What Is Information Gain in Wordle? Bits and Halvings, Explained

Information gain is the number of times a guess halves the field of possible answers, measured in bits. Every Wordle guess comes back as a pattern of green, yellow and grey tiles, and that pattern rules out every answer that could not have produced it. A guess worth one bit halves the field on average. A guess worth six bits halves it six times over, dividing it by 2⁶, which is 64. The strongest opener in our data earns 6.22 bits against the curated answer list of 1,352 words, which is why good games are so often decided before the second guess is even typed.

One bit is one halving

Think of a bit as one perfect yes or no question. Ask whether the hidden word sits in the top half of an alphabetised answer list and either reply cuts your candidates in two. That is exactly one bit of information. Stack the questions and the field collapses quickly: one bit takes the 1,352 answers on the curated list down to 676, two bits to 338, and six bits all the way to about 21.

A Wordle guess is a bundle of those questions asked at once. Does the answer contain an A? Is there an E anywhere at all? Is the R sitting in position two? The tile pattern you get back answers all of them simultaneously, sorting every remaining answer into buckets, one bucket per possible pattern. When the buckets are many and even, whichever one you land in is small, and small is the whole point. Entropy, the formal name for expected information gain, is simply the average number of halvings a guess delivers across every answer it might face. Fractional bits are fine too: a gain of 6.22 bits is a division factor of roughly 75, because two raised to 6.22 is about 75.

The best opener, measured live

Our rankings score all 15,783 valid guesses against the 1,352 answers on the curated list, and TARES currently sits on top at 6.22 bits. The search covers every valid guess rather than just the answers, because the most informative probe does not need to be a plausible answer itself. It only needs to split the field well.

Convert 6.22 bits into a division factor and you get roughly 75. Divide the 1,352 possible answers by it and a typical game leaves about 18 candidates after one guess. Our tables also quote a plain average of 28 answers left, a little higher because averaging raw counts lets the occasional unlucky pattern, one that leaves a big clump of lookalike words, drag the figure up. Entropy averages the halvings instead, so it reflects the typical night rather than the worst one. Both readings tell the same story. One blind opening guess removes roughly 98 per cent of the field.

Notice what plays no part in that calculation: nothing about vowels, nothing about lucky letters, nothing about hunches. Information gain only cares how evenly a guess splits the answers, which is why some famous vowel-heavy openers rank surprisingly badly. We pulled that thread in should you guess vowels first.

How to use it in your own games

You do not need to compute anything at the breakfast table. The practical lessons fall out of the theory. Open with a proven high-information word from our ranked list of starting words rather than a sentimental favourite. Aim your early guesses at splitting the field, not at being right, because a probing second word that cannot be the answer often earns more than a hopeful stab. And treat grey tiles as wins, since ruling a common letter out of 1,352 answers deletes a large slice of them in one move.

When you want the full method, our entropy research walks through the maths and the complete opener tables. And to watch information gain at work on a real puzzle, the Wordle Solver recomputes it after every guess you enter, so you can see the candidate count fall in real time.