Creating quizzes and question banks with AI

Good quiz questions are slow to write. The question itself is the easy part. The hard part is the wrong answers: options that are tempting to someone who has not quite got it, but clearly wrong to someone who has. Most owners run out of patience after ten questions and end up with quizzes that test whether members were awake rather than whether they learned anything.

AI assistants can draft dozens of questions in the time it takes you to write three. They can also produce questions with two correct answers, a key that is simply wrong or an option so long it gives the game away. The workflow below uses AI for volume and variety and keeps you, the expert, in charge of accuracy.

Start from objectives, not from the lesson text

If you paste a lesson and ask for a quiz, you will mostly get recall questions about whatever sentences stood out: a date, a term, a number. Those are easy to write and rarely worth asking. Questions work better when they check the skill a lesson promises.

So begin with your learning objectives. If you do not have them yet, writing learning objectives with AI walks through it. Rosa, who runs a sourdough baking school, has an objective that says members will “judge whether a dough is ready to shape from its look and feel.” That points her toward questions built around descriptions of real doughs, not questions asking her members to define fermentation.

A prompt that produces usable questions

The more precisely you describe what a good question looks like, the less editing you do later. This prompt sets out the rules most quiz writers learn the hard way.

Write [number] multiple-choice questions for the learning objective below. The learners are [describe level]. Base the questions on this lesson content: [paste lesson text or transcript].

Rules: each question describes a realistic situation where possible, rather than asking for a definition. Give four options with exactly one correct answer. Make each wrong option a mistake a real learner might make, and keep all options a similar length. Do not use “all of the above,” “none of the above” or negatives like “which is NOT.” For every option, write one sentence explaining why it is right or wrong. Tag each question with the objective and an easy, medium or hard label.

Objective: [paste objective]

A good result for Rosa might read: “Your dough has risen by about half, the surface is slightly domed and a floured finger poke springs back slowly, leaving a small dent. What should you do?” The options would include shaping now (correct), waiting several more hours, adding flour and punching it down. Each wrong option matches a real beginner habit, and each explanation teaches something.

For quick drafting, ChatGPT, Claude, Gemini and Microsoft Copilot all handle this kind of task. Run the same prompt in two of them on one of your own lessons and compare the wrong answers in particular, because that is where quality varies most. If the prompts themselves feel awkward to write, prompting basics covers how to give clear instructions.

Build a bank, not just a quiz

A question bank is a larger pool of questions, tagged by objective and difficulty, from which quizzes can be drawn. It lets you offer retakes with different questions, mix easy and hard items, and swap out a weak question without rewriting the whole quiz. Many membership platforms can draw questions at random from a pool, which makes a bank even more useful.

Once you have a few solid questions per objective, ask the AI to produce variations:

• The same skill applied to a different situation or example.
• A harder version with less information or an extra complication.
• A different format: putting steps in order, matching terms to descriptions or a short written answer.

Ask for variety in names, places and contexts as well. A professional association for event planners, for example, would want scenarios covering weddings, conferences and charity galas, not the same hotel ballroom twenty times.

The expert review pass

Every question needs a person who knows the subject to check it before a member sees it. AI makes errors with total confidence, and a wrong quiz key teaches the wrong thing and damages trust. Read each question with this checklist:

• Is the key right? Work out the answer yourself before looking at the AI's choice.
• Is only one option right? Watch for a wrong option that is defensible in some situations.
• Is there a giveaway? The longest, most detailed or most carefully hedged option is often the correct one.
• Is it fair? Could a learner who did the lesson answer it, or does it rely on something you never taught?
• Is the wording clear? Remove double negatives, unfamiliar idioms and trick phrasing.

One useful extra check: paste the questions without the answers into a fresh chat and ask the AI to answer them and explain its reasoning. Where it hesitates, picks a different answer or argues for two options, the question is probably ambiguous. This does not replace your review, but it helps you find the questions that need a second look.

Make the feedback do some teaching

The explanations you asked for are not just for your review. Show them to members after they answer. A short note on why a wrong answer is wrong, with a pointer back to the right lesson, turns a quiz from a test into practice. For more on that idea, see quizzes that help members learn, not just test them.

Edit the explanations into your own voice. They are the part members read most closely, especially after a wrong answer, so they should sound encouraging and specific.

Getting started

1. Choose one lesson and confirm its learning objectives.
2. Run the question prompt for one objective and ask for ten questions.
3. Answer each question yourself before checking the key, and fix or delete anything doubtful.
4. Test the survivors in a fresh chat to catch ambiguous wording.
5. Ask for variations to grow a small, tagged bank for that objective.
6. Edit the explanations into your voice and switch them on as feedback for members.

Repeat for one objective at a time. A small bank of questions you trust is worth far more than a large one you have never read.