How to tell if your opponent is cheating
You are almost certainly looking at the wrong number. Here is the one that means something.
You lost a game that felt wrong. Somewhere around move twenty the position stopped making sense, your opponent found something you did not think their rating could find, and the rest was technique. So you opened the game review, saw a number in the nineties, and here you are.
That number is the weakest piece of evidence in the entire game.
Accuracy measures move quality, not cheating
An accuracy score does describe your opponent: it is a measure of how close the moves they played were to the ones an engine would have chosen. What it does not do — what it cannot do — is tell you how hard those choices were to make, and that is the part your suspicion depends on.
A game with an early queen trade, a symmetrical pawn structure and nine recaptures hands both players a high accuracy, because most of the moves in it had one sensible answer and both sides found it. A sharp, unbalanced game where every move is a choice between four plausible plans produces a lower number from two far stronger players.
So a 94% is a real statement about the moves, and a poor one about the player. Forcing positions inflate it, quiet accurate play inflates it, and neither is evidence of anything on its own.
The same trap catches the estimated rating some game reviews print next to a performance. It is a rough read on one game’s move quality and was never built as a cheat detector. Treat it as evidence and you will end up suspecting half the people you play.
The gap is the part worth looking at
Start with the kind of position most of a chess game is made of.
Searching to depth 28, Stockfish puts h3 first at +0.4, then d4 and a4 level at +0.2, then Bc2 at +0.1. Four completely different ideas — a waiting move, the main central push, a queenside break and a bishop retreat — inside a third of a pawn of each other.
The two in the middle are worth a second. They are a single centipawn apart, and the engine swaps their order from one run to the next depending on which line it happens to resolve first. If a machine searching ten moves deep cannot separate two entirely different plans, no conclusion about the human who picked one of them is available either.
Play the engine’s top choice here and you have proved nothing. Play its fourth and you have not made a mistake. Positions like this one are the bulk of every game, and no move played in them is evidence of anything, by anybody.
Now the opposite kind of position.
Kg7 holds — the king walks diagonally, closing on the black pawn and the queening square at the same time, and the game is a draw. The other three legal moves lose by force. There is no second-best here, no partial credit, no reasonable alternative that costs a tenth of a pawn.
That is a gap. And a gap is the only structure in chess where finding the move actually tells you something, because not finding it is what happens by default.
This one is a composed study rather than a game, which is precisely why it is useful — it is the cleanest example that exists. In real games the gaps are smaller and much rarer, but the question you are asking is the same one:
not whether your opponent played good moves, but whether they played the only move, in positions where the second choice was far worse, more than once.
Once is a good day; everybody has them. Several times in one game, in positions that took you two minutes each to even understand, is the sort of pattern worth passing to the people who can check it properly. It is not a conclusion you can reach from one game, and this is not a method for identifying a cheat — it is a way of deciding whether a report is worth filing. If you want to review a game honestly, that is the review: not the whole game, just the two or three moments where the position had one answer, and what the alternatives were worth. Our guide on reading the eval bar covers how to read that difference.
Timing is a signal, not something you can conclude from
Humans spend time unevenly and cannot help it. A recapture is instant. A position where three plans compete costs a minute. Stretched over forty moves, that unevenness is a signature, and it is a hard one to fake deliberately.
Somebody reading moves off a second screen has a different signature. Every move costs roughly the same, because every move costs the same work: read the position across, wait, read the answer back. The profile flattens out. Obvious recaptures start taking eight seconds. Genuinely hard positions stop taking thirty.
Move times are on the game page, and they describe something accuracy does not. But be honest with yourself about the caveats: bullet flattens everyone, premoves flatten everyone, and plenty of people simply play in a steady rhythm. Timing patterns are one of the things detection systems weigh, alongside a great deal you cannot see. In your hands, on one game, it is a reason to report and nothing more.
What the site sees, and you do not
Chess.com describes its fair play system as weighing over a hundred gameplay factors, combining automated statistical analysis with human review. The published shape of it is that the models compare the strength the moves themselves imply against what the account has ever shown itself capable of, and that move-matching against engines is only one input among many.
