Can ChatGPT play chess?
Not reliably. Large language models predict text, not legal moves, so they drop pieces, hallucinate positions, and lose the board over a long game. Where they shine is explanation: opening drills, tactical motifs, and curated puzzles checked by a real engine. Treat the AI as a coach with a referee, not a full-strength opponent.
Can ChatGPT actually play a full game of chess?
Not to the end, not reliably. In the opening it plays plausible book moves, because millions of annotated games taught it what a normal first ten moves look like. Past that point it starts producing illegal moves, resurrecting captured pieces, and forgetting where things stand — because it has no internal board, only a stream of text describing one. The further a game gets from memorized theory, the faster it falls apart.
Why the numbers punish it: a chess game has on the order of 1043 legal positions and roughly 10120 possible games. A real engine navigates that space with an explicit board and search. A language model stores none of it — it re-derives the position from the written move list every single turn, so one misread compounds into a game that has silently diverged from the real one.
Why do language models lose the board?
They lose the board because a language model predicts the next token from prior text, with no persistent representation of the sixty-four squares. Each move is re-inferred from the notation written so far, so there is nothing to catch the moment a bishop is quietly teleported or a pinned piece "moves" anyway. Nothing in the objective rewards a legal move over a fluent-looking one; it rewards text that reads like chess. That's fine for describing a game and fatal for playing one.
What is ChatGPT actually good at in chess?
It's good at explaining, not calculating. Ask it to name the idea behind the Italian Game, walk through why a fork wins material, or talk you through a plan in a fixed position, and it draws on an enormous library of chess prose to teach clearly. The strengths are real and narrow: opening ideas rather than deep lines, tactical motifs (pins, forks, skewers, back-rank mates), and drilling positions where the correct answer is already known and can't be argued with.
How do you get honest chess practice from an AI?
You add a referee — a real engine, or a pre-sealed answer key the AI cannot overrule — so the coaching stays sound even when the model's confidence outruns its accuracy. That's the design behind Counterplay: every puzzle is checked by an engine before it ships, and the AI's job is to explain the verified solution, not to invent one on the spot. It's an openings coach and a tactics trainer, honest about being a teacher rather than a grandmaster opponent. The rest of the chess and strategy shelf follows the same rule: use the AI for what it's genuinely good at, and let a checker handle what it isn't.
Frequently asked questions
- Can ChatGPT play chess?
- It can imitate chess and play a reasonable opening, but it is not a reliable opponent. Because a language model predicts text rather than tracking a board, it starts producing illegal moves and losing pieces once a game leaves known theory. It's useful for coaching and explanation, not for a full, fair game to checkmate.
- Is ChatGPT good at chess puzzles?
- Not on its own — it can confidently give a wrong solution and then justify it, because nothing forces the position to be legal or the line to be sound. Give it a fixed answer key that was worked out and checked in advance, though, and it becomes a good explainer of why the right move works. That is exactly how Counterplay's puzzles are built.
- What is ChatGPT's chess Elo?
- There is no stable rating, because it plays illegal moves and its strength collapses as the game lengthens. Estimates that circulate online come from short games in opening territory and don't survive to an endgame. Treating any single Elo number as its 'chess strength' is misleading — the failure is structural, not a rating.
- Do Claude or Gemini play chess better than ChatGPT?
- All three share the same core limitation: they predict text and have no internal board, so all three drop pieces in long games. Differences between them are marginal next to that. The honest fix is the same for every model — add a referee. Counterplay uses an engine to check every puzzle so the coaching stays sound.