How to get better at chess: what 150,000 games show
Most advice on how to get better at chess is a list: learn tactics, study openings, play longer games. The trouble with a list is that it treats an 800 player and an 1800 player the same. They aren't. We counted 150,000 online games, with an engine evaluation of every move, and asked a narrower question: what does the next rating up do differently?
The answer changes as you climb. Below 1200, most of the gap is in the first ten moves. Around 1600 it moves into the middlegame. Past 1600 it's in quiet, level positions where nothing is hanging. Almost none of it, at any rating, comes after move 25.
How we counted
A slip is a move that costs at least 10 points of winning chances, out of 100. The phases are by move number: the opening is moves 1 to 10, the middlegame moves 11 to 25, and late game everything from move 26. We added up the winning chances each group loses per 100 moves, and split the difference between neighboring rating groups by where it happens.
The main numbers are from rapid games (10 to 25 minutes), the time control most online players play, with about 86,000 player-games. Classical and blitz tell the same story and are noted where they differ. Ratings are Lichess ratings, which usually read higher than chess.com's for the same player; more on that at the end.
Where the next rating is hiding
Share of the gap to the next rating group, by phase of the game (rapid):
| Climbing from → to | Opening | Middlegame | Late game |
|---|---|---|---|
| Under 800 → 800s and 1000s | 75% | 33% | −8% |
| 800–1199 → 1200–1599 | 63% | 51% | −14% |
| 1200–1599 → 1600–1999 | 57% | 68% | −24% |
| 1600–1999 → 2000+ | 33% | 63% | 4% |
The shares add up to 100%. A negative share means the stronger group loses more there, mostly because their games last longer.
Two more things hold at every step:
- Nearly all of the gap is with queens on the board. With queens off, players a band apart lose about the same per 100 moves.
- The late game is never where the next rating is. Move for move, players at 900 and players at 2000 slip about as often after move 25 (8 and 8 in 100 in rapid). Strong players do win more won endings, but that isn't what separates one rating group from the next.
So the path is different depending on where you are.
Under 1200: the first ten moves, and loose pieces
In rapid, players under 800 slip on 18 of every 100 opening moves. At 1200–1599 it's 8; over 2000, 2.
Most of it is one thing. Moves that leave a piece loose (a knight, bishop, rook or queen that could be won by simply counting attackers and defenders) account for about 85% of the gap between under-800 and the 800s and 1000s, and two thirds of the gap to 1200. (These situations overlap with the opening, so the shares don't add up across the two.)
Two simple habits in the first moves also fade with rating. In blitz, players under 800 move their queen in the first four moves, other than to capture, in about 20 games out of 100; above 1600, about 4. And after ten moves, players under 800 still have two or more knights and bishops at home, while their opponent has developed, in about 21 games out of 100; at 2000 and up, about 11.
What to work on: before every move, check every piece of yours that's attacked and every one that isn't defended. Develop your knights and bishops before your queen.
In your report: "Loose pieces" counts how often your turns began with a piece left loose, and what it cost, next to players at your rating. "Practice this" on that slide sets up those positions from your own games.
1200 to 1600: the middlegame, and the move after a mistake
From 1200 the opening still matters (57% of the gap) but the middlegame takes over (68%). Loose pieces still account for about half the gap.
Something surprising happens here. Stronger players leave pieces loose almost as often (about one move in eight at 1200–1599, one in ten at 2000+). The difference is that it rarely costs them: at 1200–1599, 19% of those moves are slips; at 2000+, 11%. Strong players leave a piece "loose" when they've seen that taking it doesn't work.
The same holds for free material. Above 1200, players take an undefended piece about three times in four, at every rating. When a strong player passes it up, it's usually right: 13% of those moves are slips at 2000+, against 23% at 1200–1599.
The other big share at this stage is the move after a slip of your own, about 40 to 50% of the gap. At every rating, the move right after a slip is more than twice as likely to be another slip. Part of that is chess (a mistake makes the position sharper), but the move after a mistake is where the next one hides.
What to work on: when a piece is loose, yours or theirs, ask whether taking it actually works before deciding. After you've slipped, take extra time on the very next move.
In your report: the slides on loose pieces and on playing when you're ahead show where your slips cluster. "Practice this" works on the positions you got wrong.
1600 and up: quiet, level positions
Past 1600, loose pieces drop to about a quarter of the gap. The biggest share, about half, comes from moves in level positions, where neither side is ahead. Another third comes from positions where you're already clearly better and need to press.
