Blog › AI Quiz Tools

15 AI Prompts That Generate Genuinely Good Quiz Questions

Ask an AI assistant to “make 10 questions about photosynthesis” and you will get 10 questions a student can answer by skimming the first paragraph of any textbook. They will be almost entirely recall, the wrong answers will be obviously wrong, and several will test vocabulary rather than understanding.

The problem is not the model. A vague prompt gives it nothing to aim at, so it produces the statistical average of every quiz it has ever seen. The prompts below are longer and fussier than what most people type, and that is the entire point. Each is copy-pasteable, with placeholders in square brackets for you to fill in.

The anatomy of a prompt that works

A prompt that reliably produces usable questions names most of these seven things:

  • The source material, pasted in, with an instruction to use only that source
  • The audience and level in real terms, such as “Year 9 students who have studied this for two weeks”
  • The question type and how many of each
  • The cognitive level you want, ideally named using Bloom’s taxonomy
  • Rules for the wrong answers, which is the part almost everyone leaves out
  • The output format you need for import or copy-paste
  • An answer key with a short explanation for each item

Bloom’s taxonomy is worth naming explicitly because it gives you and the model a shared vocabulary for difficulty. Its six levels run from remember, to understand, apply, analyse, evaluate, and create. Most AI-generated questions sit at remember unless you say otherwise. If you would rather use a purpose-built tool than a chat window, our comparison of the best AI quiz generators for teachers covers what each one does.

Foundational prompts

1. Generate from source text only

You are writing quiz questions for [YEAR OR GRADE LEVEL] students studying [SUBJECT]. Use only the source text below. Do not add any facts, examples or terminology that do not appear in it. If the text does not contain enough material for a question, say so instead of inventing one. Write [NUMBER] multiple-choice questions with four options each. After the questions, give an answer key with a one-sentence explanation for each, quoting the phrase from the source that supports the answer. SOURCE TEXT: [PASTE YOUR TEXT HERE]

This is the workhorse and the one to start with. Asking for the supporting quotation is what makes it trustworthy, because it turns your review into a quick check rather than a full re-read. Watch for the model quietly ignoring the source restriction when your text is short.

2. Target a specific Bloom’s level

Write [NUMBER] multiple-choice questions on [TOPIC] for [LEVEL] students. Every question must sit at the [apply / analyse / evaluate] level of Bloom’s taxonomy. That means a student who has memorised the definitions but does not understand the concept should get them wrong. Before each question, state in one line why it belongs at that level. Avoid any question that can be answered by recognising a keyword.

Making the model justify the level forces it to self-check, and the justification lines are easy to delete afterwards. If the reasons look thin, the questions usually are too.

3. Build a spread of difficulty

Create a [NUMBER]-question quiz on [TOPIC] for [LEVEL] students, ordered from easiest to hardest. Label each question Easy, Medium or Hard. Roughly 30 percent should be Easy recall, 50 percent Medium application, and 20 percent Hard analysis. The Hard questions must require combining two separate ideas from the topic. Include an answer key with explanations.

Use this when you want one quiz that neither demoralises the weakest students nor bores the strongest. Check that the Hard items genuinely require two ideas rather than just using longer sentences.

Prompts for deeper thinking

4. Scenario and application questions

Write [NUMBER] multiple-choice questions on [TOPIC] for [LEVEL] students. Each must open with a short realistic scenario of two or three sentences, set in a context the students have not seen in class, then ask what would happen, what the student should do next, or why the outcome occurred. Do not name the concept being tested anywhere in the question. Include an answer key explaining the reasoning.

The instruction not to name the concept is the important one. It stops the model writing “Using Newton’s second law, calculate…” which hands the student the method for free.

5. Questions that cannot be answered without the source

Using only the passage below, write [NUMBER] questions that a well-read student could NOT answer correctly from general knowledge alone. Each must depend on a specific detail, figure, comparison or argument that appears only in this passage. After writing them, review your own list and delete any question that could plausibly be answered by someone who had not read the passage. PASSAGE: [PASTE HERE]

This is the fix for the familiar complaint that students score well on a comprehension quiz without doing the reading. Keep the built-in deletion pass, since that is where most of the value sits.

6. Compare and contrast

Write [NUMBER] questions for [LEVEL] students that require comparing [CONCEPT A] and [CONCEPT B]. Focus on the boundaries where students commonly confuse the two. Each question should present a case and ask which concept applies, or ask what would change if one were swapped for the other. Include an explanation of the distinction for each answer.

Well suited to topics where the mistakes are predictable, such as mitosis and meiosis, weather and climate, or mass and weight.

Prompts for better wrong answers

Distractors are where most AI-generated quizzes fall apart. If the wrong options are obviously wrong, the question tests nothing but reading speed.

7. Distractors built from real misconceptions

First, list the [NUMBER] most common misconceptions that [LEVEL] students hold about [TOPIC]. Then write one multiple-choice question for each, where the incorrect options are those misconceptions stated as though they were correct. Every wrong option must be something a real student might genuinely believe. No joke options and nothing obviously absurd. In the answer key, name which misconception each distractor targets.

This turns a quiz into a diagnostic instrument. When a class converges on the same wrong option, you know exactly what to reteach.

8. Strip out the giveaway cues

Rewrite the questions below so no answer can be guessed from surface cues. Apply these rules: all four options similar in length and grammatical structure; the correct answer must not be the longest or most detailed; no “all of the above” or “none of the above”; every option grammatically consistent with the stem. Return the revised questions and note what you changed. QUESTIONS: [PASTE HERE]

Run this as a second pass over any generated set. Test-wise students score noticeably higher on quizzes that fail these rules, whatever they actually know.

