How to win at GuessTheSearch: a strategy guide to guessing autocomplete

GuessTheSearch looks like a trivia game, but it isn’t one. There is no fact to recall. The answer to every question is simply whatever Google suggests first when you type the sentence, and that makes it a game about predicting other people. The good news is that other people are more predictable than they think. After a few hundred rounds of looking at real captures, we have noticed patterns that turn a coin flip into an educated guess, and this guide collects them.

If you have never played, start with How to play. For the background on where suggestions come from, read how Google autocomplete works. Every example below is a real capture from September 2026, in US English, shown in the order Google ranked it.

Rule one: the answer is what people type, not what is true

The single most common way to lose a question is to answer it correctly. Autocomplete ranks what people search for, and people mostly search for things they are unsure about. So the #1 completion is often a misconception, because a misconception is exactly what sends someone to a search box.

The true answer, crustaceans, sits in last place. Nobody who already knows crabs are crustaceans needs to type the question; the people typing are the ones who looked at a crab’s legs and thought “spider?”. The same thing happens with “are hyenas ___”, where the #1 is dogs (hyenas are actually more closely related to cats and mongooses), and with “hippos used to be ___”, where the #1 is whales. Hippos were never whales; the two share a distant common ancestor, which is a far less searchable sentence.

Read the grammar before you read the options

Every question is one of eight sentence shapes, and each shape pulls a different kind of word out of the internet. Decoys always come from the same shape as the answer, so grammar will not eliminate options on its own — but knowing what a shape wants tells you what the winner usually looks like.

A few habits of each shape are worth memorising. “Why is … always” is where complaints live: Microsoft is always asking for my password, printers are always so difficult. “Is” questions about places are dominated by safety and geography (“is Cape Town safe”, “is Oslo the capital of norway”), and questions about food by health (“is beer good for you”). “Used to be” is the shape where one word does the heavy lifting: in our September capture, called appeared in 40 of the 44 “used to be” lists and was the #1 in 14 of them, because people want old names. If no option is a genuinely surprising former state, “called” is the percentage play. “Will … ever” leans towards the hopeful rather than the catastrophic: “will Dubai ever recover”, “will Costco ever bring back churros”.

“vs” wants a rival — but not always the one you expect

The “vs” shape looks easy: name the obvious competitor. Often that works (Domino’s vs pizza hut, Revolut vs wise). But people compare across categories and across history, and the famous rivalry can beat the commercial one.

The #1 isn’t a carmaker at all. It is the nineteenth-century feud between Nikola Tesla and Thomas Edison, a story people keep returning to. Similarly, “apple vs” is won by android, a platform, rather than Samsung, a company; and further down the list the word “apple” drifts into dog breeds and body shapes. When a subject’s name means more than one thing, expect the list to wander.

Self-searches: the internet talks about itself in the first person

First-person completions are rare as a #1 overall, but in two places they take over: complaints about pests and complaints about personal gadgets. People don’t search those in the abstract; they search about what is happening to them right now.

Six out of six are personal. The pattern repeats across themes: crows are always around me, AirPods won’t stay in my ears, and PayPal, Revolut and Meta all appear with “need my ssn” or “want my id”. In a question about insects, birds or earbuds, the option that reads like something a frustrated person would type on their phone deserves a second look.

Length and phrasing tells

Completions are compressed. Real searches drop articles, skip apostrophes and stay short, so the winner is usually one to three plain words. A few reliable tells:

• Shorter and broader beats longer and specific. “Why do printers never work” outranks “only have 4 colors”. Specific questions are asked by fewer people.
• The generic complaint beats the technical one. “Why are smart TVs…” is led by slow down, not have cameras.
• “Anymore” signals a change people noticed, and those are often high: “why don’t iPhones come with chargers anymore” ranks second.
• Search-box spelling is a clue, not an error. “fly to usa”, “jersey mike”, “in my yard” — lower case, no apostrophes, no articles. An option written like a headline is less likely than one written like a thumb.

How Easy, Average and Hard differ

The answer is always Google’s #1. Difficulty only changes where the three wrong options come from, and the five questions stay identical at every setting.

• 🌱 Easy uses completions that belong to other subjects in the theme. If you recognise which options could plausibly belong to this subject, you are most of the way there.
• ⚖️ Average mixes the subject’s own lower-ranked completions with a few borrowed from elsewhere in the theme. Spot the borrowed ones first; that halves the field.
• 🔥 Hard uses the subject’s own #2, #3 and #4. Every option is a real search about this exact subject, so knowledge of the topic is useless on its own — you have to judge volume.

On Hard, this question would show the top four. The strategy is to rank them by how many people could plausibly type each one. “Never work” and “suck” are available to every printer owner on earth; “use usb b” requires noticing a specific cable. Between the two broad ones, the milder, more descriptive phrasing won here. The general lesson: the vaguest complaint usually wins, and here the blunter synonym came a close second.

On Easy and Average, one more rule helps: below Hard, the game removes any decoy that would be genuinely true of the subject, so you will never face two right answers.

Choosing a time period

Suggestions move with the news, so the same question can have a different answer depending on when it was captured. The four periods read different slices of our daily archive:

Last Week is the best place to build instincts, because the consensus smooths out one-day spikes. Last Month is a single day, so it can catch a news event mid-surge; if you remember what was going on around then, you have an edge. Last Year rewards people who follow a subject over time: product launches, elections and sporting tournaments all reshape lists. All Time is the oldest snapshot we hold.

One honest caveat: a period the archive can’t reach yet falls back to our offline answer set, rotated so each period is still a different puzzle, and upgrades to real captures as history accumulates. That offline set is real captured autocomplete for every shape except the “so” shape outside the Stereotypes theme, which is hand-written. Our editorial standards explain the distinction, and the puzzle archive lets you replay any past day.

Protecting your streak

A streak advances once per day, and only for today’s puzzle in any theme. Replaying the archive never inflates it, and playing a second theme doesn’t count twice. Miss a single day and the streak shield forgives it automatically — once per calendar week. A second miss in the same week resets the streak, though your longest streak is kept forever. The practical advice: if you are going to skip a day, skip one, and don’t skip another before the week turns over.

Duels: speed matters as much as accuracy

Live 1v1 duels put two players on the same seven questions at the same moment, with ten seconds per question. A right answer scores between 1,000 (instant) and 500 (at the buzzer), falling in a straight line in between, so a correct answer after five seconds is worth 750. Wrong or late scores nothing. The answer is always Google’s real #1.

Do the arithmetic and a strategy appears. Seven correct answers at the last second total 3,500; six correct instant answers total 6,000. Hesitating costs more than an occasional miss. So in duels, read the stem first and form a guess before the options finish registering — the shape table above is your shortcut. If two options survive, take the shorter, broader, more first-person one and move on.

Challenges and leagues

A challenge link puts a friend on your exact puzzle: same theme, period, difficulty and five questions. Your score stays hidden until they finish, and the link lasts 48 hours. The fair way to use it is to challenge on a theme you both know nothing about; the fun way is to pick one you have been practising.

Group leagues are a shared board for a group chat, with daily scores, weekly and all-time totals, and everyone’s streak. Only your first run of the day counts, in whichever theme you play, so warm up on the archive rather than on today’s puzzle.

Where to go next

Themes have their own personalities. Our tour of the strangest animal questions is full of “is” misconceptions, and our look at tech complaints is a masterclass in self-searches. For the Stereotypes theme, read what Google thinks of every star sign, where stereotypes come from and autocomplete and bias, which explains why a suggestion describes searchers, never the people being searched about.

▶ Play today’s stereotypes round