Random Letter Generator

Q: Should I use flat or frequency-weighted letters for Scattergories?

Flat, over a restricted alphabet. The game die itself is flat over 20 faces and leaves out Q, U, V, X, Y and Z, so the Game die preset with the flat setting reproduces it exactly. Frequency weighting would hand out E and T far more often than the physical die ever would.

Q: Why did the same letter come up twice in a row?

Because each draw ignores the one before it. With 26 letters in play and repeats allowed, two consecutive draws match about 3.85 percent of the time, which is roughly once every 26 rounds. Switch off repeats if you want each letter used at most once per draw.

Q: Where do the letter frequency percentages come from?

They are the English letter frequency and word-initial frequency tables published in Robert Lewand’s Cryptological Mathematics (2000), the set most widely reproduced for English. They describe a sample of English prose, so different corpora shift the numbers, though the ordering of the top handful is stable.

Q: Are the letters generated on a server?

No. Every draw happens in your own browser using crypto.getRandomValues. Nothing is sent anywhere when you press the button, and the tool keeps working if the connection drops after the page has loaded.

One letter or a hundred. Pick which letters are eligible, then decide whether every eligible letter is equally likely or whether E should turn up as often as it does in real English. The percentage under the result tells you which question you just asked.

How many letters Eligible letters Weighting Repeats Case Also exclude

Flat and frequency-weighted answer different questions

Every free letter generator gives you a flat draw and stops there. Flat means each eligible letter has the same chance — with all 26 in play, 1 in 26, about 3.85% apiece, Z as likely as E. That is right when the letter is a rule imposed on people: a party-game round, a spelling drill, a name game where everyone must answer with the drawn letter. The difficulty is shared and nobody can claim the draw favoured anyone.

Weighting by frequency asks the opposite question: what letter would English itself hand me? Under the text-frequency setting E comes up about 12.7% of the time and Z about 0.074% — a gap of roughly 170 to one. Use it when you are imitating the language rather than refereeing a game: cipher and cryptanalysis exercises, phonics practice weighted towards the letters a learner will actually meet, or filler that has to look like text rather than noise.

The third setting, word-initial frequency, weights by how often a letter starts a word. It is a genuinely different table: T leads at about 16.7% because English is full of the, to and that, while E — the most common letter overall — falls to about 2.0%, because very few English words begin with it.

Where the percentages come from, and what they are not

Both tables are the letter-frequency and word-initial-frequency figures published in Robert Lewand's Cryptological Mathematics (2000), the set most widely reproduced for English. They describe a sample of ordinary prose, which matters more than it sounds: a different corpus — legal text, chat messages, a dictionary word list — shifts every number, though the leading group of E, T, A, O, I, N survives almost any English sample.

The word-initial table carries a sharper caveat, and the tool will not pretend otherwise. It counts word occurrences in running text, so function words dominate: T's 16.7% is largely one word, the, repeated. That is not the same statistic as how many distinct English words begin with T, which is what you actually care about when judging whether a letter is playable. If you want a fair game, the honest instrument is a flat draw over a trimmed alphabet, not a weighted draw over all 26.

Trimming the alphabet instead of weighting it

That is what the presets are for. The Game die option reproduces the letter die used in Scattergories: 20 faces, flat, with Q, U, V, X, Y and Z left off — removed not because they are rare in text but because a round built on them stops being fun. The vowel and consonant presets cover spelling and phonics work, where you need a vowel to sit inside a made-up word or a consonant to start one. Y is offered on both sides, because English uses it both ways and the choice should be yours.

The Also exclude box is applied last, on top of the preset: type letters in any order, with or without spaces or commas, and anything that is not a letter is ignored. Empty the pool completely and the tool says so instead of quietly reverting to all 26; ask for more unique letters than remain and it caps the draw and tells you.

Frequently asked questions

Should I use flat or frequency-weighted letters for Scattergories?

Flat, over the game-die preset. That combination is an exact software copy of the physical die: 20 equally likely faces, Q, U, V, X, Y and Z omitted. Frequency weighting would give you E and T for a third of your rounds, which is easier than the game intends and duller than it deserves. If your house rules already ban certain letters, add them to the exclude box rather than switching weighting.

Why did the same letter come up twice in a row?

Because each draw is independent of the last. With repeats allowed and 26 letters eligible, consecutive draws match about 3.85% of the time — around once in every 26 rounds, so a repeat inside an evening's play is expected rather than suspicious. Set repeats to "each letter at most once" and the tool draws without replacement, which is what you want for a spelling worksheet or a round-by-round session that should not revisit a letter.

How does weighting work when repeats are switched off?

The tool picks a letter in proportion to its weight, removes it from the pool, and re-normalises the remaining weights before the next pick. So a heavy letter is very likely to appear early in the sequence, but it can appear only once. This is a deliberate choice, not the only possible one — sampling without replacement has more than one reasonable definition once weights are involved, and this is the one that matches what people expect from a drum of raffle tickets.

Are the letters generated on a server?

No. Draws come from crypto.getRandomValues inside your browser and nothing leaves the page. For flat draws the tool discards values that fall in the short leftover block at the top of the 32-bit range before taking a remainder — 2³² divided by 26 leaves 22 spare values, and skipping that step would make the first 22 letters of the alphabet very slightly more likely than the last four. Weighted draws instead convert a 32-bit value into a fraction and walk the cumulative weights, so the same imbalance never arises.

What this tool will not do

It covers the 26 basic Latin letters only: no accented characters, no letters from other scripts, and no digraphs, so a drawn Q will not automatically bring a U with it. The frequency tables are a single published convention rather than a live corpus measurement, and they are English-only — using them to model German or Spanish would be wrong. Nothing is saved between visits: no history, no favourites, no undo. And while the draws are unpredictable, this is a page you could reload and re-draw at will, so it is evidence of nothing in a dispute. If a letter has to be provably fair to a sceptic, roll a physical die in front of them.