How Wordle’s Algorithm Actually Works (No Magic)

Wordle looks like it might be doing something clever — generating the daily word, evaluating your guesses, telling you which letters are right and where. Most of that is much simpler than it looks. The “algorithm” behind Wordle is two hardcoded word lists, a date-based index, and a string-comparison loop. Understanding how Wordle’s algorithm works takes about ten minutes, and once you see it, the bot strategies that minimize expected guesses become much less mysterious.

The answer list

The single most important thing to understand about Wordle is that the daily word is not generated. It’s looked up. The game ships with a hardcoded array of valid answers — somewhere around 2,300 words in the original Josh Wardle version — and selects today’s word by index.

The original answer list was visible in the source code on the page. Anyone who opened the browser developer console and inspected the JavaScript could see the entire ordered list of answers for the next several years. This was widely reported in 2022 and led to several internet “Wordle answer of the day” sites that just looked up the public array.

The list was curated by Josh Wardle and his partner before launch. They picked words that were common enough for most English speakers to know, avoided plurals that ended in S, avoided most obscure words, and removed words with offensive meanings.

The valid-guess list

Separately, Wordle accepts a much wider list of inputs as valid guesses — around 12,000 five-letter English words. This list is also hardcoded. If you type a five-letter combination that isn’t on either the answer list or the broader valid-guess list, the game rejects it with “Not in word list.”

The split is intentional. The answer list is curated for fairness and accessibility; the valid-guess list is broader to allow strategic guesses (low-frequency letters, awkward letter combinations) even when those words would never be selected as answers.

The date-to-answer mapping

The original Wordle calculated the day’s index by subtracting a fixed epoch date from today’s date and using the difference as an array index. The fixed epoch was June 19, 2021 — the day Wordle launched publicly. So Day 0 was the first puzzle, Day 1 was the next day’s, and so on.

This meant that anyone who reverse-engineered the JavaScript could trivially compute tomorrow’s word. It also meant that everyone on Earth playing on the same calendar day saw the same word, which was the central social hook — when you posted your daily score grid, everyone reading it had played the same puzzle.

After the NYT acquired Wordle in 2022, the date-to-answer mapping was moved server-side. The game now fetches the day’s word from the NYT’s servers rather than computing it from a public client-side index. The change made next-day-lookup harder but didn’t fundamentally alter the algorithm.

The feedback algorithm

When you submit a guess, Wordle compares it to the answer letter by letter and produces a five-color feedback string. The colors mean:

• Green — correct letter in the correct position.
• Yellow — correct letter in the wrong position.
• Gray — letter not in the answer (or already accounted for elsewhere).

The implementation is a two-pass loop:

1. First pass: mark all letters that match position exactly (greens). Track which answer-letters are now “used.”
2. Second pass: for each remaining guess letter, check whether it appears elsewhere in the answer among the still-unused letters. If yes, mark it yellow and consume one occurrence. If no, mark it gray.

The two-pass structure handles the tricky case of repeated letters correctly. If the answer is ROBOT and you guess BOOST, the two O’s in your guess are evaluated against the two O’s in the answer — one is yellow (correct letter, wrong position), one is gray (no remaining occurrence to match), and the order matters because positional matches are resolved first.

Why the algorithm matters for strategy

Because the answer list is fixed and known, optimal Wordle strategy is a solved problem. You can enumerate the entire ~2,300-word answer space and compute, for any given starting word, the expected number of remaining candidates after each possible feedback outcome.

Mathematician Grant Sanderson covered this in detail in a 3Blue1Brown video titled “Solving Wordle using information theory,” which showed how to compute the expected information gain of each possible guess and rank starting words by their average performance. The video became a popular reference in the Wordle bot community.

The general result is that good starting words have high coverage of common letters (E, A, R, O, T, L, S, N, I) and avoid letter repetition (which wastes feedback). Common consensus starters in optimization-minded play include SLATE, CRANE, TRACE, ADIEU, and SOARE. Optimal starters can solve the curated answer list in an average of about 3.4 guesses with perfect play.

For a deeper guide to starting words specifically, see our best Wordle starting words piece.

How bots solve Wordle

A Wordle solver bot works in two phases per turn:

1. Maintain a candidate set of answer-list words consistent with all feedback received so far.
2. Choose a next guess that maximizes expected information about which word in the candidate set is the answer.

