WordleBot F.A.Q. (2024)

What is WordleBot?

WordleBot is software programmed to solve Wordles in as few steps as possible on average. It does this by starting with a large list of relatively common five-letter words, and using information theory to eliminate as many options as possible with each successive guess.

Does WordleBot know the answers?

Nope. WordleBot plays no role in selecting Wordle solutions. Every guess it makes is solely based on the clues provided by the puzzle and its own dictionary derived from word frequency data. See below for more details on how the bot calculates solution probabilities.

What do these scores mean?

Skill is a measure of efficiency — how closely your guesses came to solving the puzzle in as few turns as possible averaged over all possible Wordle solutions, not based on the word that happened to be the solution that day (which we can know only in hindsight).

Luck is how fortunate you were. A number closer to 99 means things have gone your way, and closer to zero means it’s not your day.

The skill and luck measurements are independent of each other. You can be both lucky and skillful; lucky but not skillful; skillful but not lucky; and neither lucky nor skillful.

So if the bot comments that you’re lucky, don’t take offense — it doesn’t necessarily mean you’re not skillful.

Fundamentally, WordleBot measures your decision-making, not the result. This is far from easy to grasp because humans tend to put too much weight on outcomes instead of the quality of decisions at the time they’re made. Consider that solving in one turn with JAZZY would be the best possible result and yet would be an awful choice for a starter word. The same is true for the rest of your guesses: A good result doesn’t mean you’ve made a skilled decision, and a poor result doesn’t mean you’ve made a poor decision.

You can read more details on the mathematical calculations behind skill and luck below.

Should I try to beat the bot?

We don’t think of the bot as being in competition with you — we prefer to view it as a kind of coach. Luck is part of Wordle, so it’s always possible to solve faster than the bot. But the bot’s reliance on math instead of intuition means in the long run its paths to the answer are hard to top. There’s a lot to be learned from the way the bot plays, similar to the way top chess players train by using chess programs.

That said, some Wordle players enjoy testing their skills or just comparing their results with the bot’s — if seeing whether you “beat” the bot each day brings you joy, go for it.

At the very least, you can always use your scores to prove that you’re better at Wordle than your friends and family.

How can the bot help my game?

No one expects you to play exactly like WordleBot. But there are a few things the bot does that can improve your ability to solve the puzzle faster, or more reliably. Try to take note of the bot’s recommendations for second words based on the word you start with. It can provide insight into what letters or letter positions will eliminate as many of the remaining answers as possible. Another thing you can do that’s bot-like: If you reach the point where you think there are fewer than 10 solutions left, try to find a word that distinguishes them. Often, this means guessing one that contains as many letters in common among them as possible.

Why did the bot recently switch its opening word?

WordleBot chooses its opener first by estimating how likely it is for each word in its dictionary to be a solution, and then playing every possible Wordle game using every possible starting word. It then uses whichever opening word allows it to solve in the fewest guesses on average, while also minimizing its chances of failing to solve the puzzle.

These values change over time: partly as we make small adjustments to the bot’s dictionary (more on the dictionary here), and partly as the bot keeps playing and sees what kinds of words are most likely to be used as Wordle answers.

Those small changes can sometimes result in a new opening word for the bot. Though to be clear, the top opening words are so close to one another in terms of expected efficiency that it barely makes a difference. But the bot has no sentimental attachment to any opening word, and will take any tiny advantage it can get.

How does WordleBot estimate the probability of each solution?

The original Wordle game ran off a 2,309-word “solution list" that is no longer the official reference.

The bot's probabilities are estimated based on how common a word is, using the frequency of appearances in The Times since 2000 as a rough and imperfect proxy. This means the bot, just like humans, has to guess whether borderline words will wind up on the new (hidden) Wordle solution list. Consequently, the bot's skill scores can be viewed only as approximations when it comes to Wordle itself. The skill scores are, in a sense, a separate game.

Once words reach a certain floor of NYT frequency, they are considered “common” and the bot picks them randomly if they are matched up against each other with two words remaining.

