20 Questions Game: Rules and Winning Strategy

In the 20 questions game, one player thinks of a person, place or thing, and everyone else gets twenty yes-or-no questions to work out what it is. The rules take ten seconds to learn. The strategy is the part almost nobody plays well: every question should cut whatever is still possible into two roughly equal halves. Twenty clean halvings separate more than a million possibilities. Twenty specific guesses rule out exactly twenty things.

That gap is the whole game. Most rounds are lost in the first five questions, when someone asks "is it a car?" and burns a question that could have eliminated half the world. Below are the rules as they are actually played, a clear line between this game and the completely different "21 questions" game people keep confusing it with, and an opening tree you can memorise once and use forever.

What are the actual rules of 20 questions?

One player is the answerer. They think of a single thing and keep it to themselves. Everyone else is a questioner.

1. The answerer picks something. The traditional constraint is a person, place or thing, announced at the start as animal, vegetable or mineral.
2. Questioners take turns asking questions that can be answered yes or no.
3. The answerer must answer truthfully. Lying ends the game and no one has fun.
4. A guess counts as a question. "Is it a stapler?" spends one of the twenty.
5. If a questioner names the thing within twenty questions, they win and become the answerer for the next round.
6. If twenty questions pass with no correct guess, the answerer wins and picks again.

Four house rules are worth settling out loud before the first question, because every argument mid-game comes from one of them.

Ruling	The two common versions
Non-yes-or-no answers	"Sometimes" and "kind of" allowed, or answerer must pick yes or no
Turn order	Strict rotation, or open floor where anyone can ask
Conferring	Questioners may discuss between turns, or must ask cold
Wrong guesses	Cost a question only, or knock that player out of the round

Strict rotation plus no conferring makes a harder, faster game. Open floor with conferring makes a better group game, because the quiet players get carried along instead of sitting out four turns at a time.

How is it different from the "21 questions" get-to-know-you game?

These are two unrelated games with confusingly similar names, and most of the lists you find for one are written for the other.

	20 questions	21 questions
Goal	Identify a hidden thing	Learn about another person
Who answers	One person, about their secret thing	Everyone, about themselves
Question type	Strictly yes or no	Open-ended, anything goes
Prepared questions	None, you invent them from the answers	Yes, usually a written list
Winning	Guess it within twenty	There is no winner
Typical setting	Car, dinner table, waiting room	Date, sleepover, group icebreaker

If you landed here wanting "20 questions to ask your crush", you want the second game, and the questions are the entire product. Our roundup of party question games for groups covers that side properly, with sets written for rooms of different sizes.

The guessing game needs no question list at all, which is exactly why it survives long car journeys and power cuts. What it needs is a method.

Why does halving the field beat guessing?

A yes or no answer carries one bit of information, the smallest unit information theory deals in. One bit divides the possibilities in two. Twenty bits divide them in two twenty times over, and two multiplied by itself twenty times is 1,048,576. So twenty genuinely balanced questions can pick one item out of a field of more than a million.

That upper bound only holds when each answer really does cut the field in half. Computer science uses the same trick under the name binary search, where repeatedly halving a sorted list finds any entry in a handful of steps rather than by checking items one at a time. Checking items one at a time is precisely what a question like "is it a banana?" does. It takes a field of a million and leaves 999,999.

So there are two kinds of question, and they are not close in value:

• Splitting questions target a feature that lots of candidates share. "Is it alive?" "Did people make it?" "Is it bigger than a microwave?" Each one is worth up to a full bit.
• Guessing questions target one candidate. "Is it a giraffe?" Each one is worth almost nothing until the field is already tiny.

Real games never split perfectly, so treat 1,048,576 as a ceiling rather than a plan. A more useful working target: be inside a category of a few dozen by question twelve, and start naming candidates only after that.

There is one exploitable bias too. Answerers do not pick uniformly from everything in the universe. They pick things that are quick to say, easy to rule on, and already in the room or in the conversation. That is why "is it alive?" is a better opener than "does it exist?", and why "is there one in this house?" is stronger than its raw share of the world suggests.

What are the best opening questions?

Here is the tree I use. Five levels, and every branch at every level is a question that a reasonable answerer would say yes to about half the time. Memorise the shape, not the wording.

• Q1. Is it a physical thing you could photograph?
  • Yes (people, animals, objects, places)
    • Q2. Is it alive now, or was it ever alive?
      • Yes (people, animals, plants, food)
        • Q3. Is it human?
          • Yes → Q4. Is it one specific real person? → Q5. Are they alive today?
          • No → Q4. Is it an animal rather than a plant? → Q5. Is it bigger than a house cat?
      • No (objects, places, materials)
        • Q3. Did people make it?
          • Yes → Q4. Would it fit in a backpack? → Q5. Is there one in most homes?
          • No → Q4. Is it a specific place you could travel to? → Q5. Is it somewhere in North America?
  • No (ideas, activities, events, titles)
    • Q2. Is it something people do, rather than something they think or feel?
      • Yes (jobs, sports, activities, events)
        • Q3. Do people get paid for it? → Q4. Does it need special equipment? → Q5. Have most adults done it at least once?
      • No (concepts, feelings, works)
        • Q3. Does it have a title, like a book, film or song? → Q4. Is it older than everyone playing? → Q5. Would everyone in this room know it?

