Key takeaways
- →A question teaches when someone commits to an answer they are unsure of, then finds out they were wrong - not when they recognise a fact they just heard.
- →Write the wrong answers first: the distractors are where the teaching lives, and each one should be a belief somebody in the room genuinely holds.
- →Aim for 60-80% of the room getting it right. Above 90% you wrote a recap; below 40% the question is ambiguous or the material was never taught.
- →Cut every clue that lets people answer without knowing: the longest option, the grammatical mismatch, the absolute wording, 'all of the above'.
- →The reveal is the teaching moment - show the split, name the most popular wrong answer and explain why it was tempting before you confirm the right one.
- →One quiz changes nothing on its own; re-ask three of the same decisions two or three weeks later or the whole session decays back to baseline.
To write quiz questions that actually teach, stop starting from the fact you want people to remember and start from the mistake they are currently making. A question built on a real misconception forces someone to commit to a belief and then find out it was wrong. A question built on a fact you covered on the previous slide only checks whether anyone was awake for it.
The mechanism behind this is retrieval practice. Pulling an answer out of memory under a little pressure strengthens it far more than reading the same answer again, and the strengthening is proportional to the effort. A question so easy that the answer arrives instantly does almost nothing. A question so hard that everybody guesses does almost nothing either. Everything useful happens in the band where people have to decide, and where about a quarter of them decide wrongly.
This guide covers the six question shapes worth using, how to mine misconceptions from your own team, why you should draft the wrong answers before the right one, the stem rules that stop people answering by pattern-matching, the difficulty band to aim for, how to run a reveal that teaches, and how to prune a question bank that has grown stale.
A scoring question and a teaching question are not the same object
A scoring question sorts the room. A teaching question changes what somebody believes on the way out. Most quizzes in workplace sessions are the first kind wearing the costume of the second: they produce a leaderboard, a bit of noise, and no change in behaviour on Monday.
The difference is almost entirely in the wrong answers. If three of your four options are obviously silly, the question tests reading speed. If all four are things a reasonable person might believe, the question tests understanding, and the distribution of answers tells you exactly which misunderstanding is loose in your organisation. That distribution is worth more than the scores.
Commitment is the other half of it. Locking in an answer - even privately, even anonymously - creates a small stake in being right. When the correct answer appears, the person who committed to the wrong one gets a correction with something attached to it. The person who watched the options go past without choosing gets nothing at all, which is why a quiz where only the confident people answer is not really a quiz.
Six question shapes and when each earns its place
Most teams reach for multiple choice for everything, which is reasonable because it scores instantly and works at any size. But the shape should follow the failure you are trying to fix. If people in your organisation do the right steps in the wrong order, an ordering question exposes that in fifteen seconds and a multiple-choice question never will.
Numeric estimates are badly underused. Ask a room how long the average support ticket takes to resolve, or what share of sign-ups finish onboarding, and the spread of guesses will teach you more about shared understanding than any factual question. Set a tolerance band up front - 'within ten per cent counts' - so people are not punished for being roughly right.
Open short answer belongs at the start of a topic, not the end. Ask it before you teach anything, read four responses aloud, and you have both a genuine picture of what the room believes and a set of ready-made distractors for the multiple-choice questions you write next time.
| Question shape | Teaches best when | Good group size | Failure mode |
|---|---|---|---|
| Misconception multiple choice | There is a wrong belief you can name | 5-500 | Distractors nobody would actually pick |
| Scenario judgement | The skill is deciding, not recalling | 5-200 | Stems get long and people stop reading |
| Ordering or sequence | A process is done out of order in real life | 5-100 | Fiddly on phones above five items |
| Numeric estimate | People are wrong about scale, not facts | 10-500 | Needs a tolerance band or it feels unfair |
| Two-option forced choice | You want speed and full participation | 5-1000 | Fifty-fifty guessing hides real confusion |
| Open short answer | You do not yet know what people think | 5-60 | Slow to read back and impossible to auto-score |
Start from the misconception, not the fact
Take a security induction. The factual question is 'how often must passwords be rotated?', and everyone gets it right because it was on the slide four minutes ago. The teaching question is a scenario where a well-written, correctly-branded email from a plausible internal address asks for an urgent payment change, because the misconception in the room is that phishing looks obviously fake. One of those questions changes what somebody does in March. The other produces a tick.
