Go Fish: everybody's first card game, and nobody's first memory game
Go Fish is usually somebody's introduction to playing cards at all, which is why it has a reputation as a game of pure luck for six-year-olds. That reputation is wrong, and it's wrong in an unusually measurable way. Every question asked at a Go Fish table is a public confession — you may only ask for a rank you are already holding, so "do you have any sevens?" tells the entire table that you have a seven. Every answer is public too.
A player who keeps track of that ends up knowing, with certainty rather than suspicion, which cards several opponents are holding. Before building the computer opponents on this page, each piece of common Go Fish advice was played out as a bot and measured against a rival sitting at the same table. One idea turned out to be worth more than everything else in the game combined, and two pieces of often-repeated advice turned out to be worth nothing at all.
- Memory wins
- 91.7%
- How often a bot that tracks every question beats an identical bot asking at random.
- Free certainties
- 35%
- Share of turns at a four-player table where a question that cannot miss is available.
- Books, always
- 13
- Every game makes all thirteen. Go Fish can never end early — see below.
From 18,000 simulated games used to check this table's rules engine. Method and full figures further down.
Where Go Fish comes from
Go Fish sits in the family of games catalogued as quartet games: collect four of a matching set by asking other players for them. The two ancestors usually named are American and British and arrived within a few decades of each other in the nineteenth century.
The American one is Authors, played with a purpose-printed deck in which each quartet is four works by the same writer — a parlour game with a self-improving veneer, of the sort the mid-1800s produced a great deal of. The British one is Happy Families, whose cards were published by Jaques of London and shown at the Great Exhibition of 1851; there the quartets are families of four, each with a trade, and the cards' comic illustrations are as much the point as the game. Germany has the same structure as Quartett, still sold today as themed children's sets on everything from locomotives to dinosaurs.
Go Fish is what you get when you strip the special deck out and play the same game with the fifty-two cards already in the house: the quartets become the four suits of each rank. Where and when the name attached is genuinely murky — American card books were describing it as Go Fish or simply Fish by the middle of the twentieth century, and the various explanations offered for the phrase are folk etymology rather than evidence. It is one of those games that was clearly played long before anybody bothered to write it down.
The objective
Collect four of a kind — all four sevens, all four kings — and lay them face down in front of you as a book. Not a pair: the standard game collects all four suits of a rank, though pairs are a well-established simplification for young children and are covered further down. There are thirteen books in a deck, the game runs until every one of them has been made, and whoever has the most wins. There are no points, no penalties, and nothing to lose by holding cards at the end; the only currency is completed sets.
Setup: how many cards to deal
A standard 52-card deck, no jokers. The rest of the deck goes face down in the middle as the stock — traditionally called the pond, the ocean, or just the pool. Play normally starts to the dealer's left; here the starting seat is picked at random, which turns out not to matter. If a dealt hand already contains four of a kind, it goes straight down as a book before anybody plays.
How a turn works
- Pick a rank you are already holding and ask one other player for it: "Do you have any queens?" You may not ask for a rank you have none of. This is the rule the whole game hangs on.
- If they have any, they hand over every single one — not one card, all of them — and you take another turn. You may ask the same player again, a different player, for the same rank or a different one.
- If they have none, they say "go fish" and you draw the top card of the pond.
- If the card you fish up is the very rank you asked for, show it, and take another turn. If it isn't, keep it and your turn ends.
- Four of a kind goes down immediately as a book, whether you completed it from a handover, from the pond, or in the deal.
The three rules people get wrong
Almost every argument at a Go Fish table is one of these three, and all three are the same rule wearing different hats: the game is designed to leak information.
- You must hold the rank you ask for. Not "you may ask for anything you like." Without this rule there is nothing to deduce and no game — a player could fish blindly for whatever they felt like, and asking would tell the table nothing at all. Plenty of families play the loose version without realising they've removed the skill.
- They give you all of them. If you ask for jacks and they have three, all three come across. This is what makes a lucky question so valuable and why hits tend to arrive in clumps.
- A hit means you go again. So does fishing up the exact card you asked for. A turn is not one question — it's a chain that runs until you miss, which is why a player who knows where a card is can sometimes take three or four in a row.
Every question at a Go Fish table is a confession. You cannot ask for sevens without telling everyone you have a seven.
How the game ends and who wins
The game is over when all thirteen books have been laid down, and the player with the most books wins. That's the whole scoring system. With three or more players a tie is possible and happens fairly often — a tie came up in about 11% of simulated three-player games and 18% of four-player ones. With exactly two players it is impossible: thirteen books is an odd number, so somebody always has more.
What actually wins games
The strategy advice attached to Go Fish is mostly repeated rather than tested. Each idea below was built as a computer opponent and played against a rival in a symmetric seat at the same table, sharing every deal and every opponent, which strips out most of the luck of the cards.