What that buys you, level by level:

After question	Share of the field still live
1	one half
2	one quarter
3	one eighth
4	one sixteenth
5	one thirty-second

Five questions in, you are inside one of thirty-two regions and you have fifteen questions left for a space that is now small and, more importantly, describable. Questions 6 to 13 narrow attributes inside that region: material, colour, where you would find it, what it is used for, whether it plugs in. Only from about question 14 do named guesses start paying, because by then one candidate is a real fraction of what is left rather than a millionth of it.

The single most common mistake is jumping the tree. Someone hears "yes, it is man-made" and immediately asks "is it a phone?", which throws away four levels of free narrowing to chase a hunch. Hunches are fine. Ask them at question 15.

How do you pick a good answer word as the thinker?

Being the answerer is not a passive job, and a badly chosen answer ruins the round for everyone.

Pick something everyone playing has definitely heard of. Your own dog, a colleague from a job nobody else had, or a band with four listeners is not a challenge, it is a dead end. The point is to be hard to find, not impossible to know.

Pick from a large, flat category. Household objects, common animals, foods, jobs and famous landmarks all work, because they contain thousands of members that split cleanly. The best answers are the ordinary items nobody would name first: a doorstop, a paperclip, a traffic cone, a shoelace.

Avoid anything that will need a ruling. If you cannot answer "is it alive?" without saying "well, sort of", pick again. Leather, yoghurt and a dead celebrity all generate arguments. Save them for a group that enjoys arguing.

Fix it before the first question. Say it silently, and if you are prone to drift, type it into your phone. Answerers unconsciously slide their answer toward whatever is still consistent, and everyone can tell when it happens.

Test it with one sentence. Could you answer twenty yes-or-no questions about this without ever hedging? If yes, it is a legal answer. If it also makes you smile to imagine someone hunting for it, it is a good one.

How do you play it in a car or with young kids?

The car version has one rule change: the driver should be the answerer, not a questioner. Answering is a single word at a time and costs almost no attention, while questioners are the ones doing the thinking and the arguing. That also solves the usual car problem of one person reading from a phone while everyone else waits.

For young children, the research is genuinely useful here. Studies of children playing 20-questions style games separate constraint-seeking questions from hypothesis-scanning questions, the first targeting a feature many objects share and the second targeting one object directly. The pattern in that literature is consistent: preschoolers almost always ask hypothesis-scanning questions, and the shift toward constraint-seeking questions comes with age. The same work found that giving children the feature words they needed, rather than training them at length, was enough to improve the questions they asked and cut the number they needed.

So the fix for a frustrated six-year-old is not "ask better questions". It is to hand them the categories out loud before you start: "you can ask if it is an animal, if it is food, if it is something in the kitchen, if it is big or small". Three practical adjustments:

• Shrink the field on purpose. "I am thinking of an animal" or "it is something in this room" turns an impossible search into a winnable one.
• Score questions used, not wins. Got it in fourteen last time? Try for twelve. That rewards the halving instinct directly.
• Let them be the answerer often. Choosing something and holding it steady through twenty questions is its own skill, and it is the easier half.

Two players works fine, by the way. Nothing in the game needs a third person, which is why it holds up in a car with two seats occupied or a queue with one friend.

Need prompts rather than a guessing game for the drive? The Road Trip Questions deck is free, needs no account, and deals one card at a time so the passenger can run it on a thumb. The 45 road trip questions behind it are grouped by how far into the drive you are. If you are building a games night rather than a car journey, the trivia night guide has the running order and the scoring.

Frequently asked questions

How do you play 20 questions?

One player thinks of a person, place or thing. Everyone else asks yes-or-no questions, taking turns, up to twenty in total. The answerer must answer truthfully, and a guess at the answer counts as one of the twenty. Name it correctly within twenty and you win the round and become the next answerer. Fail, and the answerer wins and goes again.

What are good things to pick for 20 questions?

Ordinary items from big categories that everyone knows but nobody would name first. A traffic cone, a paperclip, a stapler, a penguin, a lighthouse, a dentist, a trampoline. Avoid anything personal to you, anything most players would not recognise, and anything that would force you to answer "sort of" to a fair question.

What's the best first question in 20 questions?

"Is it a physical thing you could photograph?" or, if you prefer the classic, "is it alive, or was it ever alive?" Both split the field close to evenly and neither depends on knowing anything about the answerer. The worst first question is any question naming a specific object, because it eliminates one candidate instead of half of them.

Can you play 20 questions with two people?

Yes, and it is one of the few group games that loses nothing at two. One person answers, the other asks all twenty, then you swap. With no turn order to manage, a two-player round runs faster, which makes it a good fit for a car, a queue or a walk.

Sources

• Azzurra Ruggeri, Caren M. Walker, Tania Lombrozo and Alison Gopnik, How to Help Young Children Ask Better Questions?, Frontiers in Psychology, January 12, 2021
• Paul E. Black, binary search, Dictionary of Algorithms and Data Structures, National Institute of Standards and Technology
• Encyclopaedia Britannica, Information theory