A chess game review gives you a lot at once: an accuracy score for both players, a graph that swings up and down, and a colored label on every move. It is easy to look at the number, feel good or bad about it, and close the tab. This guide explains what each part of a review measures, and how to spend ten minutes on it so you come away with one thing to practice.
What a game review measures
A review runs a chess engine, usually Stockfish, over every position in your game. For each position the engine gives an evaluation, which the review turns into a winning chance: how likely the side to move is to win from there, from 0% to 100%. An even position is about 50%. A position a piece up is well above 80%.
Every number and label in the review comes from one question: how much winning chance did this move give away? If you had a 60% chance before your move and 45% after it, the move cost you 15%. Measuring loss in winning chance, not in pawns, is why a one-pawn slip matters a lot in a level position and hardly at all when you are already a rook up.
The evaluation graph is the same curve drawn over the whole game. Steep drops are the moments that decided it.
Accuracy: what the number means
Accuracy summarizes how close your moves stayed to the engine's choice, from 0 to 100. In the free game review on this site, it is computed with the formula Lichess publishes: each move gets an accuracy from the winning chance it lost, and the game score combines a volatility-weighted mean with a harmonic mean, so a single blunder pulls the result down more than a simple average would.
Three things to keep in mind:
- Compare within one tool. Chess.com uses its own accuracy model, so the same game can score a few points differently there. Track your accuracy over many games in one place rather than comparing numbers across sites.
- Accuracy depends on the game. A quiet game with simple positions is easier to play accurately than a sharp one. A lower score in a complicated game can still be a well-played game.
- The result matters more than the score. You can lose with high accuracy if you made one blunder in a critical moment, and win with low accuracy if your opponent made more. Look at where the winning chance changed hands, not only at the total.
The move labels
Most labels come straight from the winning chance a move lost:
| Label | Winning chance lost |
|---|---|
| Best | The engine's top move, or practically equal to it |
| Excellent | Under 2% |
| Good | 2–5% |
| Inaccuracy (?!) | 5–10% |
| Mistake (?) | 10–20% |
| Blunder (??) | More than 20% |
A few labels describe special situations:
- Book: a known opening move. Nothing to review here.
- Forced: the only legal move.
- Great (!): you found the engine's best move in a position where the next best move was much worse. These are the moves where you had to see something.
- Brilliant (!!): you gave up material and the engine agrees it was one of the best moves, in a position that wasn't already won. A real brilliant is rare; many games have none.
- Miss (×): your opponent had just made a mistake or blunder, and your reply let them off the hook. Misses are worth studying as closely as blunders, because they are wins you didn't collect.
A 10-minute review routine
The engine finds the best move in a second. The point of reviewing is to learn why you didn't. This routine keeps the engine out of the way until it's useful.
- Write down your thoughts first (2 minutes). Before opening the review, note the moment you felt the game turn and what you were worried about. Your memory of the game fades fast, and it is the part an engine can't give you.
- Find the turning points (1 minute). Open the review and look at the evaluation graph. Click the two or three biggest drops. Ignore small wiggles; they rarely change the result.
- Try to find the better move yourself (3 minutes). At each turning point, hide the best move and look at the position again. Give yourself a minute per position. If you find it, you learned that you can see it with time, so the problem was time or attention. If you don't, the engine's move shows you a pattern you don't know yet.
- Name the mistake (2 minutes). Put each one in a plain category: missed a tactic, left a piece undefended, wrong plan, misjudged an endgame, ran short on time. After ten games the same category will keep coming up.
- Pick one lesson (2 minutes). Write one sentence you can use in your next game, such as "check every capture and check before I move" or "trade pieces when I'm ahead". One lesson you remember beats ten you forget.
If a turning point still doesn't make sense after the engine's move, that is what the AI coach is for: it reads the engine's analysis and explains up to six key moments in plain English, with what to practice next.
Which games to review
- Losses and draws you should have won. These have the clearest lessons.
- Longer time controls. A rapid or classical game shows how you think. A bullet game mostly shows how fast you move.
- Games you didn't understand. If you won and don't know why, review that too.
You don't need to review every game. Two or three careful reviews a week teach more than a quick look at twenty.
Common traps
- Looking only at the score. Accuracy is a summary, not a lesson.
- Agreeing with every engine move. If the best move needs a twelve-move line to justify it, note the idea and move on. Focus on moves a human could find.
- Fixing the opening only. Most games at club level are decided by tactics in the middlegame, not by move eight.
- Reviewing right after a bad loss. Wait ten minutes. You will read the game more honestly.
Try it on your last game
Pick the last game you lost and run the routine above. You can review it for free, with no sign-up and no daily limit:
- Chess.com game review: enter your username to load your recent games.
- Lichess game review: the same for Lichess, with brilliant, great and miss labels added.
- PGN analyzer: paste any game, including over-the-board games.
Stockfish runs in your browser, so a review takes seconds and costs nothing to run.