In addition, plural forms of three- or four-letter words that end in “s” or “es” are not considered possible solutions. Past-tense verbs ending in “ed” are considered unlikely to be solutions, with the exception of past-tense versions of irregular verbs like cry, spy or fry.

The word I guessed was accepted by Wordle — why is the bot telling me it’s probably not a solution?

There are almost 15,000 five-letter words in the Wordle dictionary that a player is allowed to guess. Some of these — SYBBE, YCOND, IMMEW — are rather obscure. Others could be considered obscene. Still others are just in bad taste. The bot gives more than 10,000 of these words a probability of zero.

Unlike human players, who are free to guess any of these words, we’ve limited the bot to a smaller set of roughly 4,500 relatively common words to choose from when making guesses and recommendations. This means humans can sometimes outperform the bot by using a word that is off-limits to the bot but that is a better winnower. (You may notice that the bot is then forced to acknowledge that word’s primacy in its analysis.)

Whether a word is obscure is, of course, a judgment call, and many words are open to interpretation. But keep in mind that Wordle is fundamentally a game of common words. When deciding whether a word belongs in the bot’s dictionary, we look at frequency of usage in The Times, and we crosscheck that with Google Ngram data as well as the guess frequency of that word among Wordle players themselves.

Also, when you see that a certain number of players faced the same “scenario” as you, this means players faced the same set of remaining solutions while being on the same turn. It’s not about which particular path players took to get there.

15,000 words? 4,500 words? What are these different lists?

Here’s a cheat sheet:

15,000: The words that Wordle accepts as guesses, most of them very obscure.

4,500: The subset of words the bot uses for suggestions to readers in its analysis. They’re supposed to be reasonably familiar, and they include words that the bot knows can’t be solutions, like plural words ending in “s” or “es,” but that can be helpful as guesses.

3,200: The words the bot considers plausible solutions, with probabilities assigned to each one based on Times usage frequency. No plurals here, for example.

2,309: The old, original list of solutions. No longer consulted by the bot in any way.

Why doesn’t the bot analyze my game using “hard mode”?

Some people play in hard mode style but don’t realize it, causing the bot’s suggestions to seem nonsensical.

For the bot to analyze your Wordle with the constraints of hard mode, go to Wordle itself and look for the gear icon at top right. Click on it, then slide your preference to hard mode; this will tell the bot to rate your game in that style.

Alternatively, you can upload a screenshot of your result and tell the bot — after you’re prompted — that you played in hard mode.

The bot’s preferred starter in hard mode is PLATE, not CRANE (the standard mode starter), reflecting the differences in style of play. The bot has already failed to solve several times in hard mode. It is not expected to fail in standard mode.

One note specific to hard mode: Humans can sometimes outperform the bot in avoiding hard mode traps. We prioritize giving players instantaneous results (even on the tiniest phones) over the extra seconds it would take to improve the bot’s calculations in these cases.

I have hard mode selected in Wordle, but the bot reused a letter that had already been eliminated on a previous turn.

The rules of hard mode require you to reuse any revealed yellow or green squares but don’t actually prevent you from re-using eliminated letters. So, even in hard mode, any player, either human or bot, can reuse gray squares as much as they want.

Why didn’t the bot reuse a green square on its second guess?

Let’s take as an example the Wordle from Feb. 1. The solution that day was VERGE.

WordleBot opens in standard mode by guessing TRACE, and gets a yellow in the second spot and a green in the fifth. At this point, there are 42 possible solutions left in the bot’s dictionary. The odds would be against it if it went for a solution here, so instead it tries to maximize the information gained. Since it already knows the fifth letter is E, there’s nothing to be gained by guessing a word that ends in E.

Instead, it guesses SHRUG.

Why does the bot sometimes try a yellow square in multiple spots?

Sometimes — often on its second guess and often when vowels are concerned — the most information can be gained by nailing down the location of a particular letter or finding out if the answer contains more than one of that letter. In those cases, the bot will prioritize determining a letter’s location over guessing a wider range of new letters.

Sometimes the bot suggests a better guess than the one I made, but it’s the solution word. Isn’t that obvious and unhelpful?

