same feature is often very helpful when you tyop or missspell a word.
Some google system will be crunching the odds on
P(X=wordle | Y=worldle) / P(X=worldle | Y=worldle)
where
Y := observed sequence of characters the user typed in the box
X := user's intended search term (unobserved)
if we set priors on the intended search term from observed frequency in pages indexed by google, we get something like
P(X=wordle) : 1.4e9 / Z
P(X=worldle): 3.2e6 / Z
where Z is the number of words in the internet.
Bayes theorem gives
P(X=wordle | Y=worldle) P(Y=worldle | X=wordle) P(X=wordle)
----------------------- = ----------------------------------
P(X=worldle | Y=worldle) P(Y=worldle | X=worldle) P(X=worldle)
So if there's a 1/400 or higher chance that a user accidentally adds an extra L when trying to key in "wordle", and hits enter anyway, then the expression on the right will evaluate to > 1 which means we should infer the user actually meant X="wordle" even though they keyed in Y="worldle"