Modern chess apps love numbers, and the most addictive of them is the accuracy score. Finish a game and you're told you played at 87% — or 61%, or a suspicious-feeling 94%. It's natural to want that number to go up. But it's also easy to misread it, and chasing it directly can make you a worse player rather than a better one.
This guide explains what an accuracy score really measures, exactly how the review on this site calculates it, and how to use it as a tool instead of a verdict.
What accuracy measures
In one sentence: accuracy measures how close your moves were to the best moves an engine could find, averaged over the game.
- Each of your moves is compared with the engine's preferred move in that position.
- If your move doesn't make your position worse, it scores close to 100.
- If it drops your chances significantly, it scores low.
- The game's accuracy is the average of those per-move scores.
What it does not measure: creativity, practical decision-making under time pressure, how difficult the position was, or your overall playing strength. It answers a narrow question: in this game, how much did you give away compared with perfect play?
How our review calculates it
After a game against the computer on our play page, tap Review game. A Stockfish engine running inside your own browser goes through every move. The calculation has three steps:
- Evaluate the position before and after each move, and convert each evaluation into a win probability between 0% and 100%. Working in win probability rather than raw pawn units means a mistake in a balanced position counts for more than the same mistake when you're already a queen up — which matches how games are actually decided.
- Measure the drop in win probability caused by each move.
- Score each move from that drop, and average the scores for the whole game.
Each move is also labeled by the size of the drop:
| Label | Drop in win probability |
|---|---|
| Best move | Matches the engine's choice |
| Excellent | Under 2 percentage points |
| Good | 2 to under 5 |
| Inaccuracy (?!) | 5 to under 10 |
| Mistake (?) | 10 to under 18 |
| Blunder (??) | 18 or more |
Tap any move in the list to see it on the board: a red arrow shows what you played, a green arrow the engine's better move. For more on what the labels mean in practice, see blunders, mistakes and inaccuracies.
To keep things fast on phones, the review spends a short, fixed amount of time on each position. That's plenty to catch real mistakes, but it's not deep analysis of every subtle choice.
Why different sites give different numbers
Every platform computes accuracy its own way: different engines, different depths, different formulas for converting evaluations into win probability, different thresholds for the labels. The same game can easily score 85% in one place and 78% in another.
So the rule is simple: only compare accuracy scores from the same tool — and ideally, only compare your own games over time.
Why 90% can still lose
It sounds paradoxical, but it happens all the time:
- One blunder can decide the game. Thirty near-perfect moves and one move that hangs the queen still average out high — but you've lost your queen.
- Short games are noisy. In a fifteen-move game, every move carries a lot of weight in the average.
- Simple positions inflate scores. When every reasonable move is fine, high accuracy says very little.
- And 60% can win, if your opponent's mistakes were bigger than yours.
That's why the first thing to look at after a game isn't the percentage. It's your biggest mistake.
Using accuracy productively
Find the one move that mattered
Each game, find the single worst move — usually a blunder or mistake — and understand why it was wrong. That habit alone produces steady improvement. Our guide on how to analyze your games walks through the process step by step.
Track trends, not single games
Individual scores bounce around. A rising average over your last ten or twenty games is a far more meaningful signal than any single result.
Count blunders instead of percentages
For beginners and club players, blunders per game is often a more useful number than accuracy. Going from three blunders a game to one is a huge improvement, even if your percentage only moves a few points.
Compare like with like
Accuracy in blitz will be lower than in slower games — that's normal. Compare fast games with fast games and slow with slow.
Common misconceptions
"High accuracy means I'm a strong player." Not necessarily. Easy positions and weak opposition produce high scores. Results over many games reflect strength far better.
"The engine is always right." Engines are vastly stronger than humans, but with a short time per move they occasionally misjudge complex positions. If a move is marked as a mistake and you can't see why, play out the alternatives on the board and check.
"Only the best move is good enough." Many positions have several moves of nearly equal value. "Excellent" and "Good" moves are perfectly fine choices.
"The computer opponent should score 100%." The review runs with different settings from the engine you played, and on easier levels the computer deliberately plays weaker moves so the game is fair. Its accuracy won't be perfect either.
A healthy routine
- Play a game at a time control that lets you think.
- Look at the result, then at the biggest mistake, not the percentage.
- Understand that one moment — what you missed and why.
- Check your accuracy last, and only as part of a longer trend.
More on reviewing games in the game review section.
Frequently asked questions
What's a good accuracy score?
There's no universal number, because every tool calculates it differently. On the same tool, comparing with your own past games is the fairest benchmark.
Why does the same game get different scores on different sites?
Different engines, depths, formulas and label thresholds. Only compare scores within one tool.
Is accuracy related to rating?
Loosely. Stronger players tend to have higher average accuracy, but you can't convert one number directly into the other.
Should I review every game?
Ideally, review at least the games you lost and any game with a big turning point. Even a two-minute review focused on one mistake is worthwhile.
Why does the engine opponent not have 100% accuracy?
Because the review uses different settings and time than the engine that played, and lower levels intentionally play weaker moves.
