There is no shortage of articles telling you the best opening word for a five-letter guessing game. Almost all of them are computed against the New York Times answer list, which is a fine list to optimise against if you are playing that game. Word Guess draws from its own pool of 760 words, so those answers are, at best, approximately right here.
Working out the real one takes about twenty-five seconds of computation, and turns up something more interesting than a single word.
Two different questions
"Best opener" hides an ambiguity, and the two readings give different answers.
The first is information. A guess sorts the 760 possible answers into buckets by the feedback pattern it would produce. A guess that splits them into many small buckets has told you a lot; one that dumps most answers into a single bucket has told you little. Measured in bits, this is the guess's expected information, and the best opener by that measure is STARE at 6.032 bits.
The second is speed: which opener actually finishes the game soonest. To answer that you have to play it out. Every opener below was played against all 760 answers, with an identical solver taking over after the first guess — from then on it always guesses whichever remaining candidate splits the surviving pool best.
Show the numbers
| Opener | Mean guesses |
|---|---|
| TRACE | 3.057 |
| SLATE | 3.061 |
| CRATE | 3.072 |
| STARE | 3.088 |
| IRATE | 3.107 |
| CRANE | 3.129 |
| RAISE | 3.139 |
| ABOUT | 3.321 |
| AUDIO | 3.350 |
The winner on information is not the winner on speed. STARE carries the most bits, but TRACE finishes fastest, averaging 3.057 guesses against 3.088.
The gap is small — about 31 thousandths of a guess — but it is consistent, and the reason is worth understanding. Expected information is a one-move measure. It asks how well a guess splits the pool right now, and it is indifferent to whether the buckets it leaves behind are easy or awkward to finish from. An opener can score marginally higher on bits while handing you slightly nastier endgames, and over 760 answers that difference shows up.
The famous opener is not the best one here
CRANE is probably the most recommended opening word on the internet, on the strength of analyses done against the NYT list. Against our list it ranks 16th by information, averages 3.129 guesses, and is one of only two openers tested that ever needs a seventh guess — it fails to finish within six on 0.13% of answers, where TRACE never does.
This is not a criticism of CRANE. It is a demonstration that the answer depends entirely on the list, and that copying a number computed for a different game is how you end up with advice that is nearly but not quite right.
For contrast, the table includes two openers people genuinely type. ABOUT — a real answer, and the first word in our list alphabetically — averages 3.321 guesses. AUDIO, the classic vowel-loading opener, averages 3.35. Both are around a quarter of a guess worse than TRACE, which across a hundred games is roughly twenty-five extra guesses spent.
Why the good openers all look the same
Line up the top of the list — STARE, TRACE, IRATE, SLATE, CRATE, RAISE — and they are visibly shuffling the same letters. That is not a coincidence.
Show the numbers
| Letter | Share of answers (%) |
|---|---|
| E | 51.4 |
| A | 42.2 |
| R | 40.7 |
| T | 35.5 |
| S | 29.9 |
| I | 28.7 |
| O | 27.5 |
| L | 27.2 |
| N | 23.6 |
| H | 20.3 |
E appears in 51.4% of the possible answers, A in 42.2%, R in 40.7%, T in 35.5%, S in 29.9%. A word made of five letters drawn from that group tests five high-probability hypotheses at once, and every strong opener is some anagram of that idea.
It also explains why the vowel-stuffing instinct is a trap. AUDIO spends four of its five slots on vowels, and U and O are not especially common — together they cover far less of the pool than R, S and T do individually. You get a confident read on the vowels and learn almost nothing about the consonants, which are what you will actually be stuck on at guess four.
The endgame is where games are lost
A good opener is worth about a quarter of a guess. Playing the middle badly is worth several, and there is one mistake that dominates.
When four answers remain and you have two guesses left, the instinct is to guess one of the four. That is a one-in-four shot, and if it misses it usually eliminates only itself. The alternative is to spend a guess on a word that cannot be the answer but separates the four remaining candidates by their differing letters — after which the last guess is a certainty rather than a coin flip. Trading a 25% chance now for a 100% chance next turn is correct whenever you have the turn to spare, and it is the single largest source of avoidable sixth-line losses.
Our solver does this by default, which is part of why it finishes within six guesses on 100% of answers with TRACE.
Method, and the duplicate-letter trap
The scoring function used throughout is copied from the game's own, including its handling of repeated letters — which is where implementations of this analysis usually go wrong.
Guess ORDER against the answer ROBOT. The first O is present but misplaced, so it marks amber. The second O also appears in ORDER, and a naive implementation marks it amber as well. It should be grey: ROBOT's only unmatched O has already been consumed by the first one. Getting this wrong shifts a large fraction of feedback patterns and quietly corrupts the entire ranking.
To be certain the analysis scores exactly as the game does, the ported function was compared against the page's own on all 1,562,560 guess-and-answer pairs — every accepted guess against every possible answer. Zero mismatches.
The rest is unremarkable: 2,056 accepted guesses ranked by expected information over the 760-word answer pool, then the leaders played out against every answer. The script is tools/sim-word-guess.js, the raw output is src/data/word-guess.json, and the tables under each figure hold the full numbers.
The practical summary
- Open with TRACE if you want the fastest average finish here — 3.057 guesses, and it never needs a seventh.
- STARE, SLATE and CRATE are all within a hundredth of a guess of it. Pick whichever you find easiest to type and stop thinking about it — the opener is the smallest decision in the game.
- Skip the vowel-loading openers. AUDIO and ADIEU feel thorough and test the wrong letters.
- Spend a guess to guarantee the next one when several candidates remain and you have turns in hand. This is worth far more than any opening word.
And if you would rather not optimise anything: the openers in that whole table span about three-tenths of a guess. You are allowed to just type a word.
