Can a Computer Always Win Wordle?

Short answer: almost always, but not literally always. Give a computer the full list of 15,783 valid guesses and normal-mode rules and it will finish practically every game inside six turns, because each well-chosen guess cuts the field of 1,352 curated answers by a large and predictable factor. The honest exception is hard mode. When the surviving answers differ by a single letter, the rules can outlaw the one guess that would separate them, and a solver that reaches that corner with the wrong letters already banked can simply run out of turns.

What six guesses buy a perfect player

Wordle is an information game. With 1,352 answers on the curated list, the puzzle starts with about 10.4 bits of uncertainty, and a guess is good in exact proportion to how many of those bits its feedback earns back. That is what entropy measures: a high-entropy guess splits the remaining candidates evenly across many feedback patterns, so whichever colours come back, the field shrinks hard. Here are the top openers when every valid guess is scored against the curated answer list.

A single opening word earns 6.22 of the 10.4 bits on offer, taking the average field from 1,352 answers down to about 28. A strong second guess performs the same compression on whatever survives, so by the third turn a disciplined player is usually choosing between a handful of words. This is not theory we admire from a distance: our Wordle Solver ranks guesses with these exact precomputed scores and re-ranks the survivors after every row of feedback, and our entropy research explains how the numbers are calculated.

Trap families: where “always” breaks down

Entropy values are averages, and averages hide tails. English keeps clusters of five-letter words that differ by exactly one letter, and the curated answer list is full of them. EIGHT, FIGHT, LIGHT, MIGHT, NIGHT, RIGHT, SIGHT and TIGHT all end in IGHT. BATCH, CATCH, MATCH, PATCH and WATCH all end in ATCH. Because the members share four of their five letters, feedback that confirms the shared ending does almost nothing to tell them apart. The distinguishing information has to be bought one letter at a time.

In normal mode that is a nuisance. In hard mode it is a trap. Hard mode forces every guess to keep confirmed greens in place, so once IGHT is locked in, every legal guess is itself a member of the family. Suppose your first two guesses reveal the ending while saying nothing about the eight possible first letters. You now hold four turns and eight candidates, at one candidate per turn. If the answer sits in the wrong half of the list, the game is lost with perfect play, because no better line ever existed. Our hard mode guide covers how to lower the odds of walking in.

A normal-mode computer simply refuses to play that game. Freed from the reuse rule, it burns a turn on a probe such as FELTS, which tests F, E, L and S against the family in one guess, then MANOR, which tests M, N and R. Between them the two probes separate all eight candidates, and the answer falls on the next turn with a guess to spare. That escape hatch is why trap families cost a normal-mode solver a little efficiency and almost never the game.

So, can it always win?

In normal mode, for every practical purpose, yes. The opener alone buys more than half of the bits the puzzle starts with, free probing collapses every family of look-alikes, and we know of no realistic line that defeats a solver which keeps maximising information across the full guess list. In hard mode the guarantee gets shakier. The IGHT family alone can hold more members than you have turns remaining when you meet it late, and the rules forbid the probes that would split it. Whether perfect play from guess one can always steer clear of that corner on our curated list is a question only a full simulation can settle. What the data already shows is the mechanism: the trap is real, a computer wins the overwhelming majority of hard-mode games, and the games it might drop are corners the format builds, not mistakes the machine makes.