Generating an exam with AI, and checking it properly
AI will produce twenty questions in a minute. A four-minute review catches the failures that generated papers reliably produce.
27 August 2026 2 min read
Generating exam questions is the single most useful thing AI does for a teacher, because writing questions is slow, repetitive, and easy to check. The saving is real. What makes it safe is knowing that generated papers fail in particular, predictable ways rather than randomly.
Ground it in your own material
A question generated from general knowledge of a subject is a question about that subject somewhere in the world. A question generated from the chapter you actually teach is about your syllabus, in your terminology, at your level. The difference is large in any curriculum and much larger in Arabic, where the same concept carries different accepted terms across textbooks. Always supply the source document.
Say what you want, in the terms that matter
"Make an exam on this chapter" produces twenty recall questions. The instructions worth giving are the ones a colleague would need: how many marks in total, the split across recall, application and analysis, which topics and in what proportion, the question formats, and the time the paper should take. A generated paper is only as balanced as the brief.
The four-minute review, in order
- Solve the questions you did not immediately verify. Wrong answer keys are the most common serious failure and the least visible: the key looks authoritative and nothing marks it as suspect.
- Check the distractors. Generated options tend to include one absurd choice and one accidentally defensible one. Both need rewriting.
- Cover the stem and read the options. The leak test from any well-written MCQ still applies, and generated questions fail it often: the correct option is regularly the longest and most qualified.
- Check the terminology against your textbook. The most frequent Arabic failure is a fluent, correct explanation using a term your students have never seen.
- Count the cognitive levels. Ask for 20% analysis and verify you got it. Generated papers drift toward recall regardless of the brief.
What generation does not give you
It does not know which questions your group will find hard, because it has never taught them. Difficulty comes from your records of what students actually got right. That is the argument for keeping those records, and the reason a generated paper is a starting point rather than an exam.
Sources
Frequently asked questions
What is the most common error in AI-generated exam questions?
A wrong answer key. It is the most serious failure and the least visible, because the key looks authoritative and nothing marks it as suspect. Solve every question you did not immediately verify.
Should I give the AI my own material?
Always. A question generated from general subject knowledge is about that subject somewhere in the world; one generated from your chapter is in your terminology at your level. That matters especially in Arabic, where accepted terms differ between textbooks.
Can AI judge how difficult a question is?
No. It has never taught your group. Real difficulty comes from your own records of what students actually got right, which is why a generated paper is a starting point rather than a finished exam.
Topics in this article
Related posts
Surviving exam season without burning out
For a private tutor, exam season is the month the income arrives and the month the work becomes unmanageable: extra revision groups, more marking, more anxious…
When students use AI: what to change about the homework
Your students have access to AI, they know how to use it, and some of their homework is already produced by it. Three responses are available and two of them…
