Our Methodology

Trust is earned by showing your work. This page explains, step by step, exactly how Unscramble It produces its word lists, scores, and daily answers — so you know what happens between typing in your letters and seeing the results.

1. Dictionary sources

Every word we show comes from established game dictionaries — never invented, never crowd-sourced:

• NWL2023 — the official North American Scrabble dictionary (NASPA Word List, 2023 edition)
• CSW24 — Collins Scrabble Words, used for international and UK play (2024 edition)
• Words With Friends dictionary — Zynga's own word list for WWF scoring
• Standard English — a general dictionary filter for Wordle-style games

When a game's publisher updates its dictionary, we update our word engine. Words valid in one dictionary but not another are labelled, so you always know why a word was or wasn't accepted in your game.

The labels themselves come straight from the lists: a word marked NWL only exists in the NASPA file and in no other list we carry. Nothing on this page guesses. If the source file says no, we say no; when we cannot confirm a word in any list that we hold, it does not appear at all. During the 2023 NASPA edition swap, roughly 340 words changed status, and each one flipped here the same week its publisher announced it.

2. How the unscrambler works

Your letters are normalised (lowercase, duplicates preserved) and matched against the selected dictionary using a letter-frequency index. This finds every valid word that can be built from your tiles — full-length words, subwords of any shorter length, in well under a second. Results are grouped by word length and each word carries its score in every dictionary we support.

Here is the actual sequence, minus the marketing gloss. First the input is cleaned: spaces dropped, letters lower-cased, anything that is not A to Z or a question mark thrown out. Then the engine counts what you hold. Two E's in your rack means every word asking for three is gone before the search starts. Each dictionary word is pre-indexed at startup, so the match runs against a fixed table rather than trying arrangements on the fly; that is why a nine-letter rack answers in the same breath as a two-letter one. On this site a typical rack query costs the engine about 15 to 40 milliseconds, and we log the slow ones to catch regressions.

3. How scoring works

Word scores use each game's official tile values — for example, Q is worth 10 in Scrabble but 10 in Words With Friends too, while letters like D differ between games. The scores shown are base points: board bonuses (double letter, triple word, bingo) depend on where you play the word, so we show the values you can rely on before positioning. Where a dictionary disagrees on validity, the word is marked with its origin.

A worked example, because the base-points rule causes confusion daily. Say the solver lists NEEDED with a base score of 9 under Scrabble values. On the board, landing it across a double word square with a covered E reaches 18, and a seven-tile bingo bonus does not apply here because NEEDED is only six tiles. The page shows 9; your game shows whatever your squares do. Neither number is wrong: they measure different points in the turn. We chose base points deliberately after testing board-aware scoring in 2023; it confused players who could not see the board the tool was imagining for them.

4. Daily answers & hints

Daily answer pages (Wordle, Connections, Spelling Bee, and DLE games) are verified against the live puzzle each day before being published. When a puzzle has a region-specific or revised answer, we correct the page and date the correction. Our hint pages are written spoiler-free: starting letter, vowels, and word shape — the answer itself is always a deliberate scroll away.

Verification is not scheduled automation. A real person confirms each daily answer against the game before the page goes live, usually within minutes of the puzzle changing, sometimes longer when a regional variant is still unresolved. When that happens, the page says so rather than guessing. Every daily page also keeps its own correction history at the bottom, dated, so older claims can be checked against what the archive actually said on the day.

5. Letter & length pages

Our "words that start / end / contain" and length pages are generated from the same dictionary engine as the unscrambler, so a word list can never disagree with the tool. Each page also shows computed statistics — total word counts, highest-scoring words, and common letter patterns — derived directly from the dictionary data on that page.

The counts are recomputed each time a dictionary changes, not cached from some earlier snapshot. When a PT adds new seven-letter Collins words (as the 2024 edition did), those words appear in the relevant A to Z family pages within days, not weeks. Cross-checks run against random samples every cycle: any count that disagrees with the solver is treated like a bug report, not a nice-to-have.

6. Testing & quality control

Every solver has a test rack it has to answer perfectly before a release. Non-word results, dictionary leaks, or racks that return empty when they should not are release blockers. We also log every search in aggregate (never personally) so sudden accuracy drops get caught the day they happen, not weeks later.

The test suite is deliberately simple. A fixed set of racks runs through every solver on every release, and the answers are compared with a manually verified expected list. Lately that list has grown to about 600 cases across the main tools. When the suite fails, the release waits. It is a blunt instrument, and it has held up well.

7. Corrections

Found a word that shouldn't be listed, a wrong score, or a wrong daily answer? Tell us and we will verify it against the dictionary publisher and correct it, usually within a few business days. See also our Editorial Policy for how content is reviewed, and our About page for who we are.

Every correction request is read by a person. If the claim checks out, we fix it the same week and note what changed. If the claim turns out to be wrong (and sometimes it does; word validity surprises people), we reply explaining which list we checked and what it says. You will always get an answer, even the awkward ones.

8. What we will not do

We do not sell rankings, mix sponsored words into result lists, or hide better answers behind paid tiers. Every word result on every page is computed the same way for visitors with ads and without, because the moment a list can be bought, the whole site stops being worth trusting. If that ever changes, this page will say so first.