ESPN's AI Poker Tell Detector Draws Skepticism From Pros

The 2026 World Series of Poker Main Event, which drew over 9,000 entries, became the testing ground for an experimental AI system designed to detect poker tells—the subtle physical cues that might reveal a player's hand strength or intentions. In early July, during the tournament's live broadcast, ESPN began displaying a new overlay tool created by Luke Geel, an AI engineer for the US Air Force. The overlay showed live metrics on player movements, including eye movements, blink rate, posture, chip handling, and hand fidgeting, alongside a "hand strength model" chart that predicted whether a player held a strong made hand, a drawing hand, a bluff, or another hand type.

The system's training was narrow but intensive. It analyzed every hand captured on camera at three designated broadcast tables, building a tells database by correlating player movements and behaviors with actual hand outcomes. This approach aims to digitize the work that professional tells specialists do—reading opponent psychology through body language. However, the poker community's reaction has been decidedly skeptical. Michael Gagliano, a 17-year poker professional who reached the Main Event final table and was competing for the $10 million top prize, pointed out a critical flaw: the vast majority of the tournament's 9,000+ entries never appeared on the recorded tables. Even players who did sit at those tables were present too briefly for the system to accumulate sufficient data on their tendencies. Gagliano said he reviewed every second of ESPN's live streams during the two-and-a-half-week break after the final table was reached in mid-July, searching for tells on his remaining opponents, but the limited screen time made it difficult to extract actionable information—a problem the AI faced equally.

Shaun Deeb, a two-time WSOP Player of the Year who finished 15th in the 2026 Main Event, offered a more fundamental critique. Deeb noted that poker tells extend far beyond what cameras can capture: leg tells, checking tells, verbal tells, breathing tells, and pulse tells. Most of those are invisible on film. He also highlighted the gap between behavioral observation and hand interpretation. "How strong is two pair to one player versus another player?" he asked. A player might project confidence with a weak hand if they mistakenly believe it is strong, or show nervousness with a strong hand because they are responding to the high-stakes environment of the Main Event rather than their actual cards. An AI observing only visual and audio patterns can track confidence or weakness but cannot reliably deduce the intention or situational context behind it.

Geel has been transparent about the tool's limitations. He told WIRED via email that a larger sample of hands would benefit his system and that he has run blind tests on other poker competitions with mixed results. For broadcast entertainment, the tool may hold some appeal, though Deeb remained unconvinced, calling it "a swing-and-a-miss" that ESPN added simply to mimic other sports coverage. Notably, Omaha Productions—the company licensed by ESPN for WSOP and other sports broadcasts—announced that the tool would not be used for the final table, though it provided no explanation for that decision. Looking forward, Deeb and other top professionals acknowledged that as AI improves, such tools could theoretically become more powerful, particularly in high-roller tournaments where hundreds or thousands of hours of footage of the same top pros already exist. However, Deeb noted that the human approach—hiring specialized tells readers to watch opponents in person—remains superior to filmed analysis. Even if an AI provided a perfect tells analysis, a player must still spot and exploit those tells during live play while managing their own exposure. Deeb concluded simply: "I would take my team versus the AI and make a bet on it."