How to Analyze a Chess Game on the Analysis Board
- Export the PGN. On Chess.com, Lichess or any other site, open the finished game and use its Share, Export or Download option to copy the PGN. The tags (players, ratings, result) show above the board; clock comments are ignored.
- Paste it and pick a depth. Standard, depth 18, suits most games. Deeper settings, up to 30, find more in sharp positions but take longer, because every move is searched twice: once before it was played and once after.
- Press Analyze Game. Stockfish 19 runs in your browser and works through the game move by move. The progress bar shows how far it has got; a longer game or a deeper setting takes longer.
- Step through the result. Use the arrows or click any move in the list. Each move shows its label, its evaluation and, when it was not Best, Excellent or Great, the engine's move as a green arrow.
- Ask why. Explain Move on any move gives you the reasoning in plain English, which is the part an engine score cannot.
The board reads games, not single positions. For one position, set it up on the chess calculator, or run the engine on its own board with Stockfish online.
How to Read the Evaluation Bar
The evaluation is measured in pawns and always given from White's side. +1.00 means White is better by roughly a pawn's worth, −1.00 means the same for Black, and M means the engine has found a forced mate. Engines count in centipawns, hundredths of a pawn, so +1.00 is 100 centipawns. The bar fills toward the better side and is full at five pawns.
A pawn is not worth the same everywhere, so the move labels run on win probability instead. Strictly it is the expected score, with a draw counted as half a win. This is what each evaluation is worth to the side that has it:
| Evaluation | Win probability |
|---|
| 0.00 | 50% |
| +0.50 | 57% |
| +1.00 | 64% |
| +2.00 | 76% |
| +3.00 | 85% |
| +5.00 | 95% |
The curve flattens at the top. Going from 0.00 to −0.80 in a level position costs about 0.11 of your win probability, which is an Inaccuracy. The same 80 centipawns from +5.00 to +4.20 costs about 0.03, which only counts as Good: you were winning before the move and you are still winning after it.
What Each Move Label Means
Every move is judged by expected points lost: how much win probability it cost, from the table above. That is why a label here can disagree with a tool that judges by raw centipawns, and why the same slip is punished harder in a close game than in a won one.
| Label | Mark | Points lost | What it means |
|---|
| Brilliant | !! | — | Reserved for a sound sacrifice. Not awarded at the moment: telling a real sacrifice from a piece that simply hangs needs more than the engine's score, so the board leaves the label out rather than guess. |
| Great | ! | — | The engine's own move at a turning point: it won a position that was equal or worse (under +0.50 before, +4.00 or better after), or saved a lost one (worse than −3.00 before, −0.50 or better after). |
| Best | ★ | 0 – 0.01 | The engine's first choice, or within a hundredth of a point of it. |
| Excellent | + | ≤ 0.025 | Not the top move, but nothing measurable was given up. |
| Good | ✓ | ≤ 0.06 | A reasonable move that costs a little. |
| Inaccuracy | ?! | ≤ 0.12 | Enough to notice. Usually a plan that was slightly worse than the one available. |
| Mistake | ? | ≤ 0.25 | Up to a quarter of a point of win probability, gone. |
| Miss | × | > 0.10* | Your opponent blundered and you did not punish it. |
| Blunder | ?? | > 0.25 | The move that decided the game, or one of them. |
* Miss is conditional rather than a fixed band: it only applies on the move straight after your opponent blundered, which is why the same loss can be a Mistake in one position and a Miss in another.
Chess Game Review: Your Own Games, Move by Move
A game review is worth more when you do part of it before the engine does. Play through the game once on your own and write down two things: the move where you think it turned, and what you were afraid of at that point. Then run the analysis. Where your guess and the engine agree, you read the game well. Where they differ is the lesson.
Take Scholar's Mate, the short loss almost every player suffers once: 1.e4 e5 2.Bc4 Nc6 3.Qh5 Nf6?? 4.Qxf7#. Black's third move develops a piece and attacks the queen, so it looks natural, and that is exactly why a review is useful. The board labels 3...Nf6 a Blunder, the bar jumps to M for White, and the green arrow shows a move that meets the threat to f7, such as 3...g6, which blocks the queen's diagonal and attacks her, or 3...Qe7, which defends the pawn. The note to keep is not "Nf6 was bad" but "I did not check what the queen and bishop were both aiming at."
Real games are rarely that clean, but the method holds. Fix the blunders first, then the misses, and only then worry about inaccuracies. If the same mistake shows up in three reviews, it is a pattern, and patterns are what chess puzzles and the checkmate patterns lesson are for. For a longer plan, see how to get better at chess.
