How to write good quiz questions

When a question checks two things at once, a wrong answer is impossible to interpret. Did the person not know the first fact, the second, or how they fit together? True-or-false items are especially exposed: a statement with two claims is false as soon as one of them is, so someone who knows half the content can get it wrong while someone who guessed gets it right.

Before writing, name the single point the question should check, in a few words: “where photosynthesis takes place”, “when to report a suspicious e-mail”. If you cannot name it, or you need two names, split the question. Two short questions take less time to read than one tangled one, and each tells you something precise.

The stem is the question itself, before the answers. A useful test: cover the options and read the stem alone. Someone who knows the subject should be able to answer it in their head. If they cannot, because the stem is only a topic (“The French Revolution:”) or a vague “Which statement is true?”, the real question is hidden in the options.

Put what people need in the stem, once, and keep the options short. State the conditions that make the answer correct, such as “at sea level” or “on a company laptop”. Prefer positive wording. If a negative is essential (“Which of these should you not do?”), keep it rare, make the word stand out, for example in bold, and never stack two negatives. Then read the question aloud: if you stumble, so will someone reading it on a phone.

A wrong option only does its job if someone who has not mastered the point could reasonably choose it. “A kangaroo” as the capital of Australia gets a laugh, but it turns a four-option question into a three-option one, or fewer. Plausible options come from real confusions: the larger city that is not the capital, the step people often swap. How to make each wrong answer reveal a specific misconception is covered in our retrieval-practice guide.

You need fewer options than you might think. A 2005 meta-analysis by Michael Rodriguez, covering 80 years of research on multiple-choice items, concluded that three options are generally optimal, and Andrew Butler’s 2018 review of multiple-choice testing also recommends three plausible options. In Pomlet a choice question can have two to six answers: a fourth one is worth adding only if it is as credible as the others.

Make the options look alike: same grammatical form, same kind of thing (all years, all cities, all actions), similar length and level of detail. Put numbers and dates in order, and check that no two options overlap.

Experienced test takers know a few tricks. The longest, most carefully qualified option is often correct. “All of the above” is a good bet. An option that repeats a word from the stem is suspicious, and sometimes the grammar of the stem fits only one answer. When these tricks work, the question measures test-taking skill rather than knowledge.

Such flaws are common and distort results. In ten exam papers taken by nursing students at a Hong Kong university, Marie Tarrant and James Ware (2008) found at least one writing flaw in 47.3% of the questions. Re-scored on the flawless questions only, the papers gave a slightly lower pass rate (90.6% instead of 94.3%) but more scores of 80% or above (20.9% instead of 14.5%): the flaws had penalised high achievers more than borderline students. Butler’s review advises avoiding “all of the above” and “none of the above”, as well as complicated formats such as “A and B, but not C”.

This one gives itself away twice: “Why lock your screen when you leave your desk?” – “Habit”, “Company policy”, “All of the above”, “So that nobody can read your files or send messages in your name while you are away”. Better: “So nobody can use your logged-in computer”, “To save battery”, “To install updates faster”.

A quiz where everyone gets everything right is pleasant but tells you little; one where most people fail mostly discourages them. Plan the curve: open with one or two questions almost everyone can answer, so people settle in and check that their device works, build towards a few harder ones, and finish with something achievable. In a revision quiz before an exam, this curve gives the least confident learners an early success.

Difficulty should come from thinking, not from obscurity. Butler recommends questions that require real cognitive work, such as comparing two concepts or applying a rule to a situation, and warns against trick questions. A detail nobody needs is hard but pointless; applying what was learned is hard and worthwhile. On a quiz night, reward reasoning or a good estimate rather than memorised trivia.

In Pomlet each question can award standard points, double points or no points at all: double the score of one or two demanding questions, and set a warm-up or opinion question to “No score”. When a question needs reasoning, give it a longer time limit, or “No time limit”, and consider switching off the speed bonus, which is on by default and gives more points for faster answers.

The format should follow the thinking you want, not the other way round. Pomlet offers 11 elements; here is when each one fits, and what to watch for. Note that in Pomlet “Multiple choice” means several correct answers; a question with exactly one correct answer is “Single choice”.

Each question can carry an optional explanation of up to 300 characters, shown after the answer. Write it at the same time as the question, not the evening before the session: one or two sentences on why the right answer is right and, if there is room, why the most tempting wrong answer is not. If it does not fit, the question is probably too broad: writing the explanation is the quickest way to catch a flawed question. When to show feedback is covered in the retrieval-practice guide.

Images, audio and video can carry a question or distract from it. Use an image when people need it to answer (a map, a graph, a painting) or when it clarifies the situation, and check it on a phone; answer choices can be images too. Pomlet accepts uploaded audio (up to 20 MB) and video (up to 30 MB), played as excerpts of up to 120 seconds with fades; in a live game the sound plays on the host screen, and participants can opt in on their own device. Check that you are allowed to use each picture or recording in your context.

Before running the quiz, check each question against this list; it also works on drafts from the AI quiz generator, which always need a human review.

Then try it. In the editor or on the launch page, “Test” runs the whole session with simulated participants – timers, answers, ranking and report – and nothing is saved: ideal for typos and explanations. To check the real flow on real devices (phone screen, sound), launch a session and ask a colleague to join with the PIN and think aloud: where do they hesitate, which option tempts them, could they find an answer without knowing the subject? A colleague who knows the topic spots wrong answer keys; one who does not spots giveaway cues.

After the first real session, open the report. A question that almost everyone got right may be too easy or may leak its answer; one that split the room unexpectedly may be ambiguous. Rewrite it before the next use.