Random Letter Generator: Why Nobody Picks One at Random

The round where nobody said Q

We played a categories game at a family lunch, the kind where someone calls out a letter and everyone races to name a country or a film starting with it. Three rounds in, the letters somebody had picked off the top of their head were M, then S, then T. Nobody had said Q, or X, or Z, and nobody was doing it on purpose.

That is not a family quirk. Asking a person for a random anything has been studied for fifty years, and the answer is always the same: people cannot do it. What surprised me is that the machines cannot either.

Ask for a random digit and you will probably get 7

The classic experiment is about numbers. In 1976 the psychologists Michael Kubovy and Joseph Psotka asked 558 people for the first digit from 0 to 9 that came to mind. If the answers were random, each digit would take about 11 per cent. Seven took 28.4 per cent, close to three times its share.

Their follow-ups are the interesting part. When the range was moved to 6 through 15, sevens dropped to 17.3 per cent. When seven was given as an example in the question, it fell to 16.6 per cent. Asked for a number in the twenties, 27.7 per cent said 27; asked for one in the seventies, only 15.5 per cent said 77. Their conclusion was not that people love seven. It is that people are doing an impression of somebody choosing at random, and seven is what that impression sounds like.

Real randomness does not sound like that. It repeats, it clumps, it leaves long gaps. A genuine run of coin flips produces five heads in a row far more often than people believe. Asked to fake one on the spot, a person produces something that looks patternless to them, which is a different thing entirely and a measurable one.

Letters go wrong in their own way

There is no letter version of the 1976 study. I looked, and the well-documented work on this bias is about digits. But a second effect takes over with letters, and this one is easy to check for yourself.

English runs heavily to a handful of letters: E, T, A, O, I, N, S, R and H come first, while J, X, Q and Z trail far behind. That ordering is not folklore. It is why a Scrabble set has twelve E tiles and exactly one each of J, Q, X and Z, and why those four are the highest scoring. When someone blurts out a letter, the common ones are simply closer to hand. Q takes deliberate effort to reach for when there is no alphabet in front of you.

Put the two together and the family lunch makes sense. People are not only bad at randomness, they are bad at it in a predictable direction: towards whatever is most frequent anyway.

Language models have the same problem

This is where it stops being a party trick. A 2024 paper on selection bias in large language models tested what happens when a model is asked to choose an item from a list, letters included. The choices were not uniform, the skew got worse as the list got longer, and some letters were almost never chosen at all, with I and Z among the ones the models passed over. Raising the temperature setting pushed the results back towards uniform but never got there.

That is the same failure wearing different clothes. A model is not drawing from a hat. It is producing the pattern its training leaned towards, and its training is made of text written by people who are themselves bad at this. If you need a letter that is genuinely unbiased, asking a chatbot for one is not the answer, and neither is asking your friend.

It is the lesson that keeps recurring here, whether the subject is coin flips or shuffling a deck. Anything that has to feel random to a human is not the same thing as a uniform draw.

What a real draw looks like

The letter picker has no favourites. Each spin calls the browser's own cryptographic random generator, seeded by the operating system rather than by a formula anyone can predict, and the letter is chosen before the wheel starts moving. The animation shows you the result. It does not decide it. If you want to see the same generator tested, the randomness page runs up to a hundred thousand simulated draws and plots them.

Three modes:

• All (A to Z), every letter with equal odds, for word games, hangman and any pick a starting letter round.
• Vowels, just A, E, I, O and U, for phonics practice or to narrow a categories game.
• Consonants, everything else, for spelling drills built around blends.

Your last fifty picks are kept with a frequency bar under them. That list is worth a look after thirty or forty spins, because it is the fastest demonstration of the point this post opened with: some letters will have come up twice in a row while others have not appeared at all, and none of it means the wheel is broken. That is what randomness does, and it is exactly what a person faking it never produces.

What it will not do

• It cannot drop individual letters. The three modes are fixed, so there is no way to play A to Z without Q, X and Z. If you need an exact set, type the letters you want into the main wheel instead, one per line.
• It does not weight letters. Every letter in a mode has the same odds, so there is no gentler setting that quietly avoids the hard ones for younger players.
• It does not draw several letters at once. One spin, one letter.
• History lives in your browser. Clearing site data clears it, and it does not follow you to another device.

Making a letter game playable

Equal odds are fair and not always fun. Roughly half the alphabet is hard in a categories game, and landing on X twice in a round kills it. Two fixes work without pretending the draw is something it is not. Agree on one re-spin per round, decided before you start, so the hard letters stay possible without being a punishment. Or drop the letter picker for that game and type your own shortlist into the main wheel, which does let you choose exactly what is in play.

For a classroom, the vowels and consonants modes do the narrowing for you, which turns the same page into a phonics tool. The other pickers that need no signup and run on a school Chromebook are collected on the teacher page.

Spin one now

1. Open the letter picker.
2. Choose All, Vowels or Consonants.
3. Spin, and read the letter.
4. Check the history and the frequency bars if you want to watch real randomness being lumpy.

Free, no signup, no ads between spins. If you want a number rather than a letter, the number wheel does the same job with digits, and for choosing which student answers, the classroom spinner is the better fit.