What this research found

Can AI Beat Wordle?

Research topic and question

  • How many guesses does AI usually need to solve Wordle? Try it against a simple word-filtering algorithm using the same word list and answers. Who wins more often, and who needs fewer guesses?
  • Can you just play the games yourself here, with a program giving you the clues? I want to see how you do against the simple algorithm.

Research findings

I averaged 3.87 guesses; the simple algorithm averaged 3.80. Both solved all 30 games.

ResultMe (GPT-6-astra)Filtering algorithm
Solved within six guesses30/3030/30
Average guesses3.873.80
Median guesses44
Finished in fewer guesses6 games9 games

The other 15 games were ties. The algorithm used just two fewer guesses across the entire experiment—a small edge, not convincing evidence of general superiority.

Both used the same 2,315-word dictionary, randomly selected answers, and SLATE opener. I chose guesses from the clues without seeing hidden answers or program-filtered candidates. The algorithm eliminated incompatible words and favored common letters.

So, about four guesses for me in this pilot. These games shared one conversation; they don’t establish how AI generally performs.

How this research was produced

AIPOCH planned and ran this investigation end to end — searching the literature, producing the figures, and drafting the report. The full session transcript is available to inspect.

Share:

Run this kind of analysis on your own question

Start a session and see how an AI co-scientist accelerates your research. Pay-as-you-go, nosubscription required.