
For flashcards, replace the activity instructions with:
Create ten focused question-and-answer cards, each testing one clearly defined point.
Describe learning needs without including names, school details or confidential records.
Where Professional Judgement Still Matters
AI can produce an incorrect solution, an ambiguous question, a misleading diagram or a convincing explanation that omits a crucial detail.
Department for Education guidance makes clear that education professionals remain responsible for checking the accuracy and appropriateness of AI-produced content. It also notes that evidence about the effects of students’ AI use is still emerging. GOV.UK
That distinction matters. Evidence supporting a revision method does not establish that every AI tool implements it well.
In my tuition, the aim remains to discover what a student can do independently, explain what they need next and provide practice that builds towards the examination. Practical work, conversation, subject knowledge and careful feedback all contribute to that judgement.
AI can support the preparation. The student must still do the thinking.
Better Revision Begins with a Better Response to Mistakes
Can AI create better revision materials? Yes—when it helps us produce accurate, relevant practice and respond intelligently to a student’s work.
A beautiful summary is useful only if the student can explain its ideas. A flashcard deck matters only if the learner can retrieve the answers. A worksheet earns its place when it helps resolve a difficulty or develops an important skill.
At Hemel Private Tuition, this is the purpose of personalised support: understanding where the learner is now and choosing a useful next step.
The most valuable revision question may be the one we choose because of the answer the student has just given.




