How to Use ChatGPT for Poker Coaching

If you've ever tried pasting a hand history into ChatGPT, you've probably noticed something frustrating: the advice is vague. "Consider your position," "balance your range," "think about pot odds." It reads like a beginner article, not a coaching session.

The problem isn't ChatGPT. It's your data.

A standard hand history from PokerStars, ACR, or any real-money site gives you about 8 fields per action: player name, action type, amount, cards, and position. That's it. There's no context for why an opponent made their move, what emotional state they were in, how long they took to decide, or what kind of player they are. Without that context, any AI — ChatGPT, Claude, Gemini — is essentially guessing.

What Good Poker Data Looks Like

Compare a standard hand history action to one with full behavioral metadata:

Standard export:

Seat 5: PlayerA raises $8.50

Rich behavioral export:

{
  "player": "Nadine Hewitt",
  "actionType": "raise",
  "amount": 8.50,
  "decisionTimeMs": 2847,
  "timingTag": "tank",
  "strengthBucket": "top_pair_weak",
  "decisionReason": "barrel_continuation",
  "fearDiscount": 0.72,
  "callerResistanceProfile": "station_caller",
  "emotion": {
    "state": "tilted",
    "intensity": 0.65
  }
}

With the first format, ChatGPT can tell you basic pot odds math. With the second, it can tell you that Nadine is continuation-betting with a weak hand while on tilt, that she took nearly 3 seconds to decide (suggesting uncertainty), and that she's sizing into a station who won't fold — which means her barrel is likely burning money. That's the difference between generic advice and real coaching.

The Workflow

The actual process is simple once you have the right data:

1. Play a session and export your hands

You need a source of hand histories with behavioral metadata. Standard online poker sites don't provide this level of detail — they export action logs, not decision context. Simulation tools that model opponent psychology can generate richer data. The Pool, for example, exports 33 fields per action including decision reasons, emotional states, timing data, and opponent profiling.

2. Paste the JSON into ChatGPT

Structured JSON is the ideal format for AI analysis. ChatGPT (and Claude) can parse JSON natively — it understands the field names, the relationships between data points, and can cross-reference patterns across hundreds of actions. Plain text hand histories work too, but the analysis is shallower because the data is shallower.

3. Ask specific questions

This is where most people go wrong. "Analyze my play" is too broad. The more specific your question, the better the coaching. Try prompts like:

4. Ask follow-up questions

The initial analysis is just the starting point. The real value comes from drilling into specific hands: "In hand #4843, why was calling the river bet a mistake given that Nadine's fear discount was 0.92?" This kind of contextual follow-up is only possible when the data includes the psychological and strategic metadata.

What You Can Learn

With rich hand data and the right prompts, AI coaching can identify patterns that are almost impossible to see on your own:

Timing leaks. Are you acting faster when you're bluffing? Do you tank longer with marginal hands? If your opponents have timing profiles in the data, you can study their patterns too — and learn to exploit snap-calls and long tanks at the table.

Emotional blind spots. Do you play differently after losing a big pot? If the data tracks emotional states, the AI can compare your decisions when calm versus when you're likely tilted, and quantify exactly how much that tilt costs you per session.

Sizing tells. Are you betting smaller with strong hands and bigger with bluffs? Most low-stakes players have unconscious sizing patterns. Structured bet-fraction data makes these patterns visible instantly.

Opponent-specific adjustments. If the data includes caller resistance profiles and player archetypes, the AI can tell you exactly how to adjust against stations versus nits versus loose-aggressive maniacs — with specific sizing and frequency recommendations backed by the data from your actual sessions.

The Data Quality Problem

The honest truth is that AI poker coaching is only as good as the data you feed it. Standard hand histories from real money sites were designed for hand replayers and basic tracking software — not for AI analysis. They contain what happened but not why it happened.

The next generation of poker tools needs to close this gap. Whether it's through richer exports from online platforms, overlay tools that add context, or simulation environments that generate behavioral data natively — the players who figure out how to feed better data into AI tools are going to have a meaningful edge over those who are still pasting basic hand histories and getting basic advice back.