ESPN · WSOP Main Event · 2026

How My Poker Tell AI Became an ESPN WSOP Feature

I built an experimental model that looks for correlations between a player's visible behavior and hand strength. Omaha Productions turned that work into a television storytelling feature for ESPN's 2026 World Series of Poker Main Event coverage.

What Appeared on ESPN

The broadcast graphic summarized an AI estimate of whether a player's visible behavior looked more consistent with a bluff or a strong hand. The segment appeared as the World Series of Poker returned to ESPN, with coverage produced by Omaha Productions. The system was a post-action broadcast device: it was not placed in front of players and was not a tool they could consult during a hand.

Sportico's Jacob Feldman first reported the project and its use in the broadcast. The story was then discussed by poker publications, science and technology outlets, and social accounts around the world. The press index collects those independent reactions in one place.

How the Model Worked

The project began with historical poker footage whose outcomes were already known. I assembled examples around a specific player and measured visible cues such as blink rate, gaze, mouth position, posture, hand movement, and how often the player checked their cards. The model then searched for patterns that differed between labeled hand-strength categories.

That distinction matters. The system did not infer a universal tell that applies to every poker player. It looked for player- and sample-specific correlations. Building a useful dataset required manual review, careful labeling, and about six months of experimentation; it was not a one-click process that could accept any YouTube link and reliably find tells.

What the Result Means—and What It Does Not

A probability estimate is not proof. Visible behavior can change because of fatigue, table conversation, lighting, an opponent, or simple noise. A small historical sample can also produce correlations that disappear on new footage. The responsible interpretation is that the model found patterns worth presenting as entertainment and further investigation, not that it solved deception or could identify every bluff.

• It measured behavior visible in recorded video.
• It used known historical outcomes as labels.
• It produced estimates, not certainty.
• It was a retrospective broadcast feature, not real-time assistance.
• Results for one player or dataset should not be generalized to everyone.

Why the Project Matters

Poker broadcasts already reveal cards, action, pot size, and equity to viewers. Behavioral analysis adds another possible layer of storytelling: whether a player's physical routine changed when the pressure rose. The same methods may also help players review their own footage for repeated habits, provided the limits and uncertainty are made explicit.

The public reaction was divided. Some viewers saw a novel sports graphic; others questioned the sample size, usefulness, or future integrity risks. Those critiques are part of the story. A responsible poker AI project should make the difference between measurement and prediction clear and should never market probabilistic correlations as mind reading.

Frequently Asked Questions

Who created the poker AI used in ESPN's WSOP coverage?

Luke Geel, an independent AI engineer and the founder of Geel AI, built the experimental behavior-analysis model.

Did ESPN's poker AI know the players' cards?

No. The model analyzed visible behavior from video. The broadcast could show cards to viewers, but hidden-card access was not an input to the behavioral estimate.

Was it a real-time bluff detector?

No. It was a retrospective broadcast feature. It was not provided to players and should not be described as a live decision tool.

Can the model analyze any poker player?

Not reliably from an arbitrary clip. The experiment depended on enough usable footage, known outcomes, consistent camera views, and player-specific analysis.