You finish a game, run the review, and it says 98% accuracy. Magnus Carlsen averages somewhere in the low-to-mid 90s in classical chess. So... are you better than Magnus? You already know the answer, but it's worth understanding exactly why the number works this way, because once you do, you'll stop chasing it.
What accuracy actually measures
Accuracy is not a percentage of "correct" moves. An engine evaluates the position before and after each of your moves, converts those evaluations into a winning probability, and scores each move by how much winning probability it threw away. Lose nothing, score near 100. Lose a lot, score near zero. The game number blends all your per-move scores together.
The key detail hides in that definition: the score is about what you LOST, not about how good the move was. And how much a move can lose depends entirely on the position it was played in.
Why lopsided games produce absurd accuracy
Suppose your opponent hangs their queen on move 6 and you take it. From that moment you are winning around 99% of the time no matter what you play. Almost any legal move keeps the win probability near 99%, which means almost any legal move loses nothing, which means almost every move you make for the rest of the game scores a near-perfect accuracy automatically. Shuffle your king back and forth in a completely won position and the formula shrugs and hands you 100 after 100.
That's the whole trick behind beginner 98% games: the number isn't lying, it's just answering a question nobody asked. It measures how little you spilled from the position you had, and a completely winning position is nearly impossible to spill. Meanwhile Magnus plays razor-sharp equal positions against the strongest players alive, where every single move can lose something, and "only" scores 93.
So what's a good accuracy?
Against a well-matched opponent in a fighting game, roughly: below 70 means real blunders got through, 70 to 85 is a normal club-level game, 85 to 92 is a genuinely clean game, and above 95 usually means the game was one-sided rather than that anyone played like a machine. Comparing accuracy across different games is close to meaningless. Comparing it against the same opponent strength over many games is where it starts to say something.
This is also why Rookie's "rough rating" estimate doesn't trust accuracy alone anymore: it also looks at your blunder count and, crucially, who you actually beat. A spotless 98% against a 350-rated opponent who hung a knight on move 5 is evidence you can convert a free piece, not evidence you play like a 2600.
The habit worth building instead: ignore the headline number, open the move list, and look at the two or three moments where the win probability actually swung. Those moments are the game. Everything else is noise wearing a percentage sign.