The Best Wordle Opening Words Backed by Information Theory

Why "best opener" is a maths question

Every Wordle guess is a filter. A good opener is one that, on average across all 2,315 possible answers, leaves the smallest number of candidates remaining after the colours come back. That is exactly the definition of expected information gain in Shannon's information theory: the number of bits of uncertainty a guess is expected to remove.

A perfect first guess would split the remaining answer list into equal thirds (green/yellow/grey) for every letter — the "20 questions" ideal. No real English word does that, but a few come impressively close.

The consistent top-three

Running the numbers across the original NYT answer list yields three words that trade places at the top depending on tie-breaks:

• STARE — 5.89 bits of expected information. Covers S, T, A, R, E — five of the six most common Wordle letters.
• CRANE — 5.83 bits. 3Blue1Brown's grid-search video made this the most famous "solver-approved" opener.
• SLATE — 5.86 bits. Slightly stronger than CRANE if you weight only the answer list (not the full guess list).

All three share a design principle: five distinct high-frequency letters, one of which is a vowel other than E in a non-standard position. That last constraint is what beats vowel-heavy words like ADIEU (only 4.98 bits) — vowels alone don't help you place consonants.

Why AUDIO and ADIEU underperform

AUDIO looks appealing — four vowels! — but Wordle answers cluster around one or two vowels per word. Guessing four vowels wastes slots on letters you likely won't need in that position, and leaves you no information about R, S, T, L or N. Post-AUDIO you typically have 90+ candidates left; post-STARE, around 55.

Point-value check (Scrabble® scoring for context, though Wordle doesn't score by tile):

• STARE: S(1) + T(1) + A(1) + R(1) + E(1) = 5 points — all common tiles, which is the point.
• CRANE: C(3) + R(1) + A(1) + N(1) + E(1) = 7 points.
• AUDIO: A(1) + U(1) + D(2) + I(1) + O(1) = 6 points — cheap but information-poor.

Second-guess pairs beat single-guess optimisation

Once you commit to STARE, the best follow-up when nothing greens is typically CLOUD, DOILY or POINT. Together, STARE + POINT covers ten of the twelve most frequent letters (S T A R E P O I N T), leaving only a handful of answers alive before guess three. Pair play is why competitive solvers rehearse two-word combinations rather than hunting for a single magic opener.

Choosing your personal opener

The margin between STARE, CRANE and SLATE is under 0.1 bits — statistically noise. Pick the one you'll type without hesitation. What actually loses games is changing openers weekly: you never build intuition for how the yellows redistribute. Our Wordle solver lets you test a fixed opener against any target pattern to see how your typical grids evolve.

If you want to see the letter-frequency data yourself, our letter frequency tool breaks down every position in the answer list.

FAQs

Q: Is STARE actually the "mathematically best" opener? A: By expected-information under the standard NYT answer list it's within 0.05 bits of SLATE and CRANE. All three are optimal within measurement noise. What matters more is consistency and how you play guess two.

Q: Should I switch openers if I lose a game? A: No. Losing a Wordle with a strong opener is almost always about guess two or three, not the opener. Track where you lost bits, not what you opened with. See how to solve Wordle in three.

Q: Does the NYT curated list change the maths? A: Slightly — the NYT trimmed some obscure words. The top-three ordering shifts by fractions of a bit but the same words dominate. Full breakdown in the Wordle answer distribution.

A closer look at the maths

"Bits of information" sounds abstract, but the intuition is simple. If your opener splits the candidate list into 32 equally likely groups, you've gained 5 bits (2^5 = 32). If it splits into 64 groups, 6 bits. The 5.89 bits of STARE means the answer, after seeing your colours, sits in one of roughly 60 groups on average — down from the starting 2,315-word list.

The gap between "great openers" and "acceptable openers" is smaller than most think. RAISE scores 5.88 bits. AROSE scores 5.81. ROATE scores 5.88 but isn't in the NYT accepted list. Within about 0.2 bits of the top, dozens of words are competitive — pick one and stop optimising.

What makes a bad opener quantifiably bad

Openers with 4+ bits are respectable. Below 4 bits you're wasting a guess. Some genuinely poor choices:

• QUEUE — 3.2 bits. Repeated U and E kill information.
• FUZZY — 3.4 bits. Two Zs, three rare letters.
• MAMMA — 2.9 bits. Three Ms and only three distinct letters.

The lesson: five distinct letters, no repeats, one common vowel in slot 2 is the design pattern. Repeat letters cost information roughly linearly — a doubled letter is about 1 bit of loss.

Regional and personal variation

If you play a variant of Wordle with a different answer list (Absurdle, Dordle, or region-specific clones), rerun the numbers. Absurdle in particular rewards openers that split the worst-case candidate group evenly, not the average — RAISE and TRACE slightly outperform STARE there.

For personal play, tracking your solve-guess distribution over 50 games with a fixed opener tells you more than switching openers ever will. If you average 4.1 with STARE across 50 games, that's your true baseline. Switching to CRANE for a week won't drop you to 3.9 — you'll just have noisy data.