How to analyze your own chess games in 20 minutes (without letting the engine do the thinking)

Analyze a chess game in three passes. First, annotate it yourself with no engine running and mark every move you were unsure about. Then switch the engine on and look only at the moves where your note and the evaluation disagree. Finally, write one sentence per mistake saying what you would do differently. Twenty minutes, one game, every day.

Why clicking "Game Review" isn't analysis

Every platform now has a button that reviews your game for you. A few seconds later you have an accuracy percentage, some coloured labels and a list of moves you should have played instead. It feels like analysis. It is closer to reading the answer key before sitting the exam.

The problem is not that the engine is wrong. The problem is what you are doing while it runs: nothing. You scroll to the red moves, nod at the better line, think "I would never have found that," and close the tab. Nothing was retrieved from your own head, so nothing in it was corrected. A week later you play the same move, because the habit behind it was never touched.

I plateaued around 1600 on Lichess for a long time while playing a lot. I could read the evaluation number; I could not read the lines, and I could not say why the number moved. The review told me what the engine thought. It never asked what I had thought, which turned out to be the only thing worth examining.

How do I analyze my chess games to find my mistakes?

Commit to your own explanation before you let the engine grade it. Everything below is a way of forcing that order, because the natural order is the other way round: the engine is one click away and your own opinion is uncomfortable to write down.

The routine has three passes on a fixed clock. Eight minutes without the engine, eight with it, four writing. It is built for a player with twenty minutes and one game, not a weekend with a database.

Pick a game you lost, or a win that felt like it was slipping away. Rapid games (15+10 or longer) work best: a rapid game has a few decisions you remember making, a blitz game is a slide of half-decisions with nothing to annotate.

Pass 1 (8 minutes): your notes first

Open the game with the engine switched off; Lichess, Chess.com and LearnChess all have an engine toggle on the analysis board. Play through the moves at a steady pace and put one of three marks on any move that earns one:

• ? for a move you were not sure about at the time. Not a move you now suspect was bad: a move where, during the game, you hesitated or played something and hoped.
• ! for a move you were proud of. The idea you saw, the sacrifice you calculated, the quiet move. These matter as much as the mistakes, because the engine may say the proud move was wrong, and that is the most instructive disagreement you will get.
• ⌛ (the hourglass) for a move played with under a minute on the clock, or in any rush. The fix for a time-pressure mistake is not chess knowledge, so these get a different kind of attention later.

Next to every "?", write one line: what you were weighing and why you chose what you chose. "Took with the queen to keep the rook on the file" is enough. If you cannot remember why you played a move, write "no idea," which is itself a useful note.

Here is a position that earns a "?" in most games between 1200 and 1800. White to move. Before you scroll on, decide what you would play and write down your reason.

87654321

abcdefgh

The position comes from the LearnChess lesson on in-between moves, where almost everyone plays the same thing. The queen is attacked, the rook is free, so take it: 1.♕︎xd4. The note would say "the rook was free." That is a complete, honest, wrong explanation, and it is exactly what pass 1 is for. You cannot grade an explanation you never wrote down.

Three to six marks per game is typical. No marks at all means you played on autopilot, which is its own diagnosis.

Pass 2 (8 minutes): the engine as a grader

Switch the engine on. Do not read the whole game. Go straight to your marks, and only your marks. At each one, compare your note with the evaluation before and after the move. There are three outcomes:

1. You agree. You marked it "?" and the engine says it was fine, or you marked it "!" and the engine agrees. Spend no more time here.
2. You were unsure and the engine confirms it was a mistake. Look at the line it prefers for one question only: what did it see that your note did not mention? Above, the note said "the rook was free." The engine says 1.♕︎xd4 loses to 1...♕︎f5+, a check that also hits the rook on c2. The missing ingredient was not calculation. It was the question "what does Black do with the move after I take?"
3. You were proud and the engine says you were wrong. Stay here longest. A "!" the engine refutes is a belief you hold with confidence and will act on again next week. Find the exact move where your idea breaks, and name it.

Go through the hourglass moves with a different question: not "was it good" but "how bad was it." If your rushed moves were all fine, time management is not your problem. If two of them dropped a piece, the problem is the twenty minutes you spent earlier getting to that clock.

