Chess Game Review Explained: What Each Tool Actually Tells You

A chess game review is a post-game report that scores your accuracy, labels each move as good, inaccurate, a mistake or a blunder, and shows the line an engine preferred instead. Every major tool does that part now. What differs is whether you also get a plain-language reason the move was wrong, and something to practice before your next game.

Start with the honest answer for a lot of players: if all you want is the label-plus-best-line version, lichess already gives you that for free, with no daily cap and no paid tier of any kind, on every game you have ever played there. It is funded by donations, not subscriptions, and analysis has never sat behind a paywall on that site. For plenty of people that is the correct, final answer. The rest of this article is about what happens once a label stops being enough on its own.

Chess.com: an accuracy score, labels, and a coach behind the top tier

Chess.com's own support page defines Accuracy plainly: "a measurement of how closely you played to what the computer has determined to be the best possible play against your opponent's specific moves." The score is currently calculated by CAPS2, the successor to the platform's original CAPS algorithm. CAPS2 changed the underlying math so scores read closer to an academic grade, typically landing somewhere in the 50 to 95 range for most players, rather than the wider, harsher spread the older system produced.

On top of the accuracy number, chess.com labels individual moves. Third-party sources commonly cite the scale as Brilliant, Great, Best, Excellent, Good, Inaccuracy, Mistake, Blunder and Missed Win, though that exact ladder is not itself spelled out on chess.com's own accuracy page.

Non-premium accounts are commonly reported, across several third-party sites, to get one free Game Review per day, with paid tiers raising or removing that cap. Chess.com's own pricing page did not render current tier prices when this article was checked, and independent sources disagree with each other by a wide margin for the same tier, so no figure is printed here. Check chess.com's own membership page for the current number rather than trust one quoted secondhand.

Here is the part worth saying plainly rather than skating past: chess.com's top tier, Diamond, already adds full natural-language coach explanations on top of Game Review, per third-party tier comparisons. That is chess.com's own answer to the exact gap this article is about. It is real, and it exists today. It is also reserved for the most expensive tier, it appears to be English only, and chess.com's own forums have active, recent threads from paying users describing the coach commentary as inconsistent, vague, or simply absent. Having built the feature is not the same as having it work reliably for everyone paying for it.

Lichess: free, unlimited, and it has no paid tier to upgrade into

Lichess's analysis board runs server-side Stockfish on every game, free, with no daily cap, and computes an accuracy percentage and average centipawn loss alongside the same inaccuracy, mistake and blunder labels. There is no paid tier at all on lichess, for this or anything else on the site.

Lichess's own accuracy page says it grades moves by the change in your win probability rather than by a fixed centipawn cutoff, and criticises centipawns directly on the grounds that their meaning depends on the position. It does not publish the exact cutoffs on that page. Older discussions online quote flat centipawn bands of roughly 50 to 99 for an inaccuracy, 100 to 299 for a mistake and 300 or more for a blunder. Those appear to describe an earlier version of the algorithm, so treat them as outdated rather than as lichess's current published rule. Lichess also offers what's described online as a "Learn from your mistakes" trainer built from your flagged moves, which drills individual positions rather than naming a pattern across your games.

For what the underlying numbers actually mean, why a centipawn is not a fixed unit, what a search depth figure really guarantees, and how often a player at a given rating would have found the engine's own top move, see our fuller piece on what chess analysis can and cannot tell you.

The free clones built specifically to undercut the cap

A handful of small, independent tools, Chessigma, Chess It Up, Chess Compass, Chessdesk and chessgameanalysis.com among them, exist specifically to offer unlimited free game review with no signup, usually by importing a chess.com or lichess username and running Stockfish directly. Several run the engine client-side as a privacy pitch.

Worth being honest about what this means for the rest of this article: free is not a differentiator anymore. Lichess plus this wave of clones already own that lane, and own it well. None of them add prose beyond the label and the best line, which is the actual gap left standing.

What counts as a Brilliant move, and why players don't fully trust the label

Chess.com's own explanation of its Brilliant (!!) tag sets four conditions: the move has to be a piece sacrifice, it has to be a good sacrifice that does not weaken your position, it has to represent a comparatively narrow path to a winning position rather than padding an already overwhelming lead, and it is not awarded if you were already winning convincingly without it. Chess.com is reportedly more lenient with the sacrifice threshold for lower-rated players than for higher-rated ones.

