Number Instinct

All questions so far

Every question that has appeared, with its explanation. The archive below is the reason this page is worth reading even if you never answer one.

Why almost everyone picks 7

Pick a number between 1 and 10.

Ask a room full of people for a number from 1 to 10 and a strange thing happens: 7 wins, usually by a wide margin. The reason is that we are not really picking at random — we are picking what feels random. 1 and 10 are the edges, so they feel like a cop-out. 5 sits dead centre and feels too obvious. Even numbers feel divisible, tidy, less "random". That leaves 3 and 7, and 7 is the only number in the range that cannot be divided into anything smaller inside it, which makes it feel the most arbitrary of all. Psychologists call this the blind-spot of randomness: asked to be unpredictable, humans become extremely predictable.

The 37 and 73 problem

Pick a number between 1 and 100.

In a range of one hundred there are a hundred equally valid answers, yet the results are never flat. Two numbers rise above the rest: 37 and 73. Both are odd, both are prime, both have two different digits, and neither ends in 0 or 5. That combination is exactly what our brain reaches for when it wants to sound unplanned. Meanwhile round numbers — 50, 100, 20 — get avoided precisely because they feel chosen. The irony is sharp: the harder someone tries to be unpredictable, the more they converge on the same small handful of numbers as everyone else.

We are terrible at counting our own habits

How many times do you think you unlocked your phone today?

Most people guess somewhere between 20 and 40. Measured logs from screen-time studies routinely land far higher — frequently past 50, sometimes over 100. The gap is not dishonesty; it is how memory works. We remember episodes — the long scroll, the call, the message thread — and those are few. The five-second checks, the glance at the lock screen, the reflexive unlock while waiting for a lift: these leave almost no memory trace at all, because nothing happened worth encoding. Any behaviour that is fast, frequent and unremarkable will be undercounted by the person doing it.

The odd-number bias, in a smaller room

Pick a number between 1 and 50.

Shrink the range and the same fingerprint appears. Odd numbers take a clear majority, and within them the two-digit odds with mismatched digits do best — 17, 23, 37 and 13 tend to cluster at the top. Round tens are avoided almost entirely; so are the endpoints. What is worth noticing is that the pattern survives the change of range. That tells us the bias is not about the specific numbers at all — it is about a rule the brain applies before it looks at the options: avoid anything that looks like it was chosen for a reason.

The mirror question

What percentage of people do you think will answer this survey honestly?

This one has no correct answer, which is exactly why it is interesting. Estimates of other people's honesty tend to say more about the person estimating than about the crowd. People who answer carefully usually guess high; people who clicked through quickly often guess low, as though projecting their own behaviour outward. Social psychologists call the general pattern projection bias — using the self as the default model for everyone else. Watch the distribution and you are not really looking at how honest the internet is. You are looking at how honest each visitor assumes everyone else to be.

Why so few people say 5

Toss a coin 10 times. How many heads?

Five is the single most likely answer — it comes up about 25% of the time — yet it is chosen far less often than that. The reason is that 5 out of 10 feels like the textbook reply, and textbook replies feel wrong when someone asks you to guess. So people drift to 6 or 4, reaching for something that sounds observed rather than calculated. There is a second layer too: many expect coin tosses to "even out", so a run of heads feels like it must be repaid by tails. It does not. Each toss forgets everything that came before it, which is precisely why the middle of the distribution is fatter than intuition allows.

The game where thinking one step further wins

Guess a number 0-100. The winner is whoever is closest to two-thirds of everyone average.

If everyone answered at random the average would sit near 50, so two-thirds of that is about 33. But if you assume others reason that far, you should say 22. And if they assume you assume that, 15. Follow the chain to its end and the only stable answer is 0. Almost nobody plays it that way. Real experiments with this game land around the low twenties, which tells us something specific: most people go one or two steps into the reasoning and then stop, assuming everyone else stopped earlier. Winning here is not about being the smartest in the room — it is about correctly guessing how smart the room thinks it is.

Laughter is mostly punctuation, and punctuation is invisible

How many times did you laugh today?

Most people guess a handful. Observational studies suggest the real figure is much higher, because the great majority of laughter is not a response to anything funny. It is social punctuation — the small exhale at the end of a sentence, the polite laugh that keeps a conversation moving, the one-second reaction to a message. Recordings of ordinary conversation find that speakers laugh at their own words more often than listeners do, which makes little sense if laughter were about humour and perfect sense if it were about smoothing the exchange. We remember the two or three moments that were genuinely funny and forget the forty that were social glue.

The answer is 23, and it feels wrong

In a room of how many people is there a 50% chance two share a birthday?

Most guesses land near 180, roughly half of 365, and the logic seems airtight until you notice what is actually being counted. The question is not whether someone shares your birthday — that really would need a large room. It is whether any two people match, and the number of pairs grows far faster than the number of people. With 23 people there are 253 possible pairs, which is why the probability crosses one half there. Our intuition tracks the guests; the mathematics tracks the handshakes. Nearly every surprising result in probability comes from that same mismatch — counting the wrong thing with complete confidence.

Forty-two, if paper could be folded that far

Fold a sheet of paper in half repeatedly. How many folds to reach the Moon?

Ordinary paper is about a tenth of a millimetre thick. Each fold doubles it, and doubling is the operation human intuition handles worst. Ten folds gives roughly the thickness of a hand. Twenty folds clears ten kilometres, higher than a cruising airliner. Forty-two folds passes the distance to the Moon — around 384,000 kilometres — with room to spare. Nothing in daily life prepares us for this, because almost everything we experience adds rather than multiplies. Our sense of "a lot" was built for stacking, not doubling, and that single blind spot explains why compound interest, viral spread and pandemic curves all feel implausible right up until they are undeniable.

