How to Write Good Multiple Choice Questions

How to write good multiple choice questions

A good multiple choice question tests one idea, asks it plainly, and offers wrong answers that a learner who half understands the material would seriously consider. Most bad questions fail on that last point: three throwaway options next to one obviously correct answer measures reading speed, not knowledge. This is a practical guide to writing items that actually discriminate between people who understood the material and people who did not.

The anatomy of a multiple choice question

Every MCQ has three parts, and the names matter because they help you see which part is broken.

• The stem. The question itself, plus any setup it needs.
• The key. The correct option.
• The distractors. The incorrect options. This is where question quality lives or dies.

When a question is too easy, it is almost never because the key was too obvious. It is because the distractors were too weak.

Rule 1: put the whole question in the stem

A learner should be able to read the stem, cover the options, and know what is being asked. If the stem is a sentence fragment that only makes sense once you have read all four options, you are testing puzzle solving.

Weak: "The testing effect is..." followed by four sentence endings.

Better: "What does the testing effect describe?" followed by four complete answers. Same content, and now the learner is retrieving an answer rather than pattern matching four fragments against a stub.

Rule 2: test one thing per question

Two ideas bolted together with an "and" produce a question nobody can learn from. If a learner gets it wrong, neither you nor they know which half they missed. Split it into two questions. They are cheap.

This gets violated most often when someone is trying to hit a question count and starts merging thin material rather than admitting that a five minute video does not contain twenty testable points.

Rule 3: write distractors from real misunderstandings

This is the whole craft. A distractor should be the answer a learner gives when they have a specific, common, wrong model in their head. The best source of distractors is what people actually get wrong: the misconception the video was made to correct, the step people skip, the two terms that get confused, the number that is right for the adjacent case.

Nonsense distractors feel safe and teach you nothing. If nobody ever picks option C, option C is not doing any work and the question is really a three option question with a decoration.

A useful test once results come in: look at which distractors get chosen. A distractor that pulls a meaningful share of learners is a good one, and it is also telling you precisely where your material is unclear. That is worth more than the score.

Rule 4: keep the options parallel

Options should look like siblings: similar length, similar grammar, similar level of detail. Question writers unconsciously over specify the correct answer, because they are thinking carefully about it and carelessly about the other three. The result is the oldest tell in test taking, which is that the longest, most qualified option is usually right. Experienced test takers will exploit that without knowing a thing about your material.

Read the four options as a set before you move on. If one is noticeably longer or more hedged, either trim it or pad the others.

Rule 5: avoid "all of the above" and "none of the above"

They reward test taking tactics over knowledge. With "all of the above", a learner who is confident about two options can infer the answer without evaluating the rest. "None of the above" tells you a learner rejected four things, which is not the same as knowing the right one, and it makes the item impossible to learn from.

Both options also tend to appear when the writer has run out of ideas for a fourth distractor. That is the real signal, and the fix is a better distractor, not a filler.

Rule 6: mind the giveaways

Small habits leak the answer:

• Grammatical agreement. A stem ending in "an" that only fits one option hands over the answer.
• Absolutes. Options containing "always" or "never" are so often wrong that learners eliminate them on reflex.
• Repeated words. If a distinctive word from the stem appears in only one option, that option gets picked.
• Position bias. If your correct answers cluster in slot B or C, people notice. Randomize.

Rule 7: explain the answer

An MCQ with no explanation is a scoring device. An MCQ with a one line explanation is a teaching device, because a wrong answer becomes the moment a learner finds out what they misunderstood, which is the moment they are most able to fix it. Write the explanation as you write the question, while you still remember why the distractors are wrong.

If the question came from a specific point in a video or a document, tie the explanation to it. Sending someone back to the exact ninety seconds they missed is far more useful than sending them back to a fifty minute recording.

How many questions, and how hard?

For a comprehension check after a single video or chapter, eight to twelve questions is the practical range. Enough to cover the substance, short enough that people finish. For a graded exam across a course, twenty to forty spreads the weight so one unlucky item does not decide someone's grade.

On difficulty, aim for a mix. A handful of items most people should get right, a majority in the middle, and two or three that only someone who genuinely followed the material will get. A test everyone aces tells you nothing, and neither does one everyone fails.

Writing questions faster without writing them worse

The rules above are not hard to follow. They are just slow, which is why so many videos and training modules go out with no assessment at all. Twenty well built MCQs, written by hand from a recording, is comfortably an afternoon: rewatching, deciding what matters, drafting stems, inventing three plausible distractors each, and building the answer key.

The realistic way to keep the quality and lose the afternoon is to change what you start from. Generating a draft from the source material and editing it is a different job than writing from scratch, and it is a job you are good at, because judging whether a question is any good is far easier than inventing it. Paste in a recording and you can have a complete set of multiple choice questions with answers and explanations to react to instead of a blank page. If the material lives in a slide deck or a PDF rather than a recording, you can build the questions from the document the same way.

Then do the part only you can do. Cut the questions that test trivia. Sharpen the distractors using what you know about where your learners actually get confused, which is knowledge no model has. Fix the stem that reads two ways. That review pass takes minutes, and it is the difference between a quiz that measures something and a quiz that just exists.

A quick checklist before you publish

• Does the stem make sense with the options covered?
• Is there exactly one defensible correct answer?
• Would a learner who half understands the material be tempted by at least two distractors?
• Are the four options similar in length and structure?
• Does every question have an explanation attached?
• Are the correct answers spread across positions rather than clustered?
• Does the question set cover the whole source, not just the opening?

Seven checks, a couple of minutes, and the difference between a test people learn from and a test they game.