The best Wordle starting word, according to information theory

What makes a good opener?

A good Wordle opener is one that narrows down the answer as much as possible, no matter what colour feedback you get back. Think about it this way: if you play a word and the result tells you almost nothing new about which of the remaining answers is correct, it was a poor guess. If it splits the answer space cleanly – ideally into many roughly equal-sized groups based on the possible feedback patterns – it was a great one.

This idea of "how much information does a guess give me" is exactly what entropy measures in information theory. The entropy of a guess is the expected number of bits of information you will learn from playing it, averaged across all possible answers. Higher entropy means more information, which means fewer guesses needed on average to reach the solution.

The formula is: for each of the 243 possible colour patterns a guess can produce (3 colours × 5 positions = 3⁵ = 243), calculate the fraction of remaining answers that would produce that pattern, then compute the Shannon entropy over those fractions. The result is a single number in bits.

The numbers

The LexiLab Wordle Solver runs exactly this calculation. It scores every word in the 14,855-word guessable list against the known answer pool and ranks them by entropy. These are not estimates or heuristics – they are exact computed values against that list.

Here are the top 10 openers by entropy score:

These scores are computed against the known answer pool using LexiLab's Wordle Solver engine – see The answer list below for what that means in practice. A perfect guesser with no prior information could expect to narrow the full answer pool down to a single candidate in about log₂(2350) ≈ 11.2 bits of information. Each guess at ~5.9 bits gets you more than halfway there in a single move. Words marked probe are archaic, obscure, or otherwise unlikely to be chosen as NYT answers – they appear in the guessable list but are not known to be part of the answer pool.

The awkward truth: the best opener is a word you've probably never heard of

TARSE. You may not know it. It is an archaic term for the tarsus (the ankle bones), and while it is valid in the Wordle guessable list, it has never appeared as a NYT answer and is unlikely ever to. The same goes for TIARE (a Pacific island flower), SOARE (an archaic word for a young hawk), ROATE, RAILE, REAST, and SALET – obscure enough that NYT's editors would almost certainly never choose them. They score well precisely because of their letter combinations – specifically the way T, R, S, A, E, I, O appear together – but no one is going to put ROATE in their Wordle tweet.

This is the fundamental tension in optimal Wordle strategy: the theoretically best openers are obscure words almost nobody knows, while the words people actually like to play (CRANE, ADIEU, AUDIO) score significantly lower.

The best opener you've actually heard of

For most players the real shortlist is not "which word has the highest raw entropy" but "which word is NYT actually likely to choose." And that matters a lot: a probe like TARSE or TIARE is almost certainly not going to be the solution, so starting with one all but removes any chance of the game ending on turn one. That 1/6 result – the shared score that makes people screenshot their Wordle grid – is only on the table if your opener is a word NYT is actually likely to choose. For many players that trade-off alone settles the question, regardless of any entropy calculation.

Once you restrict to familiar words that are plausibly part of the answer pool, the rankings look very different. RAISE leads at 5th overall (5.8776 bits), and the next familiar words in the top 10 are SLATE (8th, 5.8566 bits) and IRATE (10th, 5.8304 bits). The gap between RAISE and SLATE is smaller than any rounding error in the scoring – under 0.022 bits – and even a word like CRATE, which sits just outside the top 10, is within striking distance of all of them.

RAISE is the usual recommendation, and it earns it: five of the most information-rich letters in English (R, A, I, S, E), a word everyone knows, and one you will never have to explain to anyone. But SLATE and IRATE are just as strong. Pick whichever one you will actually stick with – consistency matters more than the tiny entropy differences between them.

Why not ADIEU or AUDIO?

ADIEU deserves some credit before we knock it. It is a genuinely reasonable strategy for casual play: loading up on four vowels almost guarantees you learn something useful from your first guess, which keeps you in the game and makes a failure very unlikely. As a "low-effort, high-floor" opener it has a lot going for it. The problem is that it stops there.

Knowing which vowels are present is only useful if the remaining consonants can be narrowed down efficiently too. ADIEU uses up four of your five letter slots on vowels (A, I, E, U), leaving only one consonant (D) – a poor return on a guess. The result is that ADIEU scores substantially lower than RAISE despite both being common words.

AUDIO has the same problem. Vowel-heavy words cluster the answer space inefficiently rather than splitting it evenly. The letters R, S, T, and L appear far more often in Wordle answers than U, O, and I together, so a guess that covers the former will produce more informative colour patterns than one that packs in the latter.

The chart below illustrates the difference. After playing RAISE or ADIEU, the colour feedback you receive places the remaining answers into groups. Each bar is one such group, sorted from smallest to largest. RAISE creates 132 distinct groups with an average of 17.5 answers each; ADIEU creates only 80 groups averaging 28.9 answers each. Fewer, larger groups means less information – and more work left to do on your next guess.

Does it matter which opener you use?

In purely statistical terms, the difference between RAISE (5.88 bits) and something like ADIEU (around 5.5 bits) is about 0.38 bits. Over a series of games that translates to roughly 0.05–0.1 fewer guesses per game on average. You are unlikely to notice it unless you are tracking your statistics obsessively.

What matters more is consistency. Any opener you play the same way every game allows you to build a mental map of the patterns – what it means when the S goes gray, what it means when the R goes yellow in position two. Switching openers regularly probably costs more than the small gain from using the theoretically optimal one.

The honest answer: play RAISE, play SLATE, or play IRATE. All three are within a rounding error of optimal among known answers, and any of them will outperform the vowel-loading strategy by a comfortable margin.

---

A note on methodology

These results use the known NYT Wordle answer pool (approximately 2,350 words – see The answer list below) and the 14,855-word guessable list. Entropy is calculated using Shannon's formula: H = –Σ p(x) log₂ p(x), where p(x) is the proportion of remaining answers that produce each distinct colour pattern x for a given guess.

The LexiLab solver uses a base-3 pattern hash and a pre-allocated shared array to compute all 243 possible patterns in a single tight loop without heap allocations. The precomputed opener rankings are generated by a build script and baked into the solver's startup data. The scores shown in this article are exact, not sampled or approximated.

Note that the solver's in-game suggestions include an answer-probability bonus on top of pure entropy – a candidate that is still a possible answer gets a small bonus reflecting the chance it is literally the answer. This does not affect the opening-guess rankings (all words start equally likely to be the answer), but it changes mid-game recommendations to favour "finisher" words over pure probes as the candidate pool shrinks.

The answer list

Wordle was created by Josh Wardle in 2021 with a fixed list of 2,315 answers pre-loaded in the game's source code. When the New York Times acquired Wordle in early 2022, they removed six words from the future answer sequence and moved to server-side daily selection. Since late 2022, NYT editors have chosen each day's answer from a curated pool – which means the pre-NYT list is no longer an accurate guide to what NYT will actually pick.

In practice this has two consequences. First, some words from the original list have never appeared as NYT answers and may never do so – the editors have shown a clear preference for common, everyday words over archaic or obscure ones. Second, new words that were never in the original list do appear: editors occasionally pick words that the original Wardle sequence never included, and no community-maintained list catches these until after the fact.

LexiLab runs a daily update that tracks confirmed NYT answers and folds any new words into our list automatically. The entropy scores in this article are computed against that known answer pool. The relative ranking of openers like RAISE and SLATE is stable across any plausible version of the answer pool – but no external source can know exactly which words NYT will choose in future.