How to write quiz questions that hold up in front of a group

The failure that ruins a room is not a hard question, it is an arguable one. "What is the biggest country in Europe?" has one answer by area, a different one by population, and a third if you exclude Russia, and whichever you meant, somebody in the room meant another. The argument that follows is not fun; it is a group discovering that the game is unfair.

So before anything else, ask whether a reasonable person could defend a different answer than yours. If they could, add the qualifier that removes the ambiguity ("by area", "as of 2024", "in the original novel"), or pick a different question. A question that needs three qualifiers to be fair is usually a question that should be about something else.

The same rule catches dated facts. Populations, record holders, "the latest" anything, and the current holder of any office all rot. If you must ask about them, anchor the question to a year, and expect to revise it. The questions in the bank here that get reported most often are almost all of this kind: right when written, wrong by the time someone played them.

Multiple choice is the default: a question, four options, one correct. It is fast to answer, fast to score, and works on a phone with one thumb, which is why it carries most quiz formats. Its weakness is that the options do half the work for the player. A question nobody would know cold becomes a coin flip between two plausible options, and a badly written one becomes a giveaway.

An estimation question asks for a number and scores by distance: how tall, how many, how far, what year. There are no options to eliminate, so it tests whether somebody has a sense of the scale of a thing rather than whether they have memorised it. The scoring here gives full points inside a tenth of the range you set and tapers to zero beyond it, so the range matters as much as the answer. A range of 0 to 10,000 for "how many bones in the adult human body" (206) makes any guess between 0 and 1,000 score the same; a range of 100 to 400 makes it a real question.

A good set mixes them. Two or three estimation questions in a round of twelve change the rhythm, and they are the questions people talk about afterwards, because everybody was wrong by a different amount. The estimation guide goes into writing them in more depth.

• Estimation questions: how they score and how to write them

The wrong options are where most questions are actually won or lost. "Which planet is closest to the sun? Mercury, Venus, a banana, Thursday" is not a question; it is the answer with decoration. The test for an option is whether somebody who does not know the answer would consider it. Three options that survive that test make a question worth asking.

A few reliable ways to build them. Use the same category as the answer: for a capital city, three other capitals; for a year, three other years within the same decade or two. Use the common misconception, the thing people actually get wrong, because that is the option a room will split over. Avoid one option being much longer or more specific than the rest; players learn that the detailed one is the right one, and the answer data confirms it — questions where the correct option is the longest are answered correctly far more often than their difficulty label says they should be.

Two small tells to remove: do not let the correct answer be in the same position every time (the editor shuffles options in play, but write them shuffled anyway), and do not let grammar give it away. If the question ends in "an" and only one option starts with a vowel, the option is the answer.

A round here runs a clock, usually twenty seconds. That time is for thinking, not for reading. A question of thirty words with a subordinate clause spends half the clock before anybody has got to the options, and the players on a phone get to the options last. Write the question as one sentence. If it needs context, cut the context or move it into the answer reveal, where people have time.

Read it aloud. A question that stumbles when spoken stumbles on screen. If you cannot say it in one breath, it is too long; if a word in it could be replaced by a shorter one, replace it. "Which of the following rivers is the longest in the continent of Africa?" is "Which is Africa's longest river?" with nothing lost.

A set where every question is hard is a set where every score is low and nobody feels good; one where every question is easy is a typing contest. The shape that works in a room is a curve: open with two or three that most people get, so the game feels playable; build through the middle; put the hardest two near the end, where a lead can still be lost; and finish on one that most people get, so the last feeling is a good one.

The labels matter because players read them. "Easy" means a majority of the room gets it. "Hard" means a minority does. When the label and the outcome disagree, the question feels broken. The answer data here is blunt about this: some questions labelled easy are answered correctly by under a third of players, and they draw more thumbs-down than genuinely hard questions do, because the label set an expectation the question did not meet. If you are not sure, label it medium and let the room tell you.

• The questions players get wrong most often

Every answer given in a room here is recorded against its question, which makes it possible to see which questions behave. A few patterns recur across the ones that fail.

Trick questions fail. A question whose answer depends on noticing a word ("Which of these is NOT a mammal?") is answered wrong by people who knew the facts and read fast, and they resent it. If the trick is the point, make the trick word unmissable; better, ask the question straight.

Two-right-answers questions fail slowly. They are not noticed on writing, they are noticed in play, and they show up in the data as a question with a "wrong" option that a large share of players choose. When one wrong option draws more than the correct one, the question is almost always at fault, not the players.

Questions that require the current year fail eventually. Questions with a number in the answer and a range in the options ("between 1900 and 1950") fail because the range is the giveaway. And questions that need a picture and do not have one fail because they describe the picture in the question text instead, which takes the whole clock to read.

Before a set goes in front of a group, give it to one person who has not seen it and watch them play it. Not read it: play it, with the clock. You will find the question they misread, the option they laughed at, the one they got right by elimination, and the one they got wrong because they knew more than you did. Fix those and the set is usable.

Then play it once in a room and read the results. The question everybody got right in under five seconds is easy regardless of its label; the one that split the room evenly across two options is either excellent or broken, and the reveal will tell you which. Each question on this site has a thumbs-up and thumbs-down, and a report button for a wrong answer; on your own quizzes, that feedback comes back to you. Two thumbs-down on a question is noise. Ten is a question to rewrite.

• Write a set in the editor
• Round formats to put the questions in