When the Best Move Is Not Your Move
The engine assumes your opponent finds every reply. Below master level that assumption is wrong often enough to change which move actually scores best.
· 3 min read
An engine's evaluation answers a specific question: what is this position worth if both sides play the best moves I can find? Every recommendation it makes inherits that assumption.
Against another engine, the assumption holds. Against a human rated 1400, it does not, and the gap between those two worlds is where a lot of bad advice comes from.
Two ways a move can be good
An objectively best move maximises the value of the position against perfect defence. A practical move maximises your expected result against the person actually sitting opposite you. These usually coincide. When they do not, the difference matters.
Consider a position where the engine offers +1.4 after a piece sacrifice requiring six exact moves to justify, and +0.9 after a quiet developing move. The engine ranks the sacrifice higher, and it is right about the position. But if you find five of the six moves, the sacrifice is losing, while the quiet move leaves you comfortably better with nothing to calculate.
Half a pawn of theoretical value against a large chance of collapse is not a good trade. Strong human players make this calculation constantly, which is why their choices sometimes differ from the engine's without being errors.
The signature of a move you should not play
You do not need to guess. Engine output tells you when its recommendation is demanding, if you read the line rather than the number.
The continuation is long and forcing. If the justification runs eight plies of checks and captures with no branch point where a natural move also works, you are being asked to reproduce a calculation.
The evaluation collapses on a near-miss. Play the second-best move at step three of the line and re-analyse. If +1.4 becomes −2.0, the margin for error is nil.
Second choice is close. When the top two moves are within a couple of tenths of a pawn, they are equal for practical purposes and you should take the one you understand.
Applied to the sacrifice above, all three fire at once, which is about as clear a signal as this gets.
| What you see | What it tells you |
|---|---|
| Top two moves within 0.2 | Equal in practice — take the one you understand |
| Continuation is 8+ forcing plies | You are being asked to reproduce a calculation |
| Second-best at any step loses a pawn or more | No margin for error |
| Continuation is quiet and short | Safe to play on general grounds |
| Evaluation holds after a plausible human reply | It survives imperfect defence |
What the engine is genuinely good for
None of this argues against using engines. It argues against one specific use: treating the top move as an instruction.
The productive use is diagnosis. Play through your own game, find the moves where the evaluation dropped sharply, and study the refutation. That line shows you the resource you did not see, and the pattern behind it is what transfers to your next game. A tactic you missed is worth more to you than a move you memorised.
The other use is calibration. If you consistently rate positions as better for you than the engine does, you are overvaluing something specific, often your own attacking chances or a space advantage that does not convert. Finding out which is a genuine improvement in judgement, and it does not require agreeing with the engine about any particular move.
A note on opening preparation
The same caution applies twice over in openings, where it is easy to memorise a line the engine likes and reach a position you have no idea how to play. An opening that produces structures you understand is worth more than one rated a tenth of a pawn higher whose middlegames are foreign to you.
The engine has no opinion about which positions you play well. That part is yours.
A reasonable working rule, then, for deciding whether to trust a recommendation over the board:
- If the gap to your own choice is under half a pawn, play yours.
- If the engine's line is quiet, play the engine's.
- If the engine's line is forcing and long, ask whether you could reproduce it under a clock. If not, take the second choice and accept the tenth of a pawn it costs.
Common questions
- Should I always play the engine's top move?
- In analysis, study it. In a game, not necessarily. The engine's choice sometimes requires a long sequence of accurate follow-ups to be worth anything, and if you cannot find those moves over the board you will get a worse result than from a simpler move it rates slightly lower.
- What is a practical move in chess?
- A move chosen because it is hard for a human opponent to meet, rather than because it is objectively strongest. It sets a problem, keeps pieces on for a player who is better at complications, or steers into a structure you know well.
- Does this mean engine analysis is not useful for improvement?
- It is extremely useful, but for finding what you missed rather than for supplying moves to memorise. The value is in the refutation lines after your mistakes, which show you the pattern you did not see.
Try it on your own position
The analysis board runs the same engine in your browser, free. The app adds what a web page cannot: reading a position straight from a photograph, saved boards, and analysis with no connection.
