Best and Worst Wordle Words

Soare. noun. A young hawk. If you're here for the "best" first word to use in Wordle, there it is.

What is Wordle?

Wordle is the word game craze that gained extreme popularity in January of 2021.

The game is pretty simple: guess a daily 5-letter word. You have six chances. If you guess incorrectly, the letters in your guess are highlighted green, yellow and gray:

• Green indicates that the letter was correct, and in the right position

• Yellow indicates that the letter appears in the word, but not in that position

• All incorrect letters appear in gray

  

In this post, let's look at the best and worst words to start with in Wordle.

The Strategy

Obviously some first guesses are better than others. Friends on Twitter or Facebook have recommended words like audio or rinse, since those contain very common letters. That makes sense - it's helpful to use words that yield a lot of yellow or green letters. A guess like xylyl (which is a real word!) will likely result in an all-gray, largely wasted guess and would only eliminate words that contain X's, Y's and L's.

So... what makes a good guess? In my opinion, a good guess not only gives you some correct letters, but eliminates the largest set of possible solutions. Good news! It turns out, those two goals (lots of correct letters and eliminates lots of possible solutions) go hand-in-hand - if you get a lot of correct letters, the number of remaining possible solutions becomes small.

Programming a Solution

If you look at Wordle's source code, you'll find two sets of words: possible solutions, and a dictionary of valid guesses you can make:

For convenience I've posted the possible solutions (of which there are 2,315), and valid guess words (12,972 words) in a GitHub repo.

Bad code warning

Disclaimer: My code is terribly inefficient (like, O(n^3) bad). I'm sure there's a better algorithm involving graphs and trees that runs in O(n log n) or whatever, but this runs in a few hours so it's fine...

Next, I wrote up some code. My script went something like this:

for each guess:
  for each possible solution:
    - Compute the "green-green-gray-yellow-gray" pattern that this
      guess and solution combination make
    - Figure out how many possible solutions are still valid given
      that pattern
    - Record that number

That gives you an idea of how "good" each guess is for any given solution.

Once you have a set of numbers for each guess, figure out the average number of solutions that are eliminated by guessing some particular word. Depending on the guess, the data had a pretty hard right or left-skew, so in my opinion median best represents the "average" here. Given some solution, thus-and-such guess should yield about N remaining options for solutions.

The Results

After letting my little Macbook churn on my script for 4 hours, I had results.

The top 10 best words to start out a game of Wordle are:

guess	Median words remaining	Mean words remaining	Standard deviation
SOARE	79.0	219.75	304.9
SAICE	80.0	232.78	318.85
SAINE	81.0	230.17	320.17
ARISE	91.0	254.45	355.75
RAISE	96.0	236.61	331.34
SANER	99.0	271.33	372.04
RAINE	102.0	253.77	349.27
SANTO	105.0	353.49	464.29
SATIN	105.0	347.2	465.33
SNARE	105.0	267.51	370.77

And the 10 absolute worst words to start a game of Wordle out with are:

guess	Median words remaining	Mean words remaining	Standard deviation
OXBOW	2077.0	1286.12	917.53
INFIX	2041.0	1164.58	900.65
IMMIX	2037.0	1354.85	873.22
ZIMBI	2029.0	1195.57	901.47
KUDZU	2026.0	1272.01	908.86
XYLYL	2026.0	1275.45	880.61
KIBBI	2021.0	1252.12	890.99
JUJUS	2016.0	1308.01	850.3
JUKUS	2008.0	1203.18	878.62
KUZUS	2001.0	1186.9	877.59

A full report is available here.

I mentioned that the data were skewed heavily right and left depending on what guess you had. Here's the histogram for soare:

You can see that for many solutions, guessing soare reduces the possible solutions from 2,315 to just a few dozen!