The pattern generalises. In a pricing rollout, the misconception is usually that discounts are negotiable at the same level they used to be. In a new expenses policy, it is that receipts under a threshold do not need uploading. In an engineering onboarding, it is that the staging deploy is safe to run without a review. Every one of those makes a far better question than the policy statement sitting next to it in the deck.
Write down the misconception in plain language before you write the question. If you cannot state it as a sentence somebody would actually say out loud - 'I assumed I could approve my own expenses under fifty pounds' - you do not have a question yet, you have a fact you are looking for somewhere to put.
Where to find the misconceptions
You already own the raw material. Support tickets and internal help-channel messages are the richest source, because every repeated question is a misconception with a timestamp. The last three sessions you ran will have produced questions from the floor that you answered off the cuff; those are misconceptions too, and they are usually the ones nobody wrote down.
Ask the most experienced person on the team a single question: what do new starters always get wrong in their first month? You will get four or five items in under two minutes, phrased in the exact language your distractors need. Audit findings, incident write-ups and the comments on the last policy announcement fill in the rest.
Keep the list somewhere durable and add to it as you go. A running misconception log is a question bank that writes itself, and it means you never again sit down at nine in the evening trying to invent ten questions from a slide deck.
Write the wrong answers first
Almost everybody writes the correct answer first and then pads out three wrong ones at the end, which is exactly backwards. The right answer is the easy part - you already know it. The distractors are the question.
List four beliefs people genuinely hold
Before you write the correct option, write the three or four things somebody in the room actually thinks. If you can only think of one plausible wrong answer, the topic may not be confusing enough to quiz on - or you have not talked to enough people about it.
Make every option the same shape and length
Options should look like siblings. Same grammatical form, same rough word count, same level of specificity. The single most common giveaway in workplace quizzes is a correct answer that is longer and more carefully qualified than the three around it, because the writer was concentrating hardest on getting it right.
Delete the joke option
The comedy answer feels like it lightens the mood, and it costs you a quarter of your diagnostic power. With one silly option a four-way question becomes a three-way question, and the people who most need the teaching now have better odds of guessing their way past it.
Cut 'all of the above' and 'none of the above'
Both let people reason to an answer without knowing the material: spot two options that are true and 'all of the above' follows. Neither tells you what anyone believes, which defeats the purpose of reading the results afterwards.
Check each distractor diagnoses something
For every wrong option, be able to finish the sentence 'someone who picks this thinks that...'. If you cannot, the option is filler. A distractor you can diagnose turns the results screen into a map of where the confusion sits, rather than a bar chart of noise.
Test it on someone who does not know the answer
Show the question to one colleague outside the topic and ask them to pick and explain. If they get it right by elimination or by spotting the tone of the wording, the question is leaking. Two minutes of this catches more problems than an hour of re-reading your own draft.
Rules for the stem
Top Tips
- Make the stem a complete question that could be answered with the options hidden. If someone has to read all four options to work out what is being asked, you have split one question into five and burned the timer doing it.
- Keep it under about twenty-five words for a live quiz. People are reading on phones with fifteen seconds on the clock, and every extra clause costs you responses from the back of the room rather than adding rigour.
- Avoid negative stems, and if you truly need one, capitalise the negative: 'which of these is NOT covered by the policy'. Under time pressure the word 'not' is the single most commonly missed word on a screen.
- Strip absolutes from the distractors. Options containing 'always', 'never' or 'all' read as wrong to any experienced test-taker, so they get eliminated on style rather than substance.
- Watch the grammar joins. If the stem ends with 'an' and only one option starts with a vowel, you have just given the answer away to everybody who noticed and nobody who did not.
- Never make the correct answer the longest option, and never let it be the only one with a qualifier like 'usually' or 'in most cases'. If it genuinely needs the qualifier, add matching hedges to two distractors.
- Give one piece of context, not three. Scenario questions fail when the writer adds a second character, a date and a budget figure that have nothing to do with the decision being tested.
A weak question, rewritten
The rewrite is not longer or cleverer. It moved from asking what the policy says to asking what the person will do on Friday, and it replaced two non-answers with two real beliefs. The original produces a 96% correct rate and no information; the rewrite typically splits three ways and shows you the exact gap in the policy rollout.
