An even draw is not how English behaves
Ask most generators for a random letter and you get a flat roll. Twenty six letters, one in twenty six each, Q as likely as E. Mathematically clean. Useless for anything resembling language, because a flat roll hands you Q, X and Z at the same rate it hands you E and T, and a word game built on that draw stops being playable around the fourth tile.
So there are two odds modes here, and picking the wrong one is the difference between a fair draw and a fun one.
The frequencies English runs on
Switch odds to English and the pool gets weighted by how often each letter appears in ordinary written text. E takes roughly one letter in eight. Z shows up about once every fourteen hundred. The gap is wider than people guess until they see it laid out.
- E12.7%
- T9.1%
- A8.2%
- O7.5%
- I7.0%
- N6.7%
- S6.3%
- H6.1%
- R6.0%
- D4.3%
- L4.0%
- U2.8%
- C2.8%
- M2.4%
- W2.4%
- F2.2%
- G2.0%
- Y2.0%
- P1.9%
- B1.5%
- V1.0%
- K0.8%
- J0.15%
- X0.15%
- Q0.10%
- Z0.07%
Figures follow the standard counts drawn from large samples of English prose. Bars are scaled against E at 100 percent.
Nine letters carry more than two thirds of all written English between them. The bottom six barely register.
Notice what that means for a draw of six letters. Under even odds you land a Q, X, J or Z roughly half the time. Under English odds it drops to about one draw in thirty. Neither number is wrong. They answer different questions. Even odds asks which letter of the alphabet came up. English odds asks which letter you would find pointing at a random spot on a page.
Five vowels, and how fast they run dry
The vowel pool holds A, E, I, O and U. Turn on no repeats with vowels selected and the ceiling is five letters, because there is no sixth vowel to hand you. The counter clamps itself rather than silently repeating, so the number on the slider always matches what lands on the board.
Same rule everywhere. A to Z with no repeats tops out at twenty six. A custom pool of eight letters tops out at eight. Type duplicates into the custom box and they get stripped, so BANANA becomes BAN. Weighting a pool by typing a letter twice does not work, and the readout under the board tells you the real pool size after cleanup.
Custom pools are where this earns its place in a game night. Feed it the seven tiles on your rack, draw one, and you have a starting letter chosen without anyone picking a favourite.
The people who keep this tab open
Teachers, mostly. A letter on the board starts a phonics round, a spelling drill, or a name-something-beginning-with game that needs no prep. English odds keeps those rounds playable. Even odds turns them into a slog around the third Q.
Writers use it differently. One letter, one constraint, and a sentence that has to start there. Improv groups do the same thing out loud. If the constraint you want is a whole word rather than a letter, the Random Word Generator is the better door, and Pictionary Word Generator covers the drawing-game version of the same problem.
Then there is the group that turns up looking for something adjacent. People who want letters and digits mixed together want the Random String Generator. People holding a set of letters already and wanting them reordered want Shuffle Letters. People who need a number rather than a character are after the Random Number Generator or, for tabletop work, Roll Dice.
Fine for a game night, wrong for a password
Here is the limit worth stating plainly.
The draw runs on Math.random(), the ordinary randomness built into every browser. It is fast, it is even enough that no letter shows a bias you would notice across thousands of draws, and it is predictable to anyone who knows the algorithm. Sequences from it are reproducible under the right conditions. That is fine when the stake is which letter starts your next sentence. It is not fine when the stake is an account.
Do not build passwords, API keys, session tokens, licence codes or anything else guarding access out of these letters. Use the Password Generator for that, which draws from the browser cryptographic source instead. The distinction sounds academic until someone runs a targeted attack against a token that turned out to be guessable.
One smaller caveat on the English odds mode. The weights come from general prose. Technical writing, code, chemical names and text in any language other than English carry different distributions entirely, so the weighted mode models a specific kind of text rather than letters in general.
Reading the strip along the bottom
The last five draws stay visible under the board. Not decoration. Redraw eight or nine times in a classroom and it gets hard to remember whether B has already come up, and the strip answers that without a second thought. Each entry keeps its own settings, so a draw made under vowels-only stays labelled that way after you switch back to the full alphabet.
Everything runs in the page. No draw is sent anywhere, nothing is stored beyond the current tab, and closing it wipes the history. Reload the page and the board comes back empty.