Let’s say the solution is SLANT. The bot is saying that of all the possible solutions remaining at the time of decision, SLANT was the best possible choice as a winnower. Yes, it just so happened that SLANT wound up being the solution this time, but even if SLANT had not been the solution, it would still have been the best choice.

So far Wordle solutions haven’t repeated. Does the bot take that into account?

No, it does not. This is a disadvantage for the bot against players who keep track of which words have already appeared as Wordle solutions, but presumably solutions will repeat at some point.

What does the skill measure actually mean?

In general as a solver, you want your guesses to divide the possible solutions into as many groups as possible (each with a unique pattern of colored squares). This reduces the chances you are left with a coin flip among equally possible solutions.

Your skill score measures how close you were to the bot’s chosen word relative to the worst word you could have picked for that turn. You’ll see that the bot’s top picks for each guess (99 skill scores) will be next to a golden checkmark, which represents its seal of approval.

If your guess doesn’t reduce the number of possible solutions remaining, the bot assigns it a zero. This can happen even if you seem to gain information about the solution word on your turn. A basic piece of guidance: If you sense there’s only one word left, try as hard as you can to solve — no half-measures.

If you prioritize solving in two steps (accepting the risk that you will get more fives), the bot’s skill scores will be a little less relevant to you because its goal is to solve in the fewest steps on average.

You can click on "Show Details" to see the possible remaining answers and their probabilities at each step.

I reduced the number of solutions remaining, but I still got a skill score of zero?

This can happen later in a game, if the bot believes a certain guess would force you to rely on luck to solve the puzzle, compared with another guess where a win is assured.

Why doesn’t the bot count your opening guess toward your skill score?

WordleBot doesn’t count your first guess toward your overall skill score because we want you to feel free to choose a fun word without being penalized. So go ahead, start with FUZZY, we won’t stop you. (Note, the bot does count your opening luck score toward your overall luck score.)

What does the luck measure actually mean?

Broadly speaking, luck is whether the number of solutions you eliminated with a guess is more or less than what would be expected on average. (For the more mathematically inclined, it has to do with the expected reduction in entropy.)

Suppose, for example, you guess CRANE on your first turn. The best thing that could happen would be if the hidden word were actually CRANE — you would solve the Wordle in one guess. That’s obviously very lucky, and your luck score would reflect that.

The worst result would be five gray squares; you would be left with 356 possible solutions to sift through. Considering that, according to the bot, there’s an 88 percent chance that any given day’s Wordle solution will share at least one letter with CRANE, that result would be very unlucky, despite a strategically sound choice.

Can you tell me more about making “efficient” guesses?

There’s a common fallacy that so-called “rhyming” solutions are purely about luck, even in standard mode. Below is an example of efficiency in difficult circ*mstances. Suppose you’ve narrowed the possible solutions to five: BATCH, CATCH, LATCH, MATCH or PATCH. What should you guess next?

If you guess BATCH, one of two possible things will happen.

If the hidden word is BATCH, great! But if it isn’t — which is the more likely outcome — you’re stuck with four possible solutions.

A smarter guess would divide these groups more efficiently, putting you in a position to solve the puzzle regardless of luck. Here, WordleBot would guess BLIMP. See how that changes the picture:

With this approach, you divide the solutions into five groups of one word each. You’d be guaranteed to get the answer on your next turn!

What does “expected solutions after guess” mean?

It approximates how effective your guess is at that point in the game, regardless of what the answer might be. By contrast, “actual solutions after guess” is the number of words remaining once Wordle reveals which of your letters are green, yellow or gray.

What about “expected steps until solution”? This number reflects a forecasted total of future guesses. It does not count your opening guess, for example, after your first turn.

When calculating the average guesses for WordleBot users, what score do you give to those who didn’t solve the puzzle in six turns?

The bot uses the expected steps needed for WordleBot to solve the puzzle, based on the solutions left after a reader fails. If a reader eliminated every solution but one, the score would be 7. If more than one solution remained after six tries, the score would be somewhat higher.

Where can I learn more about the math that powers WordleBot?

Many mathematicians and programmers have tackled Wordle. We’d recommend watching Grant Sanderson’s 30-minute video on solving Wordle with information theory. His (shorter) follow-up video is also worth watching.