Using past papers to guide students’ learning is something teachers know well.
It can require considerable work, though; sifting through years of papers, identifying relevant questions, reading examiner reports and understanding what candidates did well on and what they found tough.
It’s vital work, however. Within our maths department, we may issue up to 48 assessments annually, half of which are 50-minute, mid-module assessments and half 90-minute, exam-style simulations for our key stage 5 students to track progress and ensure that any issues are spotted.
Over the past year, I have been using AI to help with this work by getting it to find suitable past-paper questions, adapt existing ones, check questions I’ve written and carry out larger-scale research that would not be feasible alongside full-time teaching.
This has not replaced my professional judgement; I still check everything the AI shows, edit it as required and correct inaccuracies.
But it has made forms of analysis, checking and iteration possible at a scale and speed full-time teaching previously made unrealistic.
AI can find past exam questions
I begin with a brief: the assessment purpose, topics, intended demand and total marks.
Small consecutive prompts get a web-enabled AI to search online and review the specification, past papers, mark schemes and examiner reports, working through years of examination series in batches, and recording what it has checked or been unable to find.
I cannot manually work through every useful question across years of papers, so AI can do much of the initial searching and sorting before I review the shortlist.
I might ask for a table linking each question part to its marks, skill, cues, reasoning required and any successes or difficulties identified by examiners, with the relevant report passage and page reference alongside each judgement.
I can then check through the evidence provided.
Analysing examiner comments
I also get the AI to provide a broad analysis of examiner comments so I know what issues can crop up with certain types of questions and can work on these with my students.
For example, I might ask for six questions on one technique: one generally well answered, two with mixed performance and three reported as demanding.
If the examiner reports cannot support that split, I tell the AI to say so. I may also use older papers where appropriate, preserving recent ones for students’ independent revision.
Examiner reports can also help me to investigate why candidates struggled before I adapt a question.
Previously, I would usually work through questions manually to spot where a prerequisite skill was getting in the way of the mathematics I actually wanted to test.
A student might appear to struggle with differentiation, for example, when the real obstacle was rearranging an equation before they could begin.
AI can help to identify that kind of recurring barrier across questions and find or create an alternative that tests the target mathematics more directly.
A few prompts can leave me reviewing a co-adapted A-level question worth perhaps 12 marks rather than redesigning it from scratch.
In a busy week, that matters. I still check the mathematics and solution myself, but AI can do much of the initial detective work.
Writing questions
When I write questions myself, I can ask AI to challenge my design rather than simply solve them: are there unintended routes or awkward values? Could an early error block later mathematics or an incorrect method accidentally produce the right answer?
Previously, designing the question was only part of the job: it still needed checking and a mark scheme.
With the right prompting, AI can draft a question using the conventions of my examination board and consider several legitimate solution routes. I can then decide how those alternatives should be rewarded and refine the question or scheme.
Once the paper is assembled and I have worked through it myself, I also ask AI to act as a second checker against agreed parameters to spot any issues with how it has been constructed.
There is a more straightforward time-saving, too: AI can draft and format the complete mark scheme rather than me locating and compiling individual schemes question by question.
Throughout all of this, my professional judgement remains central. AI is not deciding what my students should learn or replacing professional discussion within my exceptional department.
However, using AI has made how we use assessment more iterative, more detailed and of greater breadth - while also reducing workload, thereby giving me more capacity to better support my students.
Shabih Bashir is head of key stage 5 mathematics at Doha College, Qatar