Time runs at a different speed depending on who is watching

Without counting, how many seconds do you think one minute is? Guess when it has passed.

Almost nobody lands on exactly sixty. Which way people miss turns out to depend on state rather than skill. When attention is occupied, estimates come in short — the minute felt like forty seconds because the brain was busy elsewhere and logged fewer moments. When someone is bored, anxious or waiting, the same minute stretches past ninety. Age tilts it too: older adults tend to underestimate longer intervals, which is one proposed reason years feel faster as life goes on. There is no internal clock ticking away somewhere. There is only a rough tally of how much happened, read backwards as duration.

We estimate the world from the people we happen to see

What percentage of people in your country wear glasses or contacts?

Guesses swing wildly, and the direction of the error is rarely random. People who wear glasses guess high; people who do not guess low. This is availability at work — instead of retrieving a statistic we cannot possibly hold, we sample whoever comes to mind and scale that up to a nation. The same shortcut explains why almost everyone believes their own profession is underpaid, their own city has the worst traffic, and their own generation had it hardest. None of these are lies. They are honest reports from a sample of one, presented with the confidence of a census.

The number where a phone stops being a tool

At what battery percentage do you start feeling uneasy?

There is no engineering reason for any particular figure — a phone at 19% works exactly as well as one at 21%. Yet answers cluster hard around 20, with a second cluster at 30 and almost nothing below 10. Part of this is that the interface tells us when to worry: the moment the indicator turns red, the number acquires meaning it did not have a second earlier. The rest is loss aversion. What we are pricing is not the remaining charge but the cost of being cut off, and that cost feels enormous, so we build a buffer far larger than the situation requires. The same instinct fills fuel tanks at a quarter and keeps money in accounts that earn nothing.

The buffer we build and then spend

How many minutes before you actually need to wake up do you set your alarm?

Very few people set an alarm for the moment they must be up. Most build a cushion, and then reliably spend it — the snooze, the scroll, the extra few minutes that were budgeted as insurance and consumed as luxury. What makes this interesting is that we do it knowing we will do it. The buffer is not really about waking; it is about buying permission to not wake yet. Sleep researchers point out the cost: fragmented sleep in the final stretch tends to leave people groggier than an unbroken block ending at the true deadline. We are trading better sleep for the feeling of extra time.

One wins, and that is a security problem

Think of a four-digit PIN. What is its first digit?

Analyses of leaked PIN datasets find the distribution is nowhere near flat. Sequences dominate the top of the list, which pushes 1 into first place as an opening digit by a wide margin, followed by 0 and 2. Years starting with 19 add another heavy cluster. The practical consequence is uncomfortable: a small handful of combinations covers a startling share of all PINs in use, so an attacker with three attempts is not guessing blind. The deeper point is the same one running through this whole page — asked to choose freely, humans produce highly structured output, and structure is exactly what a code is supposed to lack.

The line that moves as you approach it

At what age does middle age begin?

Ask twenty-year-olds and the answer sits somewhere around forty. Ask forty-year-olds and it has drifted to fifty. The boundary is not a fact about bodies; it is a moving marker that stays politely ahead of the person answering. Something similar happens with the word "old", which surveys consistently find is placed roughly fifteen years beyond the respondent own age, at every age. Because the concept is defined by distance from oneself rather than by any fixed point, almost nobody ever arrives at it. Middle age is permanently just up the road, and that is not denial so much as how the category is built.

The number that cannot be right, and usually is not

Out of 100 drivers, how many do you think are better than average?

By definition roughly half of drivers are below the median, so any answer far above fifty is describing a mathematical impossibility. Yet surveys of actual drivers routinely find large majorities placing themselves in the top half, sometimes approaching ninety percent. Part of the trick is that "better" is undefined — the cautious driver scores themselves on safety, the quick one on skill, and each quietly picks the yardstick they already win on. The effect fades in domains with clear feedback and hard numbers. It thrives wherever the standard is vague and nobody keeps score, which describes most of what people believe about themselves.

We photograph instead of remembering

What percentage of your photos have you never looked at again?

Most people guess low, then check their gallery and revise sharply upward. Taking a picture feels like preserving a moment, but studies of museum visitors found the opposite: people who photographed objects recalled fewer details of them afterwards than people who simply looked. The act of capture appears to license forgetting — the memory has been outsourced, so attention moves on. The archive keeps growing precisely because it is never revisited; nothing prompts deletion, and each new photo is cheap. What we end up with is not a record of a life but a very large pile of intentions to remember.

Waiting has no clock

What percentage of the messages you send are read within one minute?

Estimates here tend to run high, and the reason is that unanswered messages are the only ones we think about. A reply that lands in four seconds vanishes from memory; the one still sitting there after an hour occupies it completely. Silence also accrues meaning that speed never does — we fill it with theories, almost all of them about ourselves. Read receipts made this worse rather than better by converting a vague absence into a precise, timestamped one. The interval did not change. What changed is that we can now watch it.

The question that measures itself

How many seconds have you spent on this page so far?

This is the only question here whose answer is being generated while you consider it, which makes it a small trap. Reading absorbs attention, and absorbed attention produces short estimates, so most guesses come in well under the true figure. There is a second effect: the moment you were asked, you started paying attention to time, and time under observation behaves differently — it slows. Whatever number you gave was shaped by the asking. Every question on this page has been doing something similar, quietly; this one simply admits it.