# What is the best guess for Wordle? – How to play Wordle

Sean Plays

Views: 1064

Like: 30

Hi everyone!

I took some time to develop a computer algorithm to finally answer the question everyone has been asking–“What word should you start with when playing Wordle.”

I took some time to develop a computer algorithm to finally answer the question everyone has been asking–“What word should you start with when playing Wordle.”

The answer may surprise you. I also delve into why I think other options can be just as good as an optimized computer strategy

13.01.2022

I have my own wordle algorithm with a similar approach to scoredle (we just tweaked ~50% of the code to optimize for as few guesses as possible instead of information)CRATE has consistently been shown as the best starter, though in the very first version the best word it showed was SOARE.We've coded it to sort all words into a hidden list from best to worst when starting with (it plays all 2315 answers with each word, then averages the guesses+total information into a combined score)Here's the current top 25 best word list, as of the June 2nd version of the model;1. CRATE2. TRACE3. CRANE4. SOARE5. RAILE6. RAISE7. SLATE8. SLANE9. LATER10. LASER11. TRAPE12. SNARE13. STARE14. ROATE15. ORATE16. CLEAR17. SANER18. CATER19. RAINE20. IRATE21. ARIES22. AROSE23. SCARE24. ARISE25. TRADESome common ones lower on the list include;LEAST (28)STALE (29)RATIO (39)SALET (41)CANOE (56)TEARY (66)ADIEU (854)AUDIO (2123)The first word using a letter in the bottom 10 letters (G, M, B, F, K, W, V, X, Z, Q, J) to show up is GLARE, followed closely by LATKE and GRATE

Wait, doesn't this prioritize first letters more than last?

In the official NY Times "WORDLE" version, "S" is never used to make a 4-letter word plural. So, the odds of having an actual word end in "S" is very low.

This is the first word I used to try and beat Scoredle's Tares.IT WORKED

Hey dude, very interesting video, my current starting word is OCEAN for the number of vowels, and I feel like it is quite decent. I guess.

If there ever is a 6 letter Wordle, I will definitely use the word "rustle" or "rattle"

spare….fight

Congrats on this bud…fun stuff cheers, Mike

I went for "first two words", based solely on frequency of letters and memorability. SONIC ALERT.

When I apply this strategy to the full combined word list of nearly 13k words, I also get LARES as the optimal first guess.

However, when using the same nearly 13k words as guesses, but only look the resulting distribution of the subset of 2315 possible solutions to determine the optimal guess, the result is ROATE instead.

This takes about 1-2 seconds to calculate using parallelized Rust (6-9 seconds for the combined list). Code is not ready to share yet, but I'll likely put it on GH eventually. Edit: lowered calculation times, because I underestimated the power of Rust's –release flag.

You're probably aware that trying multiple words that cover unique frequent characters is a very stable (if possibly a bit boring) strategy. There are many recommendations as for the words to use. The set I was first exposed to at https://youtu.be/l92g6Yy8t5g are TUBES, FLING, CHAMP, WORDY, but there are many such sets on Reddit.

When I try these four words on all 2315 possible solutions I get this result: average 3.238 attempts; 0 failed games

I calculated my own set of good words to start (ROATE, LINDS, CHUMP, GAWKY) as follows: every possible guess (12972 total) is tried on all viable remaining solutions, and the guess that leaves the smallest viable remaining solution pool on average is considered the best. That guess is ROATE for the first round, it leaves an average of about 60 words for the next guess. Rinse and repeat to find the next best guesses. With these four words I get: average attempts = 3.060; 0 failed games.

Initially I had BEFIT as a 5th guess, but using it is a little worse: average attempts = 3.067; 2 (0.086%) failed games.

I should note that my simulator picks a random viable solution if all of them can be tried within the guesses that are left, regardless if not all of the fixed words were tried. The results vary a tiny bit from run to run because of this randomness.

Here's an example:

2315 solutions left, 1. guess 'roate', hint '🟨🟨⬛⬛🟩', secret 'prove'

19 solutions left, 2. guess 'linds', hint '⬛⬛⬛⬛⬛', secret 'prove'

8 solutions left, 3. guess 'chump', hint '⬛⬛⬛⬛🟨', secret 'prove'

3 solutions left, 4. guess 'grope', hint '⬛🟩🟩🟨🟩', secret 'prove' <- random wrong guess

2 solutions left, 5. guess 'probe', hint '🟩🟩🟩⬛🟩', secret 'prove' <- random wrong guess

After 5 guesses: The only word left in the list is 'prove'

I wrote a program to evaluate the maximal size of the word bank after the first word is chosen (Which should represent the worst case scenario). The result I got was 'Raise' with a score of 177. "Lares" scored 220, which is significantly worse (the points represent words left in the word bank that are still valid choices for the second guess, so lower is better)

. Another note is that my starting word bank, may be different than yours, but I'd imagine the results would be consistent as the gap is fairly large if I were to use a different word set).

Type these 4 words. TUBES, FLING, WORDY, CHAMP. You'll get the all the alphabets. Re-arange them and you'll have your word.

My starting word progressed from "arose" to "slate" to "sauce", each of which is the best according to a particular strategy that I use in my solver. I've been trying to progressively improve it ever since I wrote it last weekend.Your strategy sounds convincing though, need to think about/try it. Cheers for the vid!

Great video, and it makes sense about the vowel sounds and the way the brain works. I think ALOES is my new starter 🌱

Awesome vid, I decided on using NATES first, which has the 5 most common letters excluding vowels and in good placements, and im sure has a similar SD to the top words you listed, and on a complete miss scenario, use CHOIR next.

i read on twitter that a good first 1-2 combo is Irate and Clons — the past two days this has been a helpful starting point.

Awesome video! I am hosting a Wordle AI challenge and wanted to implement your approach for reference. But the performance is really bad in my implementation (probably O(n^2)?) . For the official 12000 long word list it takes me 7 hours to compute just the first guess.

Did you have significantly better performance or did you just wait for a long time?

A nice video. Thank you for explaining and all the analysis about best and worst words starting wordle. I learned about wordle from a youtube channel in which suggested that a good strategy from solving wordle is to use siren and octal as the first two tries in wordle. What do you think?

ASTER or TEARS. Most popular letters in the English language are E, S, T, A, R, O, L, N, I, C, M, Y. His suggestion is a good one since it hits 3 vowels, but wastes a spot with G. 5 letter words are unlikely to have a G. LYRES would be another good one – trading the A for Y but getting the S in slot 5, and while S and T are the most popular consonants, R and L are not far behind. E and A, followed by O are the most common vowels. Fun game!

Thanks for the video – really interesting. I've only discovered Wordle today therefore have only played it once. With no 1st word strategy, I managed to solve the puzzle on my second guess (opening word BLACK, solution CRANK). Your video got me thinking (I'm no mathematician) that the "mathematical approach" will guarantee a solution within 'x' number of goes and will give the solution in an average of 'y' number of goes but I don't know what numbers 'x' and 'y' are. I do know that the game limits the number of goes to '6'. I'm supposing that x is greater than 6 (please tell me if I'm right or wrong) and so to solve the puzzle within the game's constraints, the "human approach" to thinking of words/making guesses will need to outperform the "mathematical approach". My thinking on this is that the human brain will need to rely on finding "fits" (and luck) rather than working on a process of elimination and that therefore a first word will not need to purely give the best efficiency in elimination but give the best clues to finding fits linguistically. But then I think, is finding fits really just a bad way of elimination.

Yours, Confused of UK

My trick has been to try RAISE COUNT as the first two words. I find the phrase easy to remember because so many health disorders need this fix.