Most lists of Wordle openers are ranked by letter frequency, which is a rough guide at best. This one is ranked by playing the game out. We took 1,346 likely answers (how that set was built is explained below), played every one of the 8,636 five-letter words in our list against every answer using Wordle's own colouring rules, and counted how many answers each opener leaves standing on average. Lower is better. The best everyday opener is RATES: after it, an average of 28.3 answers remain from 1,346. The best-known opener, SLATE, is fifth on the everyday list. And the raw winner of the whole simulation is a word most players have never used, which says a lot about what these rankings do and do not mean.
The everyday shortlist
These are the ten best openers among words a player is likely to know and remember. Every one has five different letters, and nine of the ten are built from the same small set of letters: A, E, R, S, T, L, I.
| # | Starting word | Answers left, on average |
|---|---|---|
| 1 | RATES | 28.3 |
| 2 | TALES | 29.7 |
| 3 | TEARS | 30.3 |
| 4 | TRIES | 31.0 |
| 5 | SLATE | 31.5 |
| 6 | RAISE | 31.5 |
| 7 | LANES | 31.8 |
| 8 | TIRES | 32.0 |
| 9 | CARES | 32.3 |
| 10 | ROLES | 32.3 |
For scale: the opener at rank 100 on the full list leaves 37.7 answers, the median opener leaves 100.8, and the worst word in the list, FUZZY, leaves 659.2, barely narrowing the field at all. The shortlist is tightly packed. RATES at 28.3 and ROLES at 32.3 differ by four answers a game, so any of the ten is a sound choice, and the one you remember is the best one.
How the ranking was computed
The answer set comes first, because a ranking is only ever as good as the answers it assumes. Wordle does not pick answers from every five-letter word in the dictionary; it uses a curated set of familiar words. Ours is a stand-in built the same way: the five-letter words in the ENABLE list that also sit in the 12,000 most frequent words in a large count of published English, and that have a dictionary meaning in Princeton WordNet. That gives 1,346 answers. It is not the New York Times list, and the limits section below is frank about the difference.
Each opener was then played against each answer with the game's real colouring rules, including the duplicate-letter rule (greens first, then yellows from the copies that remain). Answers that produce the same colour pattern cannot be told apart after one guess, so they count together. The score is the average size of those pattern groups: the expected number of answers still standing after the opener, before you have seen a single colour. Averages hide the worst case, so one more figure is worth knowing: when RATES draws all grey, which is its joint-largest outcome, 77 answers remain. No opener makes the first guess safe. It makes it informative.
The raw list, and why it looks strange
Ranked across all 8,636 words, with nothing filtered out, the top of the table is this:
| # | Word | Answers left | What it means |
|---|---|---|---|
| 1 | TARES | 27.4 | An adjustment for the weight of packaging when goods are weighed |
| 2 | ARLES | 27.7 | Money paid by a buyer to bind a contract |
| 3 | LARES | 27.7 | Guardian deities of a Roman household |
| 4 | RATES | 28.3 | Charges, prices, speeds |
| 5 | SANER | 28.6 | More sane; in better mental health |
Nothing has gone wrong. TARES, ARLES and LARES are real words built from exactly the letters that divide the answer set best, and the simulation does not care that few people know them. It should care, a little: an opener you have to look up is an opener you will not play, and a word you cannot pronounce is hard to build a second guess around. That is why this page leads with the everyday shortlist and shows the raw list underneath it. The honest answer to "what is the best starting word" is TARES by the numbers and RATES in practice, and any ranking that hides the difference between those two answers is tidier than the truth.
Where the famous openers land
STARE ranks 12th overall (30.8 left) and would sit between TEARS and TRIES if it were in the everyday set; it just misses that set on frequency, which shows how rough any cut-off is. SLATE is 17th overall and 5th on the shortlist at 31.5, RAISE 18th at 31.5, AROSE 45th at 34.2, IRATE 110th at 38.0. Then the drop: TRACE 370th (45.8), CRATE 327th (44.9), CRANE 508th (48.3). The C words suffer twice over. C appears in only 18% of the answers, and all three give up S, the single most informative letter in the set.
