An unscrambler answers two different questions at once. An exact anagram uses every letter you gave it — LISTEN, SILENT, ENLIST and TINSEL are all the same six letters rearranged. A subset uses some of them, which is what you need when a rack has seven tiles and the board has room for four. The first list is short and closed; the second is long and depends entirely on which dictionary is being searched.
That second dependency is the thing worth understanding, because it is the difference between a word that scores and a word that gets challenged off the board.
What dictionary is being searched?
Here, Webster’s Second International of 1934, whose copyright lapsed long ago and which survives as the word list shipped with most Unix systems. It is generous, it is out of date, and it is not a tournament list.
Competitive play uses purpose-built lists that differ by region. North American tournaments use one; most of the rest of the English-speaking world uses another, which is substantially larger. Words valid in one are invalid in the other, and neither matches a general-purpose dictionary.
The practical rule: an unscrambler built on a public-domain dictionary is a discovery tool. If you are playing competitively, check anything unfamiliar against your game’s own list before committing to it.
What does the question mark do?
It stands for a blank tile — any single letter. A rack of SILEN? searches every letter in that position, which multiplies the work and produces a much longer result list.
Blanks are also where the dictionary question gets sharpest. A blank makes obscure words reachable that you would never have thought of, and an obscure word from a permissive dictionary is exactly the kind that gets challenged.
Why does the first search take a moment?
Because the dictionary is loaded on demand rather than shipped with the page, and only the parts that can be reached from your letters. A seven-letter rack can only make words of seven letters or fewer, so the tool fetches those lengths — around 88 KB — rather than the whole list.
That is a deliberate trade. Loading the full dictionary on every page view would cost every visitor bandwidth for a feature most of them do not use; loading it on the first keystroke costs one visitor a short pause. After that it is cached and searches are instant.
How does the search actually work?
By letter counts rather than by rearranging. Every word in the dictionary is reduced to a canonical key — its letters sorted alphabetically — so LISTEN, SILENT and TINSEL all become EILNST. Finding exact anagrams is then a single lookup rather than a search through permutations.
This matters because permutations explode. Seven letters have 5,040 arrangements and eight have 40,320, so checking each one against a dictionary is hopeless. Sorting the letters turns the problem into a hash lookup, and the same trick handles subsets by checking whether the dictionary word’s letter counts fit inside your rack’s.
Rack management is the other half of the same idea. A rack of six consonants and one vowel has very few subsets regardless of how good the dictionary is, so the useful play is often a short one that dumps three consonants rather than the longest word available. The unscrambler shows you what is reachable; it cannot tell you what leaves you a workable rack next turn.
What are the words worth knowing?
The short ones, because they are what turn an unplayable rack into a play. Two-letter words are the highest-value thing to memorise in any word game — there are only a few dozen valid ones, they are the difference between a blocked board and an open one, and they are the most common source of a challenge.
Beyond that, the useful category is words with awkward letters and no vowel demands. Anything using Q without U, or V and W in short combinations, is worth recognising precisely because it is rarely reachable.
Does an anagram have to be a real word?
For a game, yes, and that is the whole reason a dictionary is involved. Letters can always be rearranged; whether the arrangement is playable is a question about the list, which is why an unscrambler that returned every permutation would be useless — a seven-letter rack has 5,040 of them and perhaps a dozen are words.
For a puzzle or a name, the constraint is different again. Anagram-based pseudonyms and crossword clues are judged by whether the result reads as language rather than by whether it appears in a list, and no tool can settle that.
Questions people ask
How many letters can I enter? Up to fourteen, which covers any rack plus a hook or two off the board. Beyond that the dictionary runs out before the search does.
Is this the Scrabble dictionary? No. It is a public-domain general dictionary, which means it contains archaic words that a tournament list excludes and excludes recent additions that a tournament list contains.
Why are the results grouped by length? Because that is the decision you are actually making. The longest word is not always the best play, and seeing all the five-letter options together is more useful than a single ranked list.
Does it know about proper nouns? Largely not, and that is correct for word games, where proper nouns are generally invalid. It also means place names and surnames will not appear even when they are anagrams.
Exact anagrams are a closed question; subsets are a question about the dictionary. The word unscrambler shows both lists separately, the word counter handles the other end of a word problem, and the morse code translator is where letters go when they stop being a game.