How to Analyze Chess Games Like a GM (Without Being One)

Watch a grandmaster like Daniel Naroditsky stream a chess.com speed-run and you’ll notice something: he barely uses the engine. He talks through positions in plain English — “Black wants to play c5 here, but the d-file gets ugly,” — and only checks the engine to confirm. That’s because GMs analyze games using the candidate moves method, not engine evaluations. Here’s the 5-step framework borrowed from Alexander Kotov’s Think Like a Grandmaster (1971), updated for adult improvers in 2026.

What GMs do that amateurs don’t (the candidate-moves habit)

The amateur analysis loop: open Game Review, read the engine’s recommended move, nod, click next. Total reps spent thinking: zero. The GM loop: in every critical position, list 2-3 candidate moves, evaluate each in plain English, pick the one with the soundest justification, and only then check what the engine prefers. Total reps spent thinking: high.

Kotov’s idea, written 50+ years ago, is that you don’t get better at chess by being told the right move. You get better by being forced to choose between three plausible moves and then being right or wrong. The engine is a verification tool. It is not a teaching tool.

Step 1: Identify the critical moments before opening the engine

Most games have 3-5 critical moments — positions where the game tilts decisively. The other 35 moves are either book, automatic, or recovery. Your job in step 1 is to find the 3-5 moments without engine help.

Replay your game at 5 seconds per move. Pause when:

• Material balance changes (a piece is taken, a sacrifice happens).
• The pawn structure changes irreversibly (a pawn breaks, files open).
• One side castles or chooses not to castle.
• A long think happened — your clock ran 2+ minutes on a single move.
• An exchange offer was made or declined.

Mark these positions. They’re your critical moments. Now you analyze.

Step 2: Assess the position in plain English (no engine evals)

Before considering moves, describe the position. Use Silman’s framework from How to Reassess Your Chess — the 5 imbalances:

1. Pawn structure — who has the better long-term pawns?
2. Minor pieces — who has the better bishop/knight?
3. King safety — whose king is more exposed?
4. Space — who has more squares to maneuver?
5. Initiative — who is making threats?

Write 2-3 sentences in your own words. “White has a slightly better pawn structure but Black’s bishop pair is dangerous; my king is exposed because I haven’t castled yet.” That sentence is the foundation; every candidate move is judged against whether it improves or worsens the imbalance you care about.

Step 3: Compare candidate moves systematically

At a critical moment, list 3 moves you’d consider. Not 1, not 5 — exactly 3. Kotov was specific about this number because more than 3 candidates fragments your calculation, and fewer than 3 means you didn’t really consider alternatives.

For each candidate, write what it does in plain English. Example:

• Move A: Bg5 — pins the knight, but the bishop can be chased by h6.
• Move B: e5 — gains space, locks in my bishop on c1.
• Move C: Nbd2 — solid development, slow.

Now pick. Once you’ve picked, then open the engine. The engine might prefer C even though A felt sharpest. The lesson is in the gap: why did you favor the sharp move? What about your evaluation of “the bishop can be chased” was actually fine?

Step 4: Find the “lesson move” not the “best move”

This is the most underused step. Engines tell you the best move. The best move is often a 5-move tactical sequence you’d never spot in a real game. The lesson move is the move that, if you’d known the principle, you would have found.

Example: engine says +1.4 with Bxh7+. The whole line goes Kxh7 Ng5+ Kg8 Qh5 Rfe8 Qxf7+ Kh8 Ne6. You’re never going to calculate that in a real game at 1400 rapid. The lesson move is: “open the h-file when you have queen and knight near it.” That’s drillable. The exact sequence isn’t.

Translate every “best move” into a “lesson move.” That’s where actual learning happens.

Step 5: Convert insight into one drill

After analyzing 1 game with steps 1-4, you should have one drill. Not five drills, not “I need to study tactics.” One concrete, time-boxed activity for the next 5 days. “Drill 25 puzzles tagged kingside-sacrifice.” Or “play 3 rapid games where I write down candidates before move 15.” Or “watch 1 Naroditsky game on the Caro-Kann.”

Without the drill step, analysis stays intellectual. The drill is what writes the new pattern into your decision making.

Automating GM-level review with AI (the 60-second alternative)

The 5-step framework is the gold standard. It’s also slow — 30-60 minutes per game done well. Most adult improvers can do this for 1-2 games per week, max.

The shortcut for the other 98 games: feed them to a tool that does steps 1-3 across the archive automatically. MyChessPlan flags critical moments, identifies imbalance patterns across 100 games, and reports the recurring lesson moves — the same outputs as the manual GM workflow, just aggregated and automated. Sample report here.

Use both. Manual GM-level review on 2 games a week for the deep learning. Automated 100-game view monthly for the pattern-level diagnosis. Our coach-style guide covers the manual workflow in detail; this post covered the GM-specific candidate-moves layer on top.

Three GM analysis habits worth stealing immediately

Beyond Kotov’s candidate-moves method, modern GMs share three small habits that compound into big improvements when amateurs adopt them:

1. Annotate one move per game in plain English. Pick the most interesting move you played and write 2 sentences about why. Not the engine’s move — yours. After 30 games, you have a notebook of your own thinking patterns. Naroditsky does versions of this on his streams; Vladimir Kramnik built his entire repertoire from longhand annotations as a teenager.
2. Compare your move to a database move. Use the chess.com Opening Explorer or Lichess masters database. After your game, see what 2400+ players played in the same opening position. The gap between their move and yours is a 30-second free lesson. Aronian and Nakamura have both said in interviews this was their core teenage study habit.
3. Replay a single GM game weekly. Same GM, same opening, weekly. Pattern transfer happens through volume in narrow areas. Watching one Caruana Petroff game won’t help. Watching 12 of them across 12 weeks will.

What separates 2000 from 2200 (and why it matters even at 1500)

The IM-to-GM gap isn’t that GMs see more — it’s that they evaluate faster and they trust their evaluations enough to act on them. At 1500, the equivalent shift is from “I don’t know what to play” to “I’m going to play this and find out.” Decisiveness, not ignorance — making the best move you can identify in 30 seconds beats agonizing for 4 minutes and playing the same move with 90 seconds less on the clock.

The candidate-moves method enforces this. Three candidates, brief evaluation of each, pick one, move on. Repeat 30 times in a game and your time management resolves itself. Our rating-band plateau guide covers why decisiveness training matters most at the 1500-2000 band specifically.