That is a specific, sensible-sounding rule. In practice, players do not always buy it. Chess.com's own forum has a running, multi-year thread of complaints with titles like "Fake Review Summary Showing Brilliant moves," "The problem with the new 'Brilliant' moves feature..," "Brilliant moves aren't all that brilliant," and "Will chess.com ever fix brilliant moves?" One complaint, paraphrased in search results and not independently confirmed word for word, describes the engine flagging a queen recapture as Brilliant simply because every alternative was a few points worse, not because the move was hard to find. Another describes close to the opposite failure: when every legal move is already winning, an objectively poor move sometimes registers as only a mistake or an inaccuracy, as if the labels go soft once the outcome is already decided.

Label without cause is the complaint that keeps coming back

Search chess.com's own forums for Game Review and the same complaint recurs across years, not from one disgruntled thread: "game review stopped giving explanations," "game review no explanation," "anybody find that the game review makes no sense sometimes," and, bluntly, "Game Reviews = Terrible." One thread asks it directly: "Why does the coach occasionally just...not explain why a move was evaluated the way it was?" A separate thread, "Blunder/Mistake count doesn't make sense?", is about the labels feeling arbitrary rather than the missing prose.

One forum summary reduces the whole complaint to a single line, close to a direct quote though not independently confirmed word for word: the review tells you if you blundered, not why you blundered. Another example cited in the same discussions has the coach's commentary landing on something as thin as "that's an unfortunate error," again paraphrased rather than confirmed verbatim.

A label is not a lesson, even a correct one

Say a review correctly tags a move as a mistake and shows the better line. A 900-rated player reads it, nods, and loses the same way three games later, because nothing connected that one mistake to a pattern. Was it the third time this month a piece got left hanging after a check? The second time a plan got abandoned mid-attack? Naming the recurring cause, not just the one move, is what turns a label into something to actually practice.

Almost nothing in this space does that today. DecodeChess is the closest thing this research turned up: it produces written, explainable analysis of individual positions rather than a bare label, a genuinely different approach from chess.com or lichess. It is still position by position, not built around naming what recurs across a full game, and its pricing page did not resolve reliably when checked, so confirm the current cost directly on decodechess.com. Aimchess takes a different angle again, building personalized training drills from your game history rather than prose explanations of individual mistakes; its own store listings are the more reliable place to check pricing than any third-party summary, including this one.

What you actually get, free versus paid

Where Chessdrive fits, and where it doesn't

Chessdrive, the app we build, runs the same kind of evaluation as the rest of this list, Stockfish through the multistockfish package at MultiPV=3, then adds an LLM coach that writes out why a flagged move was wrong in prose rather than stopping at the label. That overlaps with what chess.com's Diamond tier and DecodeChess are each already trying to do in their own way; this is not the first attempt at an explanation layer.

The part not found anywhere else in this research is the language. Chessdrive's coach answers in your own language across a list of twenty languages, without needing chess.com's top membership tier to get an explanation at all. Standard notation stays in English in every language, since that is how notation actually works, but the sentences around it do not. Games can be played inside the app or imported, and Chessdrive is live on both iOS and Android. To be precise about our own limits, since this page criticises vague free tiers: the free allowance is a single review on the house rather than a daily one, and the second review asks you to subscribe.

How to actually use a review, whichever tool you're on

Whichever tool produced the labels, the useful next step looks the same. Read the flagged moves in order of how much they cost, not in the order they happened. Ask whether the better move was something a player at your level could plausibly have found, or whether it required calculation nobody at your rating does over the board; if the latter, file it as trivia and move on. Look across your last five or ten reviewed games for a repeated shape: the same piece left hanging after a check, the same plan abandoned halfway, the same opening line misplayed into a known trap. Turn whatever repeats into one specific thing to watch for next game, not a list of ten, then check a week later whether it stopped happening.

A correct label was never the hard part. Every tool on this page can produce one. What still separates them is whether anything after the label gives you a reason to play differently next time.

Last verified: 28 August 2026, against the sources cited above. Pricing, free-tier limits and feature gates on third-party products change without notice; confirm current details on each product's own page before relying on them.