ROVER beat 43% of the players who met it.
Not a single letter in that word is unusual. R, O, V, E and R again — nothing you would call obscure. Yet it defeated almost half the people who drew it, while OKAPI, a word most players have never heard, defeated a statistically similar number.
Two completely different words, nearly identical failure rates, for opposite reasons. We went looking for why, using every completed game on this site.
What we measured
Every finished game on WordlyPlay between 25 April and 14 August 2026:
- 6,223 completed games
- 33,081 individual guesses
- 696 distinct answer words
- 1,762 different opening words
Players solved 83.7% of games, taking 4.41 guesses on average when they won. For the word-level statistics below we only used answers played at least 20 times, which leaves 146 words with enough data to say anything meaningful.
The words that beat the most players
| Word | Times played | Failure rate | Avg. guesses when solved |
|---|---|---|---|
| ROVER | 30 | 43.3% | 4.59 |
| OKAPI | 21 | 42.9% | 5.17 |
| BLAZE | 45 | 37.8% | 4.75 |
| JOLLY | 51 | 37.3% | 4.94 |
| JOKER | 42 | 35.7% | 5.07 |
| WATCH | 36 | 33.3% | 5.08 |
| JAZZY | 27 | 33.3% | 5.44 |
| TASTE | 43 | 32.6% | 4.31 |
| FIBER | 52 | 30.8% | 5.00 |
| EQUIP | 33 | 30.3% | 4.74 |
| QUEST | 43 | 30.2% | 4.33 |
| VIVID | 41 | 29.3% | 4.72 |
| UMBRA | 38 | 28.9% | 5.07 |
| VIGOR | 28 | 28.6% | 5.00 |
| ZESTY | 39 | 28.2% | 4.57 |
At the opposite end, four answers in our sample were never failed by anyone: LEMON, KNELT, PLANT and GRILL all recorded a 0% failure rate. MOUNT, HEART, GHOST, CRISP, TORCH and NOBLE all came in under 3%.
Look at those two lists side by side and the obvious explanation falls apart. JAZZY and EQUIP have awkward letters, fine. But ROVER, JOLLY, WATCH, TASTE and FIBER are built from the most ordinary letters in English — and KNELT, which nobody failed, contains a silent K.
The pattern: it's about how many words look almost the same
So we tested a different idea. For every word we counted its one-letter neighbours — how many other valid five-letter words you can produce by changing exactly one letter.
ROVER's neighbours include COVER, HOVER, LOVER, MOVER, ROVED, ROVES, RIVER and ROWER. Twenty-eight of them in total, measured against our 34,376-word dictionary.
The difference between the hardest and easiest words is not subtle:
| Group | Average one-letter neighbours |
|---|---|
| The 10 hardest words | 12.6 |
| The 10 easiest words | 5.5 |
Hard words have more than twice as many near-identical siblings. Individually the gap is starker still — ROVER has 28 neighbours, JOLLY has 23, JOKER has 21. KNELT and GHOST, neither of which anyone failed, have one each.
This is what actually ends a Wordle game. You are rarely stuck because you cannot find the letters. You are stuck because you found them all by guess three, and then burned guesses four, five and six choosing between COVER, HOVER, LOVER and ROVER with nothing left to tell them apart.
The average solve took 4.41 guesses. On a word like ROVER, players who solved it needed 4.59 — and 43% never got there at all. The letters were never the problem.
There are two kinds of hard word, and they need opposite responses
What convinced us the neighbour count was real is that it explains its own exceptions.
Trap words: too many lookalikes
ROVER, JOLLY, JOKER, WATCH, FIBER. Common letters, crowded neighbourhoods. You identify the shape of the answer early and then guess blind between candidates. The failure feels unfair, because you were close the entire time.
What to do: the moment you have a pattern with multiple candidates, stop guessing answers. If _OVER fits four words, do not try them one at a time — spend a guess on a word containing C, H, L and M. You will lose that guess and win the game.
Blind-spot words: nothing looks like them
OKAPI, EQUIP, JAZZY. These have almost no neighbours at all — OKAPI has zero. They are hard for the reverse reason: once you have the letters, only one word fits, but it is a word that never occurs to you. An okapi is a forest-dwelling relative of the giraffe.
What to do: when two guesses leave you with almost no confirmed letters, widen coverage rather than narrowing. Play a word with entirely fresh letters, including the ones you have been avoiding.
Both situations end in a lost game. Telling them apart by guess three is most of the skill.
Which opening words actually survive this
If crowded neighbourhoods are what kills games, the opener that matters is the one that leaves you with fewest ambiguous positions. Here is how the most-used openers on our site actually performed, ranked by guesses needed:
| Opening word | Times used | Win rate | Avg. guesses |
|---|---|---|---|
| AISLE | 33 | 97.0% | 3.69 |
| RAISE | 51 | 96.1% | 3.88 |
| TEARS | 69 | 94.2% | 3.95 |
| SLATE | 100 | 92.0% | 4.05 |
| AUDIO | 565 | 91.5% | 4.38 |
| CRANE | 225 | 78.2% | 4.39 |
| ADIEU | 143 | 91.6% | 4.52 |
| STARE | 136 | 80.9% | 4.65 |
CRANE is the most widely recommended opening word on the internet. On our site it produced an average of 4.39 guesses, against a site-wide average of 4.41 — statistically indistinguishable from opening with anything at all. Its win rate was 78.2%, the lowest in the table.
The vowel-heavy openers did better. AISLE, RAISE and ASIDE all carry three vowels, and all beat the consonant-loaded favourites. That fits the neighbourhood theory: vowels are what separate COVER from CAVER, and knowing them early is what stops you entering a trap blind.
We would not overstate this. AISLE was used 33 times and CRANE 225, so the CRANE figure is the more reliable of the two. But the direction is consistent across every vowel-heavy opener in the data, and it runs against the standard advice. If you want the theoretical case for the usual picks, our analysis of starting words by letter frequency makes it — this is what happened when real people used them.
Method and limitations
Figures cover games completed on WordlyPlay between 25 April and 14 August 2026. Word-level statistics use answers played at least 20 times (146 words); opener statistics use openers played at least 25 times. Neighbour counts were computed against the 34,376 five-letter words in our valid-guess dictionary.
Some caveats worth stating plainly. Sample sizes differ a lot between words, so treat individual percentages as indicative rather than precise — the 10-word group averages are the more reliable finding. Players choose their own difficulty, so word pools are not identical across games. And our answer list is our own rather than the New York Times', so specific words will differ from the official game, even though the structural pattern should not.
If you want to test any of this, every mode is free and needs no account. You can also see the words that have historically broken the most streaks for the wider picture.