9. Near-miss distractors

For each question below, replace the weakest distractor with one that is nearly correct: right idea but wrong scope, right process but wrong order, or correct except for one specific condition. It must be defensible enough that a student who half-understands the topic would choose it. State briefly why each new distractor is tempting but wrong. QUESTIONS: [PASTE HERE]

Use with care. Pushed too far this produces two defensible answers, so its output needs the closest reading of anything here.

Prompts for formats and levels

10. Output as an importable table

Write [NUMBER] multiple-choice questions on [TOPIC] for [LEVEL] students. Output them as a table with exactly these columns: Question, Option A, Option B, Option C, Option D, Correct Option Letter, Explanation. One question per row, no commentary before or after the table, and no commas inside any cell.

This saves more time than anything else on the page, because a clean table pastes into a spreadsheet and then into a quiz platform’s bulk import. The rule about commas prevents broken columns if you save as CSV. For the full document-to-quiz workflow, see our guide to turning a PDF or textbook chapter into a quiz.

11. Adjust the reading level

Rewrite the questions below for students reading roughly [NUMBER] years below grade level. Keep the concept and the difficulty of the thinking exactly the same. Shorten sentences, replace uncommon words with everyday ones, and remove unnecessary clauses. Do not simplify the subject-specific vocabulary students are expected to learn. Show the original and the rewrite side by side. QUESTIONS: [PASTE HERE]

The distinction between simplifying the language and simplifying the thinking is what makes this useful for mixed-ability classes and for students working in an additional language.

12. True or false, and short answer

Write [NUMBER] true or false statements about [TOPIC] for [LEVEL] students. Each false statement must be false for a specific, identifiable reason rather than obviously untrue. Give the answer and a one-sentence correction for each false one. Then write [NUMBER] short-answer questions on the same topic answerable in under 20 words, with a model answer and the key term that must appear for credit.

Naming the key term needed for credit makes marking faster and far more consistent, particularly if a colleague is marking alongside you.

13. Diagnostic pre-test

I am about to teach [TOPIC] to [LEVEL] students. Write a [NUMBER]-question diagnostic quiz to find out what they already know and what they get wrong. Cover the prerequisite knowledge they should already have, plus the two or three ideas within the topic students most often misunderstand. For each question, tell me what a wrong answer would reveal about that student’s gap.

That last instruction is what makes this worth running. You get a quiz and a guide to reading the results together. It pairs well with the planning approach in our guide to creating an effective assessment.

14. Turn one exam question into practice variants

Here is a past exam question: [PASTE QUESTION]. Produce [NUMBER] variants that test the same underlying skill but change the context, the numbers and the surface details, so a student cannot pass by memorising the original answer. Keep the difficulty and the command word the same. Provide worked answers. Then write one variant that is slightly harder and explain what makes it harder.

Strong for revision, and it stops students pattern-matching to a single memorised worked example.

The prompt that fixes the other fourteen

15. Make the AI critique and repair its own quiz

Review the quiz below as a strict assessment moderator. For each question report: whether more than one option could be defended as correct; whether the answer relies on outside knowledge rather than the source provided; whether any distractor is implausible; whether the correct answer is cued by length, grammar or specificity; and what Bloom’s level it actually tests rather than the one intended. List every problem, then output a corrected version of the whole quiz. Be harsh. QUIZ: [PASTE HERE]

Running this as a separate pass, ideally in a fresh conversation, catches a surprising amount. Models are consistently better at criticising a finished draft than at getting it right first time. It does not replace your own review, but it clears the obvious problems before you start reading.

You still have to check the output

No prompt here removes the need to read every question before it reaches a student. A wrong answer key in a graded assessment costs far more time to unpick afterwards than the ten minutes of review would have taken.

Four things to look for specifically:

  • Check every marked answer against your own source material, not against the model’s explanation. The explanation can be confidently wrong in exactly the same way the answer is.
  • Read all four options as an able student hunting for an argument. If you can build a case for a second option, rewrite it.
  • Scan for giveaway cues: the longest option, the only one that reads grammatically after the stem, the only one carrying a qualifier such as “usually”.
  • Treat every specific date, figure, statistic or named example with suspicion. This is where models hallucinate most often.

Once the questions are clean you still need somewhere to run them. AI tools write questions but do not deliver a quiz to a live class, which is a separate job covered in our comparison of AI quiz generators against Quizizz and Kahoot. If you are a student rather than a teacher, several of these prompts work just as well for building your own practice sets, alongside the options in our roundup of free AI study tools for students.

Frequently Asked Questions

Which AI assistant works best for writing quiz questions?

The major general-purpose assistants all handle these prompts reasonably well, and the gap between them is smaller than the gap between a vague prompt and a specific one. A more useful test is whether the tool lets you upload your source material directly, because generating from your own text produces better questions than generating from the model’s general knowledge.

How much source text should I paste in?

A single section or sub-topic of a few hundred to a couple of thousand words usually works best. Paste in a whole chapter and the questions spread thin across it with uneven coverage. Splitting a long chapter into sections and running the prompt on each gives tighter, better-targeted questions.

Can I get questions in a language other than English?

Yes. Add an instruction such as “write the questions and answer key in [LANGUAGE]”. Quality varies by language and tends to be strongest in widely used ones. Review matters even more here, since subject-specific terminology is a common failure point in translation.

Is it acceptable to use AI-generated questions in a graded exam?

That depends on your school or institution’s policy, so check it first. Where it is permitted, the practical standard is the same as for any question you did not write yourself: you are responsible for its accuracy and fairness. For anything carrying marks, review every item against your source and have a colleague read it as well.