This is the hardest part of chess to train, because nothing is hanging and no tactic is waiting. It's choosing a plan and not drifting. It's also why converting wins climbs so steeply here: in blitz, the typical player rated 1200–1599 wins 59% of games after being three pawns ahead; at 1600–1999 it's 69%; at 2000 and up, 79%.
What to work on: in your games, find the level positions where your evaluation slid without one obvious mistake, and ask what plan the better move belonged to. When you're winning, keep checking your opponent's best reply.
In your report: "Your openings", the closing score and the guide (ask it about any game) are built for this stage.
What doesn't decide your rating
Some things look the same at every rating, so they're poor places to look for the next rating:
- A long think followed by a quick blunder happens in about 5 blitz games out of 100 in every group.
- How many openings you play. Among 100 players with 200 games each, players with a narrow repertoire and players with a wide one were rated about the same, and did no better than their ratings predicted. Below 800, players came out of their main openings better than their rare ones (12 of 17 players); above 1200, their main openings gave them no edge (23 of 60).
- More time on the clock, below 1200. Players rated 800–1199 slip about as often in classical games as in blitz (13 in 100 moves either way). Over 2000, they slip about 30% less in classical. Extra time helps once you know what to look for. (Different players choose different time controls, so treat this as a pattern, not a law.)
Style isn't rating
A rating says how often you win. It doesn't say how you play. Our reports place every player on two lines, and the 100 players we measured show how independent they are:
- Forcing or quiet: how much more often you check or capture than the position calls for. This has almost nothing to do with rating. There are very forcing and very quiet players under 800, and players over 2000 sit close to the middle, where you play what the position asks for. Neither end is better; the advice differs. A forcing player should check what each capture gives up; a quiet one, what each quiet move allows.
- Loose or safe: how rarely your turns begin with a piece you can't afford to lose. This one is skill, and it climbs closely with rating.
- Fighting back is a third trait that barely follows rating: how often you draw or win a game in which you were clearly worse. Among players at the same rating, those who fight back scored a little above what their rating predicted, and they let fewer won games slip.
So two players rated 1400 can need opposite advice. That's the case for looking at your own games rather than anyone's list.
Mate in one: what the usual numbers hide
You'll read that beginners leave mate in one far more than strong players. That's true but misleading. Most of those mates are allowed in positions that were already lost, and strong players resign those games rather than play on: 59% of losses at 2000+ end in resignation in blitz, against 28% under 800, where 42% end in checkmate.
Counting only positions that weren't already lost, players under 800 leave mate in one in about 7 games out of 100, and players over 2000 in about 1. Still a big difference, and still the cheapest one to fix: before every move, look at every check your opponent could give.
Find your own path
Averages describe groups. Your games have their own order: a 1300 player might lose most in the opening, another in the move after a mistake.
Plysight does this count on your games. Enter your chess.com or Lichess username, or upload a PGN file, and it reads your last 200 games, finds where your slips happen, and puts each pattern next to players at your rating. Then "Practice this" sets up the positions from your own games, with a guide that answers from them. Free, no account.
Common questions
What should I work on to get better at chess? It depends on your rating. Below 1200, the first ten moves and pieces left loose. From 1200 to 1600, the middlegame and the move after a mistake. Past 1600, quiet, level positions and finishing won games.
Does studying endgames help me climb? Not as much as you might expect. Move for move, players at 900 and at 2000 slip about as often after move 25, so that isn't where the next rating is. Strong players do win more won endings.
Does playing longer games help? It depends on your level. Players rated 800 to 1199 slip about as often in classical games as in blitz. Over 2000, they slip about 30% less. Extra time helps once you know what to check.
Is a Lichess rating the same as a chess.com rating? No. Lichess ratings usually read higher than chess.com's for the same player, by an amount that varies. The ratings in this article are Lichess ratings.
About these numbers
- The 150,000 games are Lichess games from August 2026 with Lichess's own engine evaluations, players' names removed. Rapid: 2,742 player-games under 800 up to 30,707 at 1200–1599. Classical has too few games under 800 to count.
- Per-game counts (mate in one, openings, converting wins) come from 19,959 blitz games by 100 public Lichess players, 20 per rating group. Style traits are from the same 100 players.
- Results: players often ask for analysis of games that upset them, so the 150,000 games are used only for how often things happen per move; who won comes from the 100 players' games.
- Ratings are Lichess ratings. The games include no ratings under 400 and few under 600.
- Written and counted by Plysight. Updated twice a year.