Notice that the rewritten options are all defensible. A quiz question where the wrong answers are embarrassing to have chosen will quietly push people towards not answering at all, particularly if scores are named. Plausible distractors keep the room honest because getting one wrong feels like a reasonable mistake rather than a public failure.
✗ Checks whether they read the slide
- •Stem: 'Which of the following is true about our expenses policy?'
- •A) We take compliance seriously
- •B) Receipts must be uploaded within thirty days of the spend, including under the twenty-pound threshold
- •C) All of the above
- •D) None of the above
- •Correct answer is twice as long as any other option
- •Two of the four options are not answers to anything
✓ Checks whether they will do it right
- •Stem: 'You buy a fifteen-pound taxi on Friday and lose the receipt. What do you do?'
- •A) Claim it - it is under the twenty-pound threshold
- •B) Claim it and note in the description that the receipt was lost
- •C) Request a duplicate receipt, then claim it with the duplicate
- •D) Do not claim it - no receipt means no claim
- •Every option is something somebody in the room currently believes
- •The answer split tells you exactly which belief to correct
Pitch the difficulty at the edge of what they know
Treat these as structural defaults rather than research findings: four options, a target correct rate between 60 and 80 per cent, twenty seconds for recall and forty-five for a scenario, and five to seven questions in a round. They are starting points that stop you writing a twenty-five question marathon at 95 per cent accuracy.
The correct rate is the number to watch. Above 90 per cent, the question is a recap and you should either retire it or raise the difficulty by strengthening one distractor. Below 40 per cent, one of two things is true: the question is ambiguous, or the material was genuinely never taught. Both are useful to know, and both are your problem rather than the room's.
Live results make this diagnosable in the moment. Session Flo shows the per-question breakdown as answers land, so you can see a question splitting 45/40/10/5 and decide on the spot that the next ten minutes are about the difference between the first two options. That is the whole value of running the quiz live rather than sending a form.
The reveal is where the teaching actually happens
Most of the learning value of a quiz is thrown away in the ten seconds after the timer stops. The typical reveal shows a green tick, awards the points and moves on, which teaches the people who were already right and abandons everybody else.
Write the explanation when you write the question, not in the room. One or two sentences, saved next to the question, covering why the popular wrong answer is wrong. Improvised explanations drift into re-teaching the whole topic, which is how a seven-question round takes forty minutes.
If a question splits close to evenly between two options, stop the round and talk about it properly. An even split is not a failed question - it is the single most informative result you can get, and it has just told you what the next part of the session should be about.
Close the answers and show the split first
Put the distribution on screen before the correct answer. The room reads it, sees that a third of them agreed with each other, and is now genuinely interested in which third was right.
Name the most popular wrong answer
'Forty per cent of you said claim it under the threshold.' Saying it out loud, without a note of surprise, makes being wrong a group event rather than a private one.
Explain why it was tempting
Say what makes the wrong answer reasonable before you dismantle it. This is the sentence that changes the belief; skipping it leaves people with a correction they have no reason to accept.
Confirm the answer in one sentence
One sentence, one reason, no slide. If the correct answer needs three minutes of justification, the question was carrying more content than a quiz item can hold.
Move on inside ninety seconds
Cap the whole reveal at about ninety seconds. Rounds die when question three turns into an eight-minute discussion and the room stops expecting to move.
Pacing, order and how many questions to run
Five to seven questions per round, two or three rounds in a session, with something else in between. A twenty-question block is where quizzes go to die: attention holds for roughly the first six, then answer speed rises, accuracy falls, and you end up measuring stamina rather than understanding.
Order matters more than people expect. Open with a question most of the room will get right - it establishes that answering is safe and gets the last few phones out. Put your hardest, most diagnostic question third or fourth, while attention is at its peak. Close on a scenario question rather than a recall one, because the last item is the one people carry out of the room.
Set the timer to the work, not to a uniform default. Twenty seconds is generous for 'which of these needs a receipt' and cruel for a four-line scenario. If you are not sure, watch the response curve: when 80 per cent of answers have landed and the count stops moving, you have your number for next time.
One quiz teaches nothing on its own
A single quiz at the end of a training session produces a satisfying spike and then decays. Retention falls off steeply in the days after a session, and a one-off test does not change that curve much on its own - what changes it is meeting the same material again after you have started to forget it.
