A first guess has two jobs and only one row

The opening guess cannot win the game. It can only do two things: tell you which letters are in the hidden word, and — where the tiles come back green — tell you where they sit. Everything after that is downstream of how much you learned on row one.

So the question "what is the best opener?" is really "which five letters, in which positions, leave the smallest number of possible answers once the colours come back?" That is a solvable question, and the answer is not the one most lists give.

The letter-frequency trap

The classic advice is to open with a word full of common letters, which usually produces something vowel-heavy: ADIEU, AUDIO, OUIJA-adjacent nonsense. It feels productive because four of the five tiles are vowels, and vowels are obviously important.

But a vowel-heavy opener spends the row on the least informative letters available. There are only five vowels. Almost every English five-letter word contains at least one. Learning that A is in the word therefore removes relatively few candidates — the news is not surprising. Learning that K is not in the word, or that a rare consonant occupies a specific position, cuts the pool far harder, because consonants are rare enough that their position carries real signal.

The other half of the trap is position. A letter's usefulness is not one number. S is extremely common as a first letter and comparatively rare as the last letter of a five-letter answer (plurals are usually excluded from answer lists, which kills a huge share of final-S words). An opener that ignores position is optimising a statistic the game does not score.

What actually ranks an opener

If you take a realistic answer list — a few thousand common five-letter words — and score every allowed guess by how evenly it splits that list, the same small family of openers keeps rising to the top: SLATE, CRANE, RAISE, SOARE, TARES, STALE. They share a profile rather than a vowel count:

  • Five distinct letters. A repeat inside the opener wastes a tile that could have tested something new.
  • Two or three vowels, placed where vowels actually occur. E and A are common in the middle and at the end; I and U are rarer overall and weaker spends.
  • Consonants that are common and positionally informative — S, T, R, L, N — rather than showy rarities like J or Z.
  • Plausible as an answer itself. This matters more than people expect. If the opener is a word the answer list actually contains, you have not thrown away the row on a pure probe.

Measured in bits, the best openers land around 5.7–5.9 bits against a common-answer list, which sounds abstract until you translate it: they leave something in the neighbourhood of 40 to 60 candidate words out of two-odd thousand. A mediocre opener like ADIEU leaves two to three times as many. That is the entire difference between the lists — and it is a real one, because those candidates are exactly what you will be enumerating by row four.

The opener is not a lucky guess. It is a purchase of information, and the best ones buy roughly six bits for five letters.

The plan beats the guess

Here is the useful reframe: stop choosing an opening word and start choosing opening pair. Two well-chosen guesses can test ten distinct letters, which is enough to place most answers in a very small set by row three. Greedy single-guess rankings ignore this, because they score each guess as if it were the last one.

Good pairs share no letters and cover both the high-frequency consonants and the vowels most likely to appear:

  • SLATE then CHOIR — ten letters, four vowels, and R/C/H do real work.
  • CRANE then SPILT — covers S, P, L, T and I on the second pass.
  • TARES then BLIND — heavier on consonants, useful when your first row comes back mostly grey.

The second guess should only repeat a letter from the first when that letter was yellow. Repeating a grey letter is the single most common waste of a row, and it happens because people reach for a familiar word instead of checking their own board.

When the conventional advice stops paying

An opener that maximises average information is a statement about the average. Real games are not average, and there are three specific places where the rule bends.

1. The answer list is curated, not the dictionary

Most Wordle implementations, including ours, draw answers from a hand-built list of common words rather than every five-letter entry in a dictionary. That changes the statistics. Words built around "professional" letter distributions — SLATE, CRANE — are well-represented in a curated list. Obscure consonant clusters are not. If you know the list skews common, a natural English word is a better probe than a mathematically perfect string of rare letters.

It also means some openers are secretly answer-adjacent. CRANE is a plausible answer; SOARE is not. All else equal, prefer the word that could itself be the target.

2. Repeats break the five-distinct-letters assumption

Openers with five different letters cannot tell you whether the answer doubles a letter — and a small but real slice of answers does (ABBEY, FUZZY, MIMIC, SISSY). You cannot fix that on row one without giving up coverage, and giving up coverage to insure against a rare case is a bad trade. The right move is to notice the possibility later: if four rows of elimination leave you with a pattern that no five-distinct-letter word fits, stop searching for a new letter and start testing repeats.

3. You are playing hard mode

Hard mode forces you to reuse every green and yellow letter in later guesses. That constraint changes which opener is best, because an opener whose letters are awkward to build around — a lone Y in position five, say — can trap you. In hard mode, prefer openers whose letters sit in flexible positions and pair naturally with common suffixes.

What does not help

A short list, because these circulate constantly:

  • Word shapes. Patterns like "double hump" describe typography, not the answer distribution. They feel clever and predict nothing.
  • Yesterday's answer as a hint. Answers are drawn independently. There is no memory in the game.
  • Reading other people's opening words. Fine as a starting point, useless as a strategy — the value was never the word, it was the enumeration that follows.
  • Googling the answer. You already know why this does not count.

A six-row script that works

If you want a concrete routine rather than a theory: open with a two-vowel, three-consonant word (SLATE is a fine default). Second row, take the ten-letter coverage guess that shares no letters. Then count candidates out loud before guessing again — literally list them. From row three onward, every guess should be chosen because it eliminates the most remaining words, not because it is the word you would like the answer to be. That last discipline is what separates players who finish in three from players who lose on row six with two candidates they never named.

Try it yourself

Play one round and write down your first two guesses before you make them on the board. After the game, ask how many candidates each row actually eliminated. Most players find their second guess was worth about half of what it could have been.

Play Wordle  →

Editorial note: Letter statistics vary with the word list a particular version uses, and no single opener is optimal for every list. Treat the rankings here as a reasoning framework, not a fixed answer key.