Using LLMs (Gemini / ChatGPT) for Chess Analysis & Interactive Coaching: What are your thoughts?

Hi everyone!
What are your thoughts or experiences on using LLMs like ChatGPT or Gemini for chess analysis? If you don't use it, why not, or what do you think speaks against it?

For anyone overwhelmed by traditional engine lines, just feeling to lazy for analysing or being simply lost, this approach really takes you by the hand.

️ Since these models have improved significantly, you don't really see many illegal moves anymore.
I've been experimenting with Gemini (Thinking Mode) lately, and as a beginner/intermediate player, I'm genuinely impressed.
It’s surprisingly good at summarizing complex engine evaluations for any position in simple terms and walking me through my mistakes. (Occasional it misses a detail, but corrects itself instantly if pointed out).

️ ️ My 2 Main Prompts (Feel free to steal these!) for ether general analysis or an interactive lesson:

Prompt 1: Universal Game Review & Quick Summary

You are an experienced chess coach (Grandmaster level) helping me analyze my game.
About me: I play 10-minute Rapid games with a rating of around [yourELO)] Elo.
My color: [Black / White]
Here is the PGN of my game:
[INSERT PGN HERE]
Please cover the following points:
1.Game Summary: Overall dynamic of the game.
2.Key Moments: Identify 2–3 critical mistakes or missed opportunities for both sides.
3.Strategic Mistakes: Flawed plans (e.g., poor piece placement, pawn weaknesses).
4.Tactical Motifs: Missed tactics.
5.3 Concrete Takeaways: Key lessons for my next games.

Prompt 2: Interactive Trainer (Guided & Step-by-Step)

You are an experienced chess coach (Grandmaster level) helping me analyze my game through active learning.
About me: I play 10-minute Rapid games with a rating of around [number] Elo.
My color: [Black / White]
Here is the PGN of my game:
[INSERT PGN HERE]
How to Structure Our Session:
1.Initial Overview: Give a short Game Summary without spoiling mistakes or best moves yet.
2.Interactive Analysis: Walk me through the game. When you reach a critical moment (mistake, missed tactic, strategic error):
-STOP immediately. Do not reveal the evaluation or correct move yet.
-Ask me a targeted question to guide my focus (e.g., 'Look at Black's king safety...').
3.How to React to My Answers:
-If CORRECT: Confirm, explain why in depth, and move to the next key moment.
-If INCORRECT: Point out why it doesn't work, give a stronger hint, and let me try one more time (only reveal the answer if I get it wrong twice).
4.Final Wrap-Up: Summarize strategic mistakes, tactical motifs, and 3 concrete takeaways.

What’s your take? Have you tried interactive prompts like this before?

No. Just no. there are a lot of errors and there are lots of people willing to coach you for a rate that won't make you go broke nowadays.

I think LLMs are only going to get better at what they do - chess included.

Recently, just for fun, I tried using ChatGPT to analyze and evaluate a position. The result was inconsistent and riddled with errors: incorrect variations, impossible piece captures, and so on.

Asking ChatGPT to act like an International Grandmaster won't turn it into one. It will simply adopt the language of a Grandmaster (assuming it has "read" enough books by Grandmasters to mimic their style, like a parrot). Chatbots possess neither deductive reasoning nor analytical ability; they merely produce statements that imitate the form of reasoning.

Have been wibe coding some local experiment for myself, when I equipped the LLM (gemini-3-flash-preview here in the example, there are better models of course) with stockfish as a tool it became more consistent, but not perfect.