WordleBot sets a new preferred opening word

Your Wordle starter word can make a big difference to your game. Photo / Getty Images

In the Upshot’s recent analysis of 515 million Wordle games played over a year, we noted that a few changes would soon be coming to WordleBot’s dictionary. As of Wednesday, those changes are now live — including a new opening word in both regular and hard mode.

WordleBot plays no role in the words selected to be Wordle solutions and doesn’t know the solutions. But, unlike human players, who are free to guess just about every five-letter word in the English language — 14,855 of them — we’ve limited the bot to a smaller set of roughly 4,500 relatively common words to choose from when making its guesses and recommendations.

We listened to feedback from New York Times readers who wanted certain words added (like BARRE and MOREL and MERCH) and certain ones deleted (PWNED has been owned, and EGADS to YEESH; all three are out). We also have new data on reader guesses that has helped us get a feel for which words are more familiar to more people.

Perhaps the most noticeable effect of all these changes will be that, starting today, the bot’s preferred opening guess has shifted — ever so slightly — to TRACE, from SLATE. And in hard mode, the bot’s favourite word is now neither TRACE, nor SLATE, but TROPE.

As always, the difference between these starter words (or, for that matter, between these words and the bot’s other favourites, CRATE, CARTE, CRANE and SLANT) is extremely small.

How small? On average, the bot expects to solve a Wordle in 0.0006 fewer guesses on average when it starts with TRACE rather than SLATE. If you play Wordle every day, it would take you more than 4 1/2 years to save one guess by opening with TRACE instead of SLATE. By comparison, ADIEU — the most common starting word among Wordle players — trails TRACE by about one-fifth of a guess, adding up to 74 extra turns for the bot over the course of a year.

In addition to reader guesses, we rely on data sources like usage frequency in the Times. All else being equal, the more common a word is (as measured by how frequently it’s appeared in the Times since 2000), the more likely it will be included in the bot’s dictionary.

This dictionary, in which each word is assigned a probability of being a solution, affects the skill scores assigned by the bot. (With two plausible words remaining, for example, we don’t want to penalise players with a zero skill score if they choose MERCH over MARCH.) The bot is making educated guesses at which words could be solutions just the way humans do, and it continues to learn as each daily Wordle solution is published, so there’s no guarantee that TRACE will be best for the Wordle solutions to come. But in our more expansive dictionary, it’s now the bot’s best bet.

This article originally appeared in The New York Times.

Written by: Josh Katz and Toni Monkovic

©2024 THE NEW YORK TIMES