A word on the numbers. The evaluation is in pawns. A swing of a third of a pawn, say +0.5 to +0.2, is noise at our level, and chasing it is how pass 2 takes an hour. A swing of about 1.5 pawns or more is the one to care about. It usually means something concrete happened: a pawn was lost, a piece landed on a bad square for good, or a tactic appeared. Those are swings you can name in a sentence. Above, Stockfish at depth 22 rates 1.♖︎c8+ at about +1.8 for White and the position after 1.♕︎xd4 ♕︎f5+ at worse than -11. That gap is the whole game, and a note that said "the rook was free" would never have found it.

Set a timer. Eight minutes is enough for five marks. When it goes, stop, even mid-line. The clock is what brings you back tomorrow.

Pass 3 (4 minutes): one sentence per mistake

For each mark where the engine disagreed with you, write one sentence in a fixed shape:

I played X because [your note]; it loses to Y; next time I check Z.

The shape does the work. The first clause comes from pass 1, so it is what you actually thought. The second comes from pass 2 and is one concrete move, not "the engine didn't like it." The third is the only part you invent, and it has to be a check you can run at the board in ten seconds, not a resolution to "calculate more."

The position above gives this sentence: I played ♕︎xd4 because the rook was free; it loses to ♕︎f5+, which also hits my rook on c2; next time, before I take a free piece, I look for my opponent's checks. Here is the correct move, so the sentence has a picture.

87654321

abcdefgh

Four minutes is enough for three or four sentences; write up the biggest swings and let the rest go. One well-understood mistake teaches more than eight half-read lines.

What to keep

Keep the sentences, nothing else. Not the games, not the engine lines, not the accuracy score. A plain text file is fine, as long as it is one place and you add to it after every game.

After ten games the list starts to repeat itself, and that repetition is the real output of the routine. Three of ten sentences end with "before I take a free piece, I look for checks." Four say "I had under a minute." You have just found your plateau type, the diagnosis the plateau post runs from your last ten losses, and you got there with the reason attached.

The repeats also tell you what to study. If several sentences name a piece that ran out of squares, the trapped pieces lesson is fifteen minutes well spent. If most are one-move oversights under a running clock, the fix is a check short enough to run on every move. Your games chose the lesson, not a list.

Reread the file before your next game, so that the "next time I check Z" clauses are the last chess thought in your head before move one.

Where should I analyze my chess games?

Any tool with an engine you can switch off will do. The facts below were checked on 4 November 2026.

Lichess. Free, no paid tier. The analysis board runs Stockfish in your browser without limit, which is all pass 2 needs. The deeper server-side "request a computer analysis", with the accuracy report, is listed on Lichess's own features page at 40 games a day per account, far more than this routine uses. Its "learn from your mistakes" mode is a decent pass 2 on its own: it makes you find the better move before showing it.

Chess.com. The free (Basic) plan includes one full Game Review per day, according to Chess.com's membership page on the check date; Platinum and Diamond lift that to unlimited. One review a day is exactly this routine's budget, so the limit is not the problem. Reading the review before thinking is.

LearnChess. Disclosure: I build it. The analysis board runs real Stockfish free on every plan, with a toggle to keep it off during pass 1. Any move can branch into its own line in a tab, so you can test your pass 1 idea next to the engine's. The one-click game review, which grades every move and marks the turning point, is counted on the free plan (one a day, plus your latest 25 games at the first import) and unlimited on the paid tiers.

Pick the one where your games already are. The routine matters far more than the tool, and the best tool is the one you will open tomorrow.

If you want the engine's verdict explained in words

The gap in every tool above is pass 2's hardest moment: the engine prefers a move and you cannot see why. I wanted an easy way to decipher engine lines and to ask questions about the board in plain language, and that is what the LearnChess coach on the analysis board is for. Click any move and ask why. The coach reads the exact position and the live engine evaluation and answers in a sentence: the check comes first because the king lands on the queen's diagonal, so the capture arrives with tempo. That is the sentence pass 3 needs, when you cannot write it yourself. Free accounts get ten coach messages in total; the paid tiers get more. Spend them on the "!" moves the engine refuted, where a sentence changes what you believe.