The letter counts behind that, over the 1,346 answers: E appears in 50% of answers, S in 45%, A in 40%, R in 35%, O in 30%, T in 29%, I in 27% and L in 27%. An opener is a net spread across those letters, in positions where they separate answers well. The vowel-heavy strategy fails that test, whatever its reputation: AUDIO ranks 5,239th and leaves 116.3 answers, ADIEU 1,971st with 69.6. Knowing the vowels early feels productive. The simulation shows it mostly is not, because an answer's vowels are the part you could have guessed anyway.
Choosing yours, and what happens next
Pick from the shortlist, or from the top hundred of the full list, and stay with it long enough to learn its patterns. The opener is the only guess you make with no information, so it is the only one that can be optimised in advance; everything after it belongs to the colours. When your opener draws a poor pattern, all grey is the famous one, the second guess matters more than the first ever did. The Wordle strategy guide takes the game from there, including the duplicate-letter traps that cost the most turns, and the Wordle Helper will list exactly which answers your colours leave standing.
Limits of this ranking
Three, stated plainly. The answer set is a frequency-built stand-in, not the game's real list, and the game's list has been edited since launch and will be again; treat the top of the table as a group of near-equals, not a strict order. The score is an average, so it says nothing about any single day's answer. And the simulation assumes you may play any follow-up word, which is standard-mode logic; hard mode, where clues must be reused, changes the value of later guesses, not of the opener itself.
Questions
Is TARES really the best Wordle starting word?
By this one measure, yes: against our answer set it leaves the fewest words standing on average, 27.4 of 1,346. But the gap between TARES and the best everyday word, RATES at 28.3, is less than one answer per game. A starting word only helps if you remember it and can follow it up, so the everyday shortlist on this page is the list we expect most players to use.
Why does CRANE rank so low here when other sites love it?
CRANE is not bad. It ranks 508th of 8,636 in our simulation, leaving 48.3 answers on average, and its letters are all common ones. Two things hold it back. C appears in only 18% of our answers, and a word with no S anywhere gives up the most informative letter in the set: S appears in 45% of them. Lists built on different answer sets rank it higher, which is a reminder that every ranking is a property of its answer set, not a law.
Should I play the same starting word every day?
There is no scoring reason to rotate. The opener's job is to divide the answer set, and it does that the same way every day. Playing one word consistently has a side benefit: you learn what its colour patterns usually mean, and your second guess gets faster. Rotate if you get bored, not because the maths asks you to.
Are vowel-heavy words like AUDIO and ADIEU good openers?
The simulation says no, clearly. AUDIO ranks 5,239th and leaves 116.3 answers standing on average, and ADIEU ranks 1,971st with 69.6. Vowels are common, but knowing which vowels an answer holds separates it from far fewer rivals than knowing about S, R, T and L. Four of the five letters in AUDIO are vowels, and the fifth, D, is not a strong separator either.
What should my second word be?
It depends entirely on the colours the first word drew, which is why there is no fixed second word worth memorising. The follow-up should use new letters, not re-test old ones, unless a green needs confirming. The Wordle strategy guide works through the main cases, and the Wordle Helper will list exactly which words still fit your clues.
Sources
- ENABLE word list (public domain): the 8,636 five-letter words simulated as openers: github.com/brokensandals/wordlists
- Word frequency counts from published English text (Google Books n-gram data), collected by Peter Norvig, used to build the 1,346-word answer set: norvig.com/ngrams/count_1w.txt
- Princeton WordNet, for the meanings in the tables and the answer-set test: wordnet.princeton.edu
- The simulation, the colouring rules it uses, and every figure on this page are described in full on How We Verify. Figures computed 28 October 2026.
Update log
28 October 2026: published. Rankings computed from the site's word list and the answer set described above; two independent runs of the simulation agreed on every figure printed here.