So build the repeat into the plan rather than hoping for it. Re-ask three of the same decisions two to three weeks later, at the start of a session about something else. Keep the underlying decision identical and change the surface: different amount, different day of the week, different supplier. If people can only answer the version they have seen before, they memorised the item rather than the rule.
Track which questions people get wrong the second time. Those are the ones worth building a five-minute segment around, and they are usually far fewer than you feared. The rest can retire, which is how a question bank stays sharp instead of accumulating fifty items nobody has looked at since the first rollout.
Session Flo's recap makes the repeat cheap: the questions, the answer splits and the explanations go out after the session, so the second exposure has already happened before you re-run anything live.
Pruning a question bank that has gone stale
Run this pass over a bank once a quarter and it takes about twenty minutes. Most banks shrink by a third, and the third that goes is almost always the batch someone wrote in a hurry to hit a round number of questions.
The last item is the one people skip and the one that decides whether any of this was worth doing. A quiz that is never revisited is entertainment with a scoreboard attached. A quiz whose hardest three questions come back a fortnight later, in a slightly different form, is a training programme - and it costs you four minutes at the top of a meeting you were having anyway.
- ✓Every question maps to a decision somebody makes in their actual job
- ✓No question can be answered by reading the stem and options alone
- ✓The correct answer is not the longest, most qualified option
- ✓Every distractor is a belief you have heard somebody express
- ✓No 'all of the above', no joke option, no negative stem without capitals
- ✓Anything scoring above 90% correct twice running is retired or hardened
- ✓A one-sentence explanation is stored with each question for the reveal
- ○Three questions are scheduled to come back in two to three weeks
Frequently asked questions
How many options should a multiple-choice quiz question have?
Four is the practical default for a live session: it keeps guessing at 25 per cent, fits on a phone screen without scrolling, and is readable inside a twenty-second timer. Three works fine when you genuinely only have two plausible wrong beliefs - padding to four with a filler option is worse than having three real ones. Above four, reading time starts eating the clock and response rates drop at the back of the room.
How do I write good distractors for a quiz question?
Take them from things people have actually said. Support tickets, questions asked in past sessions, and the answers to an open text question you ran before teaching the topic are the three best sources. Test each one by finishing the sentence 'somebody who picks this thinks that...' - if you cannot, it is filler. Then match the length, grammar and specificity of all options so the correct one does not stand out by shape alone.
Should quiz questions be anonymous or named?
Anonymous when the questions are diagnostic and you want an honest picture of what people believe, named when the quiz is a social event and the stakes are low. The awkward middle is a compliance quiz with a public leaderboard: people answer defensively, guess towards the option that sounds most official, and your results stop measuring understanding. If you need both, run the diagnostic questions anonymously and the fun round with names on.
What correct-answer rate should I aim for?
Between 60 and 80 per cent per question is a good working band. It means most people can reason to the answer while enough get it wrong for the reveal to teach something. Consistently above 90 per cent and the question has become a recap - retire it or strengthen a distractor. Below 40 per cent, check the wording for ambiguity first, because a genuinely hard question and a badly worded one look identical in the results.
How long should a quiz be in a training session?
Five to seven questions per round, and no more than two or three rounds across a session, with teaching or discussion between them. Attention on a question block holds for roughly six items before answer speed rises and accuracy falls. If you have twenty questions you care about, split them across three sessions and re-ask the hardest few rather than running them all at once.
Can I reuse the same quiz questions with the same team?
You should, deliberately. Re-asking the same underlying decision two or three weeks later is one of the cheapest things you can do for retention. Change the surface details - the amount, the day, the supplier - while keeping the decision identical, so you find out whether people learned the rule or memorised the item. Retire anything they get right twice and keep building around the ones that still split the room.
Keep reading
- the evidence that retrieval beats re-reading — Why answering a question does more for retention than reviewing the same material.
- running a team quiz people actually enjoy — Once the questions are written, the format and pacing decide how it lands.
- choosing between open questions and multiple choice — When to collect free text instead of forcing a choice between four options.
- what the forgetting curve means for your training — The decay that makes the two-week repeat non-negotiable.
- build and run your quiz with Session Flo — Live scoring, per-question answer splits and a recap that repeats the questions.