What the colours actually tell you
Every Wordle guess returns three kinds of tile: green (this letter is in the answer, in this exact spot), yellow (this letter is in the answer, but not here) and grey (this letter isn't in the answer — or you've already accounted for every copy of it that is). Treat each guess as a question and each colour as an answer, and the game stops being "guess a word" and becomes "narrow a set." Before your first guess, any of the thousands of five-letter words in the answer list is possible. Each guess, whatever comes back, removes every word inconsistent with that result. Play well and the set shrinks fast; play by hunches and it barely shrinks at all.
Choosing a starting word: letter-frequency reasoning
Nobody knows the day's answer in advance, so the first guess can't be aimed at a specific word — it has to be aimed at the letters most likely to appear across five-letter English words in general. A handful of consonants and vowels turn up disproportionately often: E, A, R, I, O, T, N and S account for a large share of English letter occurrences (it's the basis of the old typesetters' mnemonic "ETAOIN SHRDLU," which ranks letters roughly by how often they were needed in a print shop's type case). A first guess built from several of these, in positions where they commonly sit — S is unusually common as a word's first letter, E is unusually common as its last — tends to come back with more colour than a guess picked at random.
That's the reasoning behind openers like SLATE, CRANE and RAISE: five distinct letters, no repeats, drawn from the commonest set, arranged in plausible positions. The "no repeats" part matters specifically for a first guess. Testing the same letter twice when nineteen others are still completely untested wastes a tile — at best it confirms a count you didn't need yet, when that slot could have tested a brand-new letter instead.
The duplicate-letter rule, precisely
This is the part players most often get wrong. Wordle doesn't colour each letter of your guess by asking "is this letter anywhere in the answer?" — it works from the answer's actual letter counts, position by position, in two passes. First, every tile that matches its exact position turns green, and that letter is removed from a working tally of the answer's remaining letters. Second, working left to right through whatever's left, a tile turns yellow only if its letter still has a remaining copy in that tally — and turns grey, and removes nothing further, once the tally for that letter has hit zero.
Play it out. Suppose the answer is CRANE and you guess EAGER. Position by position, nothing lines up — no greens at all, even though every letter in EAGER except G is somewhere in CRANE. The tally to draw on is one each of C, R, A, N, E. Scanning the guess left to right: the first E claims the answer's only E and turns yellow; A claims the answer's only A and turns yellow; G matches nothing and turns grey; the second E finds the tally for E already at zero — used up by the first E — so it turns grey, despite E genuinely being in the word; R claims the answer's only R and turns yellow. Two identical letters in a guess, two different colours, because the answer only had one to give.
The rule to hold onto: a repeated letter in your guess can only earn as many yellows or greens as the answer actually contains copies of it. A second occurrence beyond that count is grey, full stop — it is not telling you the letter is absent, only that you've already claimed every copy the answer has.
The second guess: probe or exploit
After one guess you're holding partial constraints and a decision: spend the second guess probing brand-new letters you haven't tested yet, or spend it on a word that's actually a live candidate given what you already know. Early on, with hundreds of candidates still possible, probing tends to win — a guess built from fresh common letters splits a large remaining set more evenly than a guess drawn from the survivors, most of which already share several letters with each other. Once the candidate list is down to a handful, that trade reverses: there's little new information left to buy, and a guess that could itself be the answer is worth more than one that can't be.
Hard mode's trade-off
Hard mode requires every subsequent guess to use the hints you've already been given — confirmed greens stay locked in place, confirmed yellows must appear somewhere in the guess. That removes the option of a pure probe word chosen purely to test new letters while ignoring what you already know, unless a word happens to satisfy the constraints and still introduce something new. The trade is discipline for flexibility: hard mode guarantees you never contradict information you already hold, at the cost of sometimes forcing a guess that repeats a lesson you'd already learned instead of the guess that would have taught you the most.
Why elimination beats guessing likely answers
A guess's value isn't just whether it happens to be correct — it's how well it splits the remaining candidates across every colour pattern Wordle could hand back. A guess drawn only from plausible-looking answers, if wrong, often eliminates little beyond that one word, because plausible answers tend to share the same common letters with each other and so react to feedback in similar ways. A guess built to probe fresh, information-rich letters instead tends to divide the field into more evenly sized groups whatever comes back — so even though it's less likely to be the answer itself, it usually removes more candidates when it isn't. That's the logic any solver, including the one on this site, is built around: rank surviving words by how well they'd split the remaining set, not by how plausible any one of them individually looks.
Worked example
Feed the solver a green C in the first position, a green T in the fourth, a yellow A that's confirmed present but can't sit in the second position, and confirmed-absent R and S — pattern c__t_. Against a small demo list, exactly one word satisfies every rule at once: COATI, which places its A in the third position, clear of the one spot it was ruled out from. Every other candidate on the list breaks at least one constraint — a wrong letter in a locked position, a required letter missing altogether, or a letter that was ruled out entirely. That's elimination working as intended: each rule removes words, and what's left after enough rules is the answer.
What good play looks like
Within Wordle's six-guess limit, an elimination-first approach with a well-chosen opener typically brings the candidate list down to one or two words within three or four guesses, leaving the remainder as a buffer for unlucky patterns rather than a last resort. There's no guaranteed number of guesses — a rare word or an unlucky first result can still eat the whole budget — but consistently choosing the guess that narrows the field the most, rather than the guess that feels most likely to be right, is what keeps most days comfortably inside the limit.
Frequently asked
Is SLATE, CRANE or RAISE always the best opener?
None of them is best on every single day — the actual answer is one specific word, and any fixed opener will occasionally miss badly. What makes these words good defaults is the reasoning behind them: common, distinct letters in plausible positions, which performs well on average across many days, not a guarantee for any one of them.
Should I ever guess a word with a repeated letter?
Rarely on the first guess, since it tests one letter twice while nineteen others sit untested. Later on it can make sense — if you've confirmed one occurrence of a letter and want to find out whether the answer has a second, a guess that repeats it is the only way to ask that specific question.
Does hard mode change the underlying logic?
No — the colour rules and the duplicate-letter maths are identical in both modes. Hard mode only restricts which guesses you're allowed to type next; it doesn't change how any guess gets scored.
Is elimination guaranteed to solve it within six guesses?
No strategy can guarantee that for every possible answer, but consistently picking the guess that removes the most candidates, rather than the guess that looks most likely to be right, gives you the best odds the game allows. Try it with our Wordle solver, which applies exactly this filtering to your own guesses.