What the factors are, how they are weighted and where the thresholds sit is not published, and deliberately so — a detection system that explains itself in detail is a detection system telling people how to evade it. Anyone describing the internals precisely, this page included, is guessing. What is on the record is the scale: over 130,000 accounts closed in March 2026 alone, with the site reporting that roughly 85% of closures happen without a human reviewing the case.
The asymmetry is the point. You are looking at one game. They are looking at every game an account has ever played, measured against a model of what that account should be capable of. Your single game is the very weakest version of evidence they already hold in bulk — which is the real reason reporting works and arguing does not.
What to do, in order
- Report the account. Chess.com has a report button on the profile and the game; Lichess has the flag. A report is the one action that puts the account in front of people who can actually examine it — though a report on its own does not close anything. What closes an account is what the review finds.
- Do not accuse in chat. Public accusations are against Chess.com’s Community Policy and can get your own account actioned. A message in the game chat is not a report and reaches nobody who can act on it.
- Wait. If the account is closed for a fair play violation, Chess.com adjusts your rating automatically, with the correction showing up after your next game.
- Then review the game anyway. Whatever the account turns out to be, the two or three critical positions are still worth understanding, and they are the only part of the game that was ever going to teach you anything.
If you are the one being accused
This happens constantly, and almost always after a game somebody played well. Being accused in chat is not being investigated: a message calling you a cheat reaches nobody who can act on it and has no effect on your account.
A formal report is a different thing, and worth being accurate about. It can put an account in front of the fair play team, so it is not nothing. What it cannot do is decide anything: the outcome rests on what the analysis of your games shows, not on how many people reported you or how strongly they felt about it. If your games are your own, a report is an inconvenience that goes nowhere.
The line itself is worth stating plainly, because people genuinely get it confused. Analysing your own finished games with an engine is not cheating — it is what engines are for and how every strong player has studied for thirty years. Taking outside help during a live rated game against another person is, and no amount of “only in one game” survives the statistics.
Where that leaves this app in particular is set out on our fair play page.
Most of the time, you were not cheated
The uncomfortable part: the overwhelming majority of games that feel wrong were lost to an opponent having a good day, a plan you did not see coming, or twenty minutes of your own tilt after one bad move. The suspicion is usually the mind doing what it does after a loss.
But it is a useful instinct anyway, as long as you point it at the right thing. Go and look at the two positions where the game turned. Find out what the alternatives were worth. Either you learn something about how the position worked, or you find a gap so wide that reporting is obvious. Both outcomes are better than the accuracy score.
Questions people ask
Does a high accuracy score mean my opponent cheated?
No. Accuracy measures how close the moves played were to an engine’s choices, which is a real statement about the moves and a poor one about the player. Forcing, simplified games produce high numbers for both sides, and sharp ones produce lower numbers from far stronger players. On its own it is not evidence of anything.
How many people actually cheat in online chess?
Nobody knows precisely, and the honest answer is a range. A 2025 survey of 1,924 German club players, using the randomized response technique to get truthful answers to a sensitive question, estimated that about 6% had cheated online at least once in the previous twelve months — that is club players in one country reporting any occurrence, not 6% of chess players cheating regularly. For scale on enforcement, Chess.com closes accounts on the order of 100,000 a month.
Will I get my rating back if my opponent is banned?
Yes, automatically. If an account you lost to is closed for a fair play violation, Chess.com adjusts your rating and the correction shows up after your next game. In serious cases it will go back as far as your last 100 games or three months; results older than that are treated as statistically insignificant.
Can I get banned for using a chess engine or an analysis app?
For analysing your own finished games, no — that is what engines are for, and it is how every strong player studies. For taking help during a live rated game against another person on Chess.com or Lichess, it can close your account. That is the case fair play systems are built to catch.
Should I tell my opponent I think they are cheating?
No. Public accusations break Chess.com’s Community Policy and can get your own account actioned, and a message in game chat reaches nobody who can act on it. Use the report button instead: that is what puts an account in front of the fair play team, even though the report alone does not decide the outcome.