The information-maximizing guess isn’t always a word from the remaining candidates. Sometimes you’ve narrowed the candidates to four or five possibilities, and the best move is to guess a word from the broader valid-guess list that contains letters distinguishing among those candidates — even though that guess can’t be the answer.

This is called a “probing” guess. Wordle players who don’t use bots often miss probing opportunities, defaulting to “guess one of the remaining candidates” even when probing would resolve the puzzle in fewer expected guesses.

The NYT era and curation changes

The New York Times acquired Wordle from Josh Wardle in January 2022 for an undisclosed seven-figure sum. The NYT initially kept the original answer list largely intact but began curating it more aggressively over time — removing words considered too obscure, too offensive, or too political for a general audience.

The NYT’s editorial team now maintains the answer queue, sometimes substituting different words from the broader pool when current events make a scheduled answer awkward. The substitution process isn’t public, but the queue itself is no longer simply a public hardcoded array.

The valid-guess list (the wider 12,000-word list of acceptable inputs) has remained largely unchanged.

Why the algorithm choices matter

Wordle’s design decisions look small but were intentional. The hardcoded answer list with a date index made the game synchronous globally — every player got the same word on the same day, which was the foundation of the social-sharing virality. The two-list separation (curated answers vs broader valid guesses) made the game fair without making it artificially restrictive. The five-letter length and six-guess limit produced a difficulty curve that rewarded skill without making the puzzle unsolvable.

The fact that the algorithm is deterministic and the answer list is finite is why Wordle works as a “shared experience” game in a way procedurally-generated puzzle games don’t. Wordle isn’t just a word game — it’s a synchronized one.

For more on why Wordle reached the cultural moment it did, see our why Wordle blew up piece.

What Wordle isn’t doing

Wordle isn’t using any AI, NLP, or generative model. There’s no “difficulty adjustment” based on your past performance. There’s no per-player word selection. The game’s complete logic could be implemented in a few hundred lines of JavaScript, and the social hook does most of the work.

This is part of why Wordle’s success is interesting — the game is so simple that anyone who looks at it asks “why didn’t I think of that?” The answer is usually that the obvious-in-hindsight idea required the social-sharing format Wordle invented (the colored-square score grid copy-paste) to find its audience.

For an algorithm-free break

If you’d rather not think about word lists and probability for a few minutes, the Chrome Dino game running on this site is the opposite kind of game — one obstacle, one button, no curated answer list. Both are well-designed in their own ways, and both reward the brain for completely different reasons.

Frequently asked questions

How does Wordle pick the daily word?

Wordle uses a hardcoded list of around 2,300 curated answers, indexed by the date relative to a fixed epoch (originally June 19, 2021). After the NYT acquisition, the date-to-answer mapping moved server-side, but the underlying mechanism — fixed list, date index — remains the same.

Is Wordle’s answer the same for everyone?

Yes. Every player gets the same word on the same calendar day. This is the foundation of Wordle’s social-sharing model — the colored-square score grids posted on social media are all from the same puzzle, which is what makes them shareable.

How many possible Wordle answers are there?

The curated answer list contains around 2,300 words. The broader valid-guess list (words you can type as a guess even though they won’t appear as answers) is around 12,000 words. The NYT has lightly modified both lists since acquiring the game.

What is the best Wordle starting word?

Information-theoretic analysis (notably 3Blue1Brown’s video on the topic) suggests starters like SLATE, CRANE, TRACE, and SOARE perform optimally on the curated answer list. The differences between top starters are small in practice — any word with high coverage of common letters and no repeats performs well.

Can Wordle be solved by a bot?

Yes, easily. Because the answer list is finite and known, a bot can maintain a candidate set and pick information-maximizing guesses each turn. Optimal bots solve the puzzle in an average of around 3.4 guesses, and reach the answer within six guesses on essentially 100% of the answer list.

The bottom line

Wordle’s algorithm is unglamorous — two word lists, a date index, and a comparison loop. The genius wasn’t in the code; it was in the curation, the share format, and the synchronous daily structure. Once you see how it works, the game becomes a tractable optimization problem. Whether you’d rather play it that way or just for fun is a personal choice.