Exam boards differ in ways that matter — command words, assessment objectives, mark distribution. Here's how to get AI producing model answers that actually match your board rather than generic exam-style prose.
Why board specificity is the whole game
AI has a general model of "a good exam answer" that averages across everything it has seen — including US and international qualifications. Ask for "a model GCSE answer" and you'll get something that looks plausible and matches no specification precisely. The fix is straightforward but must be done every time: give it the board's actual framework.
The anchoring prompt
Prompt — board anchoring
I teach [subject] with [AQA/Edexcel/OCR/WJEC]. Here are the assessment objectives and their weightings for this paper: [paste]. Here is the official mark scheme for a comparable question: [paste]. Here is the question I need a model answer for: [paste]. Write a top-band answer that explicitly satisfies these assessment objectives in these proportions. Do not use any structure or terminology from other exam boards.
Save this as your department's standard opening — everything else builds on it.
Command words matter more than people think
"Evaluate", "assess", "to what extent", "explain" and "describe" have board-defined meanings and expected structures. AI blurs them.
Prompt — command word discipline
The command word in this question is '[word]'. For [board], this requires: [paste the board's definition]. Write the model answer so its structure explicitly reflects that requirement, and afterwards explain in one sentence how the structure meets it.
Producing answers at realistic grades
Prompt — graded set
Write three answers to this question at grades 4, 6 and 8. The differences should reflect the mark scheme's band descriptors, not just length. Show realistic pupil-level writing — the grade 4 answer should contain the kind of errors real pupils make, not be artificially poor.
The checks that must happen before it reaches pupils
- AO coverage — does the answer actually hit the objectives in the right proportions?
- Factual accuracy — subject content errors are the most common problem
- Quotations and data — verify every one; AI invents them fluently
- Board terminology — no imported language from other specifications
- Level — has it drifted above GCSE?
- British spelling and context
A department-level habit worth adopting: nothing AI-generated goes to pupils as an exemplar until a second teacher in the department has checked it. Exemplars carry authority with pupils — an error in one propagates into thirty essays.
Where this genuinely helps
Departments that use it well aren't replacing their exam expertise — they're removing the typing. The teacher who knows exactly what a grade 7 response looks like can now get 80% of it on the page in a minute and spend their time refining rather than drafting. See AI for mark schemes and model answers.
Want board-accurate AI use across your department?
We build prompts around your actual specifications. 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 department.
Get one-to-one AI support for your schoolFrequently asked questions
How do I get AI to write board-specific model answers?
Paste in the actual assessment objectives and weightings for your paper, an official mark scheme for a comparable question, and instruct it not to use structure or terminology from other exam boards. Without this, AI produces generic exam-style prose matching no specification precisely.
Does AI understand AQA and Edexcel differences?
Not reliably. It has a general model of a good exam answer averaged across everything it has seen, including US qualifications. You must supply the board's actual framework each time — that's the difference between usable and generic output.
How should command words be handled with AI?
Paste your board's own definition of the command word and instruct AI to structure the answer so it explicitly reflects that requirement. Terms like 'evaluate' and 'to what extent' have board-defined meanings and expected structures that AI otherwise blurs.
What checks are needed before using AI model answers with pupils?
Check AO coverage and proportions, factual accuracy, every quotation and data point, board-specific terminology, whether the level has drifted above GCSE, and British spelling. Ideally a second teacher checks anything used as an exemplar, since errors propagate into pupils' work.