How to Read Your Own Game Report
The accuracy percentage is the headline, and it is the least useful number in the report. It rewards quiet positions where almost nothing can go wrong, so a dull draw scores higher than a sharp win. Two players can finish on 85% having played games of completely different quality.
Read the blunders instead, and read them in order. Most lost games are decided by one move, not by a slow accumulation, and the useful question is not which move it was, since the report already tells you, but what you were thinking when you played it. A blunder you can explain is a habit you can fix. A blunder that still looks reasonable after seeing the refutation is a gap in what you know, which is a different problem needing different work.
Inaccuracies are worth less of your time than their count suggests. A game with fourteen inaccuracies and no blunders is a game you played fine. Under about 1600, most lost points come from blunders, and time spent eliminating 0.05-point imprecision is time not spent on the moves that actually lose games.
What the AI Explanation Adds
An engine tells you that, say, 14.Rd1 was better than 14.Qe2 by 0.6 pawns. It does not tell you that the rook belonged on the open file, or that the queen move left the bishop on b2 without a defender. That second part is what you can use in your next game.
Explain Move sends the position before and after the move, its label and the game so far to our server. There a second Stockfish 19 checks the position, and an AI coach turns the result into a few plain-English sentences about what the move did and what the better move does instead. The same coach runs on the chess calculator and in Learn with AI.
Chess Analysis FAQ
How do I analyze a chess game?
Copy the game's PGN from the site you played on, paste it into the box on this page, choose a depth and press Analyze Game. Stockfish 19 evaluates the position before and after every move, labels each move from Best to Blunder and gives both players an accuracy score. Then step through the game with the arrows and press Explain Move on any move you want explained.
What is PGN format?
PGN (Portable Game Notation) is the standard text format for a chess game: tag lines such as [White "Carlsen"] and [Result "1-0"], followed by the moves in algebraic notation, like 1. e4 e5 2. Nf3 Nc6. Every major chess site can export it, usually from a Share, Export or Download option on the finished game.
What does the accuracy score mean?
It is a 0 to 100 score built from the average win probability each side lost per move: 100 × √(1 − 2 × average). Losing 0.05 per move on average scores about 95%, 0.10 about 89% and 0.20 about 77%. Quiet games score higher than sharp ones, so when you compare two games, compare their blunders rather than their accuracy.
Is the chess analyser free?
Full game review on this analysis board is part of Chess Next Move Pro: for free accounts, Analyze Game opens the upgrade screen. Single positions are free on the chess calculator, which runs Stockfish 19 at up to 800 ELO strength with 5 AI explanations a day and no account needed.
Can I analyze my online chess games?
Yes. Any game you can export as PGN works, whether it was rapid, blitz, bullet or daily, on Chess.com, Lichess or anywhere else. The clock times those sites write into the PGN as comments in curly brackets are ignored. Paste the main line only: side variations in round brackets get mixed into the game.
How can analysis help me improve?
It shows you which kind of mistake loses you games. Go through your blunders in order, work out what you were thinking when you played each one, and read the AI explanation of the better move. When the same pattern keeps coming back, such as a piece left undefended or a missed fork, drill it with chess puzzles until you see it without looking for it.
What is a chess game review?
A game review is a move-by-move check of a finished game against an engine. On this analysis board, Stockfish 19 grades every move, the evaluation bar shows where the game swung, a green arrow shows the engine's move wherever yours fell short, and an AI coach explains the difference in plain English.
How do I review my chess games move by move?
Play through the game once without the engine and note the move where you think it turned. Then run the analysis and step through it with the arrows. Compare your guess with the move the review flags: the gap between the two is usually the most useful thing you will learn from the game.
What depth should I analyze at?
Standard (depth 18) is the default and is plenty for club games. Each step deeper takes longer, and the cost adds up quickly, because every move in the game is searched twice: once before it was played and once after. Use 22 to 26 for sharp tactical games, and keep 28 to 30 for short games or a single critical sequence.
How do I read the evaluation bar?
The number is in pawns and always from White's side: +1.00 means White is better by about a pawn, −0.50 means Black is half a pawn better, and M means the engine has found a forced mate. The bar fills toward the side that is better and is full at five pawns, where the result is rarely in doubt.
What counts as a blunder?
A move that throws away more than 0.25 of your win probability. From a level position you start at 50%; a move that leaves you at −2.00 drops you to about 24%, a loss of 0.26, so it is a blunder. A Mistake loses more than 0.12 and up to 0.25, and an Inaccuracy more than 0.06 and up to 0.12.
Can I analyze a position from a FEN?
Not on this page: the analysis board reads whole games from a PGN. To analyze one position, open the chess next move calculator and drag the pieces into place, or paste the FEN there with Pro. While you review a game here, the copy button gives you the FEN of the current move, so any position can go straight into the calculator.
