Can ChatGPT analyze chess games?

Can ChatGPT analyze chess games? Yes, when it has a game to review and a chess engine to check the concrete moves. With Chess Coach connected, you can bring a finished Chess.com or Lichess game into ChatGPT, see Stockfish's findings on an interactive board and ask why a move was a mistake. A fluent explanation on its own is not evidence that a position has been checked.

The useful division of work is simple: Stockfish calculates a position; ChatGPT explains the result and answers your questions. This guide uses a real finished Chess.com game to show what that review can look like, without presenting an invented conversation as a product demo.

Can ChatGPT analyze chess games reliably?

It helps to distinguish a model commenting on pasted moves from an assistant using an analysis tool.

A PGN is a text record of a game, including its moves and information such as the players and result. A FEN describes one position. These are different inputs: a position alone does not tell the whole story of a game (Chess.com's explanation of PGN and FEN, checked October 23, 2026).

For a full game review, start with the game rather than a screenshot and an invitation to guess. Even with an engine connected, read the explanation critically. The engine checks the positions and lines it analyzes; it does not verify every sentence the assistant writes.

How to analyze Chess.com games with ChatGPT, step by step

Here is the whole route with one real finished Chess.com game: PapZap (1328) vs nacsol3 (1317), January 1, 2026, a 10-minute rapid game from the founder's account. The positions and engine findings below come from Stockfish, through the same analysis code the connector uses. The wording around them is ours; it is not a copy of a ChatGPT reply.

1. Connect Chess Coach in ChatGPT. Open Chess Coach in OpenAI’s plugin directory, click Install plugin, connect Chess Coach and sign in. Then select @Chess Coach in a Work chat. The ChatGPT setup instructions explain the steps. Chess Coach never asks for your Chess.com password, only your public username.
2. Ask for your list of games. Try: "@Chess Coach, show my last 10 finished Chess.com games. My username is [username]." The list_games tool returns your recent finished games so ChatGPT can show them as a numbered list with the date, the opponent and their rating, the result, your colour, the time control and the opening. It can also search older games by month, opponent, result, colour, time control or opening.
3. Choose one game yourself. Answer with a number from the list, or paste the game link. Pick a game you remember. In this case, the choice was the loss to PapZap, a Nimzowitsch Defense where the game turned on move 13. Looking at the list does not use your daily review; the review starts when you choose a game.
4. Read the review on the board. The review_game tool runs Stockfish on the game and returns its key moments, numbered ①②③ on the interactive board, with a red arrow for the move played, a green arrow for a better move and an evaluation graph. ChatGPT explains those moments below the board, at a beginner level by default.
5. Stop at the biggest swing. In this game, the engine's analysis shows the game close to equal before Black's 13th move and completely winning for White after it. That is the position to study.

r . b . k . . r
. . . . p p b p
p . . p . . p .
. . p P . . . n
. q . . N . . .
. . . . . N . .
P P P B . P P P
R . . Q R . K .

The queen move looks active: it attacks White's knight in the center and a pawn with no defender. Stockfish's reply shows the problem. White's knight takes the pawn on d6 with check and, from there, also attacks the queen. That is a fork, one piece attacking two things at once. Black's king must move, and the knight then takes the queen. Black cannot take the knight with the pawn next to the king either: that pawn is pinned, meaning it cannot move without exposing the king to White's rook on the open line. White played the knight check, and the game ended there: Black abandoned it.

The engine's preferred move keeps the queen on a safer square, where it attacks an undefended pawn on the queen's side, and the position stays close to equal. A good follow-up question for ChatGPT is: "Before my queen moved, which enemy pieces could give check? Show the knight's reply on the board." Then ask it to compare the two queen moves, and end with one habit: before attacking two things, check which squares around your king the enemy knight can reach.

A first request you can adapt:

@Chess Coach, review this finished Chess.com game: [game link].
Use Stockfish to find three important moments. Explain the ideas for a beginner.
Start with the biggest mistake, show the position and ask what I was trying to do.

This is a suggested prompt, not a transcript of a connector response. The same distinction between a review and a follow-up matters on Lichess; the Claude analysis guide covers that assistant's workflow, and how to read Stockfish analysis explains the evaluation graph.

A second example: a capture that leaves the queen undefended

In ChocoteBABY vs nacsol3, January 11, 2024, Black captured a pawn with the queen on move 16. The game record contains that capture, White's queen taking it on the next move, and Black's resignation. It is also one of the examples in week 2 of the founder's chess diary.

r . . . k . . r
. . p . . . p p
. . n . q p . .
. p . . p . . b
. . . . . . . .
P P B P . N . P
. . . . . P P .
R . . Q R . K .

The capture attacks White's queen on d1, but Black's queen has no defender on b3. White can simply take it. Attacking a valuable piece does not force your opponent to move that piece if it can capture the attacker instead. Stockfish's preferred move in this position is kingside castling. This is an alternative found by the engine, not a move that Black played in the game.

That gives the follow-up a precise target. Instead of “How do I get better at chess?”, ask: “After my queen captures the pawn, which piece can take my queen? Show that reply on the board. Then check castling and explain what it changes.”

The lesson is a short check before making a capture: after moving this piece, who can take it, and what recaptures? That question comes from the verified position. No invented feelings, hidden intentions or sample AI answer are needed to make the example useful.

Useful questions after the first review

• “What immediate threat did I miss?” Keep the board at the relevant position and ask for the checked reply.
• “Why doesn't my attacking move work?” Compare the intended threat with the opponent's strongest response.
• “What if I had played this instead?” Specify the move and ask Stockfish to analyze it. A new continuation needs a new check.
• “What should I check before making this kind of capture again?” End with one practical habit tied to the mistake.

You can also use the board's practice mode to revisit mistakes. Try to answer before revealing the better move. For more ways to structure the conversation, see the ChatGPT chess coach prompts and how AI chess analysis works.

Using ChatGPT without a connector

You can supply a PGN together with positions and lines already checked in an engine. Ask ChatGPT to explain only those findings, and verify any extra move it introduces. This involves moving information between tools, but it keeps the source of the chess analysis clear. Do not treat a plausible paragraph as an engine result.

Chess Coach is available in Claude’s directory and OpenAI’s plugin directory. Open its listing, install it and sign in with your Chess Coach account. The Free Chess Coach plan includes one in-depth review per day; the documentation lists the current limits.

Use all of this after the game has finished. Chess.com forbids tools that analyze positions during play, and Lichess also prohibits outside assistance (Chess.com Fair Play Policy, Lichess Terms of Service, checked October 23, 2026). Chess Coach is for understanding completed games, not choosing moves in a game against another player.