Finding word problems that match exactly the small step you're teaching is a familiar frustration. AI generates them to order — here's how to get problems that genuinely fit your scheme.
Anchor it to the small step
The reason generic AI maths problems disappoint is that they don't match where your class actually is. Naming the specific small step — and, crucially, what the children have not yet been taught — is what produces usable output.
Prompt — problems for a small step
Write 8 word problems for Year 4 on the White Rose small step '[name the small step]'. The children have been taught [list what they know]; they have NOT yet been taught [list what to avoid]. Use only numbers and methods appropriate to that. UK contexts and pounds and pence. Include the answers with the calculation shown.
That "have NOT yet been taught" line is the one most people miss, and it's what stops AI slipping in methods your class hasn't met.
Varying the structure, not just the numbers
Children get good at spotting a pattern rather than reading the maths. Deliberate structural variation fixes this.
Prompt — varied problem types
Write 6 word problems on [small step] where the unknown is in a different position each time — some where the answer is the total, some the starting amount, some a missing part. Same mathematical content, different structures, so children can't pattern-match.
Reasoning and problem-solving
Prompt — reasoning
Write 4 reasoning questions for Year 5 on [small step] in White Rose style: one 'spot the mistake' where a character has made a specific common error, one 'always, sometimes, never', one requiring an explanation of method, and one multi-step problem. Include what a good answer would contain.
The "spot the mistake" type is particularly valuable — it targets a misconception directly, and AI produces them quickly if you name the error you want featured.
Differentiating
Prompt — three levels
Take this word problem and write three versions: one with smaller numbers and simpler language, one as written, and one requiring an extra step. All should test the same mathematical understanding: [paste problem].
Prompt — reducing reading load
Rewrite these word problems for a pupil who finds reading difficult: shorter sentences, simpler vocabulary, less text — but keep the mathematical demand exactly the same: [paste].
Always check the maths. AI makes arithmetic errors, produces problems with impossible or unrealistic answers, and occasionally writes questions that can't actually be solved with the information given. Work through every answer before it reaches children — a wrong answer in a maths resource undermines the pupils who got it right.
Other things to check
- UK contexts — pounds not dollars, metres not feet, familiar settings
- Realistic numbers — AI produces problems about buying 47 apples
- Reading level appropriate to the year group
- Inclusive names and contexts
Build a bank as a team
Year groups teaching the same small steps can split the work and share the output — a term's word problems in an afternoon rather than each teacher generating their own. See reducing teacher workload with AI.
Want a term's worth of problems in an afternoon?
We show year teams how to build shared resource banks with AI. AskColin gives schools low-cost monthly support with one-to-one help whenever you need it — so when you're stuck on a real job at 8am, there's someone to ask. Practical, jargon-free, built around your team.
Get one-to-one AI support for your schoolFrequently asked questions
Can AI generate maths word problems matched to White Rose?
Yes — name the specific small step, list what the children have been taught, and crucially list what they have NOT yet been taught so AI avoids methods your class hasn't met. Specify UK contexts and ask for answers with the calculation shown.
How do I stop AI word problems being too samey?
Ask for problems where the unknown is in a different position each time — some where the answer is the total, some the starting amount, some a missing part. Same mathematical content with different structures stops children pattern-matching instead of reading the maths.
What should teachers check in AI maths problems?
The maths itself. AI makes arithmetic errors, produces unrealistic answers, and occasionally writes questions that can't be solved with the information given. Also check UK contexts, realistic numbers, reading level and inclusive names.
How can AI help with maths reasoning questions?
Ask for White Rose-style reasoning types — 'spot the mistake' featuring a specific common error, 'always, sometimes, never', explanation of method, and multi-step problems — with a note of what a good answer contains. Naming the misconception you want featured works well.