How to Analyse Your Own Chess Games (and Only Then the Engine)

Of everything you can do to improve at chess, this is the highest-yielding and the one fewest people do properly. Analysing is not running the engine over the game. It is reconstructing your own thinking and comparing it with what the position was asking for.

The difference between the two is the difference between knowing you were wrong and knowing why.

Why the engine goes last

The engine is an extraordinary tool with one flaw: it answers a question that is not yours.

The engine answers "what was the best move". You need to answer "why did I choose the one I chose". And only you can answer that second question, in that moment, before the engine's answer contaminates your memory.

It is a well-known phenomenon: the instant you see the right move, your brain rewrites history. "Ah, of course, I nearly played that." No, you did not nearly play it. That is why the order matters.

The method, in five steps

Step 1: walk through the game with nothing

Open the game, no engine and no database. Walk it from move one and stop at every moment you remember hesitating. At each one, write down three things:

• Which move you played.
• Which other move you were considering.
• Why you chose the one you chose.

If you cannot remember, write that down too: "I do not remember thinking anything here" is extremely valuable data, because it usually marks the point where you stopped playing and started making moves.

Step 2: mark the critical moments

A critical moment is a move where the game could go two different ways. There are not many: a typical game has between three and six.

They are usually: the last move of the opening before you need a plan, an important exchange, the decision to attack or consolidate, and the entry into the endgame.

That is where the learning is. The rest of the game is execution.

Step 3: evaluate the critical positions yourself

At each critical moment, write an evaluation in words before looking at any number:

"I think I am slightly better here: I have the bishop pair and his d5 pawn is weak. My problem is that his knight is heading for c4."

This is uncomfortable at first, and it is exactly the skill that separates a 1400 from an 1800. A number from the engine does not teach it to you; writing your own and comparing does.

Step 4: now, the engine

Only now. And with two rules:

Do not scroll the evaluation move by move. Go straight to the critical moments you marked and compare there.

When the engine disagrees, look for the reason, not the line. "The engine prefers Rc1" is useless. "The engine prefers Rc1 because it prepares the c4 break, and my move prevented it" is not.

If you cannot find the reason in two minutes, flag it and move on. A reason you do not understand is not learned.

Step 5: write one line

At the end of the analysis, a single sentence: what you learned from this game.

Real examples of the kind of sentence that helps:

• "I traded my good bishop for his bad knight for no reason."
• "I attacked with two pieces while three were still undeveloped."
• "In the endgame I needed to activate my king and I sat there defending."

That sentence is the only thing you will ever reread. And when you have twenty of them, you will see that three or four repeat. That is your study programme, taken from your own games rather than from general theory.

What to do with games you won

Analyse them too, for a reason that surprises people: most wins contain two or three moments where the position was worse than it looked and the opponent did not use it.

Those are mistakes that cost nothing this time. Next time they will. And they are the hardest to spot alone, because the result hides them.

The most common mistake

Skimming every game instead of working one properly. Twenty games run through the engine in twenty minutes leave nothing behind; one game worked on for forty minutes leaves a sentence that changes your month.

If you play five long games a week, analyse two. That beats skimming five.

How to read what the engine says

The engine comes last, and when it comes you have to know how to read it, because misreading it is the commonest way to waste time with an excellent tool.

The numerical evaluation is an opinion, not a verdict. A +0.7 does not mean "winning": it means that with perfect play from both sides there is a small advantage. At club level, anything between −1 and +1 is a position where the result is decided by whoever errs first.

What matters is not which move was better, but how much yours cost. A move that drops the evaluation from +0.4 to +0.1 is not a mistake: it is noise. One that drops it from +1.5 to −2 is. Look for the big jumps and ignore the rest; there is a strong temptation to "correct" ten mediocre moves and learn nothing.

The engine does not explain. It gives you the move and not the reason, and the reason is the only thing that transfers to the next game. The way to extract it is to let it run on from its own move for three or four more and look at the position it produces: the idea is usually visible there.

And the engine calculates lines you will never play. A seven-move forced sequence ending in a won endgame is not a useful lesson if you do not know how to play that endgame. What you should take from a game is two or three applicable things, not the objective truth.

What to do with what you found

Analysing without recording is analysing half. What turns twenty analyses into improvement is what happens afterwards:

A document with your errors, grouped by type. Not a list of games: a list of errors. "Traded queens while I had an attack", "missed the in-between check", "pushed pawns in front of my king". After twenty games, that document shows you your three recurring errors, and those three are your study plan.

One saved position per error. The diagram of the exact moment you went wrong. Reviewing ten of those positions once a month is worth more than a hundred new puzzles, because they are yours and because the error is still alive.

One sentence per game. At the end of the analysis, write in one line what you learned. If you cannot write it, the analysis is not finished.

And the monthly check: read the whole document. You will see that three errors explain half your losses, and that two of them no longer appear. The latter is the hardest thing to perceive any other way, and it is what sustains consistency.

How long, and which games

To be sustainable it has to be short:

Twenty minutes per game. Ten without the engine, five with it, five writing. More than that does not pay proportionally and makes you stop doing it by week three.

Two games a week. Not all of them. Choose the ones that teach: a loss in a position you understood, and a hard-won win. Losses to a one-move oversight need no analysis — you already know what happened — and easy wins do not either.

And once in a while, one whole and slow. An hour on a single game of yours, without an engine, writing what you were thinking at each key moment. It is the exercise closest to having a coach that you can do alone, and one a month is enough.

What to avoid is the all-or-nothing pattern: analysing six games on a Sunday, burning out, and analysing nothing for two months. Twenty minutes twice a week beats that by a very long way.

A minimal format that works

You do not need special software. One text file per game with this is enough:

`` Date, opponent, result, time control Opening: what I played and how far I felt comfortable Critical moment 1: move, what I thought, what the engine says, why Critical moment 2: ... What I learned: (one sentence) ``

Ten of these and you have a better diagnosis than any general advice on the internet, because it is about your games.

Related questions

Do I have to analyse every game? No, and wanting to is the commonest reason people stop analysing. Two a week, chosen — a loss that hurt and a hard-won win — return more than attempting all of them and quitting within a month.

Can I analyse without an engine? Yes, and for the first months it is even better. What the engine adds is precision; what it removes, if used first, is the exercise of searching on your own, which is where almost all the learning lives.

How long per game? Twenty minutes. Ten without the engine, five with it, five writing the conclusion. Two-hour analyses sound serious and almost never survive three weeks.

What do I do with the games I won easily? Nothing. They teach almost nothing and take the time of the ones that do. The exception is a win where at some point you stood worse: that one is worth as much as a loss.