Same-table A/B runs, 1,500 four-player games each. Head to head, the bot that remembers beats the random asker in 91.7% ±1.1 of decided games and the odds-only bot in 85.0% ±1.4. Knowing the odds is worth +0.30 books over asking blind; adding memory on top is worth a further +1.57.
That gap is the whole argument against Go Fish being a luck game, and it points straight at how to play well.
- Never ask for something you know they don't have. The cheapest habit to build, and the one most casual players never start. A "go fish" is a permanent fact about that player until they draw a card — and once the pond runs dry nobody draws again, so in the endgame the table's knowledge only ever accumulates. The last third of a game should be played almost entirely off things you already know.
- Certainties come up more often than you'd expect. A question that literally cannot miss — you hold that rank, and somebody has already admitted to holding it too — is on offer surprisingly often, and more often the bigger the table, because more players means more questions in the air.
- When you're guessing, ask the biggest hand. If nothing is certain, the player holding the most cards is the most likely to hold any given rank. A small edge, but a free one.
A bigger table is a more skilful game, not a luckier one: at four players more than a third of all turns can be played on certain information.
Two pieces of advice that measured as nothing
Both of these are repeated constantly, and both survived a first round of testing before larger samples killed them. They are included here because a strategy section that only reports the ideas that worked is not much of a test.
- "Don't tip your hand" is worth nothing. The most commonly repeated piece of Go Fish advice is that asking for a rank you hold three of announces you are one card from a book, so you should ask for your singletons instead and keep quiet. Tested at the same table over 3,137 decided games, the quiet bot and the bot that happily asked for its strongest ranks finished at 50.9% ±0.9 — a difference of 0.03 books a game. The reason is that the "warning" only matters to an opponent who actually holds one of those cards, and an opponent who holds one was going to be asked for it, or ask for it, soon regardless. Ask for whatever gives you the best chance of a hit.
- Going first is not an advantage. Across 6,000 games at each player count, the opening seat won its fair share to within about one standard error every time. Worth knowing mostly so nobody argues about the deal.
About the computer players
The opponents here play off a record of what has been said at the table — every question, every handover, every "go fish", every book — and nothing else. They never look at your hand and they never look at the pond.
But a bot with perfect recall against a human trying to hold forty questions in their head would be an unfair fight dressed up as a fair one, so the board simply shows you the same record the bots are using: under each opponent, the ranks they are provably holding and the ranks they provably don't have, kept up to date as the table learns and — when somebody draws an unknown card from the pond — as it forgets. A rank marked in gold is one you hold too, which makes it a question that cannot miss.
Watching that panel fill up is the fastest way to understand what good Go Fish actually looks like. Everything in it was deduced from things said out loud in front of you.
Playing it with young children
Go Fish is one of the few genuinely good first card games, and it's worth adapting rather than simplifying. A few things that help:
- Play for pairs instead of books with children under about five. Two cards is a much shorter wait for a reward than four, and the asking mechanic is identical. This is far and away the most common house rule in the game.
- Deal fewer cards. Five is plenty; a seven-card fan is hard for small hands to hold and harder to read.
- Say the rank out loud every time. The whole educational value of the game — matching, counting, and the beginnings of remembering what other people told you — lives in the spoken question.
- Don't correct the memory part. Children work out on their own, and usually quite suddenly, that they can just remember what was asked. It's a genuinely nice thing to watch happen and prompting it spoils it.
Common variations
Go Fish is a folk game with no governing authority, so the rules vary from kitchen to kitchen. The ones you're most likely to meet:
- Pairs, not books. Two of a kind scores. Faster, luckier, and the standard for very young players.
- Ask for anything. Some tables drop the must-hold-the-rank requirement. It makes the game friendlier for children who can't yet read their own hand reliably, at the cost of removing every deduction in it.
- One card, not all of them. A rarer variant in which the asked player surrenders a single copy. It slows the game considerably.
- No extra turn for a catch. Fishing up the card you asked for is just good luck, and your turn ends anyway. Common, and it shortens games noticeably.
- Draw a whole new hand. Instead of drawing one card when you run out, you draw a full five or seven. Keeps everybody in the game longer and makes the pond disappear much faster.
- No pond at all. With four or more players, some deal the entire deck out. There is then nothing to fish from — a miss simply ends your turn — and the game becomes pure deduction with no luck of the draw at all. This is the version closest to Literature.
- First one out wins. Rather than counting books at the end, the player who empties their hand first wins outright. It rewards making books quickly rather than making the most of them.
- Authors and Happy Families. The same game on a printed deck of named sets. If you have a set in the house it plays identically — and the named quartets are considerably easier for a child to keep track of than "the fours".