ESPN’s AI Poker Tell Tracker Sparks Debate at the 2026 World Series of Poker

ESPN’s new AI tell-detection overlay at the 2026 World Series of Poker Main Event has triggered a fresh argument inside poker: useful broadcast innovation, or overhyped technology that misunderstands how live reads really work? The system, which appeared during early July coverage, tried to estimate player tendencies from camera footage, but many pros say it lacks the data, context and subtlety needed to meaningfully predict what a player is holding.

The controversy matters because poker is built on incomplete information, and any tool that claims to decode body language could reshape how fans watch the game and how players think about being studied on camera. It also raises a broader question about where AI can add value in sports broadcasting without pretending to do the work of elite human judgment.

What ESPN showed viewers looked polished enough: movement metrics, a hand-strength model, and on-screen AI analysis of live table behavior. But the reception from players was far less enthusiastic, with several top professionals arguing that camera-based analysis cannot capture the full range of poker tells, especially in a tournament with thousands of entrants and very limited screen time for any one player.

What exactly did ESPN’s AI tool try to do?

ESPN’s broadcast feature attempted to infer whether a player was strong, weak, bluffing, or drawing based on visible behavior at the table. It pulled from camera feeds used in the tournament broadcast and layered that with automated readings of body language, blinking, eye movement, posture, chip handling and small repetitive gestures.

The result was presented as a kind of live tells dashboard, designed to give viewers another lens through which to interpret the action. In theory, it was meant to turn a notoriously subjective part of poker into something quantifiable and visual.

In practice, the tool may have done something else: it gave the audience a simplified model of a game that is built on nuance, context and deception.

Who built the system?

The tool was created by Luke Geel, an AI engineer who works for the U.S. Air Force. According to the broadcast description and Geel’s own explanation, the system was trained by reviewing filmed hands from the 2026 WSOP Main Event and using those observations to build a tells database for various players.

Geel has acknowledged that more data would improve the system. He also said he has tested similar approaches in other poker settings with mixed outcomes, suggesting the model is still experimental rather than fully reliable.

How does AI tells detection work in poker?

AI tells detection works by converting visible behavior into data points, then looking for patterns that may correlate with hand strength or decision-making. In this case, the system evaluated movements such as blinking, gaze shifts, posture changes, chip handling and small fidgets while comparing those behaviors with what happened in the hand.

The idea is straightforward: if a player tends to move one way when bluffing and another way when holding a strong hand, a model might detect that tendency and surface it to viewers or analysts.

That said, poker is not a laboratory experiment. The same physical behavior can mean very different things depending on stack depth, table image, payout pressure, opponent skill and stage of the tournament. A body language cue is not the same thing as a confession.

Why are poker professionals skeptical?

Poker professionals say the main problem is not whether cameras can spot movement, but whether those movements actually reveal the truth about a hand. The most valuable tells are often subtle, multi-layered and context-dependent, and many never make it onto a broadcast feed at all.

Shaun Deeb, a two-time World Series of Poker Player of the Year and one of the game’s best-known figures, argued that the popular image of a poker tell is too simplistic. He said real tells extend beyond obvious gestures and can include breathing, pulse, speech patterns, checking habits and physical changes that a camera simply cannot reliably capture.

Poker players say the public often thinks of tells as one obvious clue, like a nervous snack or a dramatic glance, but the reality is much broader and far more difficult to read from television footage alone.

That skepticism is not just about technology. It is about the limits of any remote observer. Even a sharp human watching the live stream has less information than someone sitting at the table. The AI, critics argue, inherits all of those limitations and then adds some of its own.

What did pros say about the sample size?

They said it was too small to support strong conclusions. The 2026 WSOP Main Event drew more than 9,000 entries, but ESPN’s cameras only followed a small number of tables, and even those tables rotated often enough that no single player generated a large amount of consistent footage.

Michael Gagliano, a veteran pro who reached the final table this year, said the broadcast structure itself limits how much usable information can be collected. He spent the two-and-a-half-week gap before the final table reviewing every second of the live coverage, but still found the amount of actionable information thin.

That is a major obstacle for any machine-learning system. Models need repetition and variety to distinguish meaningful tendencies from noise. In the Main Event, the footage may be extensive in total, but not necessarily deep enough for any one player.

What makes poker tells harder than they look?

What makes poker tells hard is that they are often about the person, the moment and the pressure, not just the cards. A player can look confident with a weak hand because they are naturally composed, or look tense with a strong hand because they are in a high-stakes spot.

That problem is especially acute at the Main Event, where life-changing money, long endurance and the stage of the tournament can all influence behavior. A hand that feels routine to a seasoned pro can still trigger anxiety if the pay jump is enormous.

Gagliano said body language may reflect the emotional weight of the moment rather than the actual strength of the hand. In his view, a player’s demeanor can be shaped by how strong they believe their hand is relative to the situation, which is not always the same as the objective hand category.

That is one of the biggest weaknesses of any AI that tries to interpret poker from the outside. It may identify confidence or fear, but confidence and fear are not the same thing as strength and weakness.

Why the Rounders comparison only goes so far

The famous movie example of a tell being reduced to a snack habit is memorable because it dramatizes a real poker tension: the battle to conceal or reveal intention. But professionals say the reality is far more complicated than a single behavioral quirk.

Deeb said the public tends to overestimate how neat and obvious physical reads are in real life. The most meaningful tells, he suggested, are distributed across many small signs rather than one dramatic clue. That means any system built mainly from video is likely to miss more than it sees.

Even if a model could detect that a player appears uneasy, that observation only has value when paired with deep knowledge of that player’s style, tendencies and the specific hand state. Without that context, the data can be misleading.

How useful is the AI as a broadcast feature?

How useful the feature is depends on whether it is judged as analysis or entertainment. If the goal is to provide viewers with a fresh angle on televised poker, the tool may serve as a conversation starter. If the goal is to produce consistently accurate reads, the experts say the system falls short.

Geel has not presented the tool as an omniscient replacement for expert judgment. Instead, the project appears to be an exploratory attempt to see whether existing broadcast video can be mined for meaningful behavioral patterns. That puts it closer to experimental sports data visualization than to a finished competitive weapon.

Still, some players see the concept as misplaced. Deeb was blunt in his assessment, saying he believed the broadcast was trying to bolt a sports-tech gimmick onto a game whose central appeal already lies in human psychology and hidden information.

One of the game’s top veterans described the feature as a missed opportunity, arguing that it was more a novelty than a genuine advance in understanding poker.

ESPN’s licensing partner, Omaha Productions, later said the feature would not be used for the Main Event final table. The company did not publicly explain the change.

Why did ESPN drop the feature for the final table?

Why the tool was absent from the final table broadcast has not been officially detailed, but the decision signals caution. The final table is the most visible stage of the tournament, with the greatest audience attention and the highest stakes for both players and network presentation.

In that setting, an experimental overlay that might overstate its certainty could draw more criticism than praise. Producers may also have recognized that the final table features fewer players and far more scrutiny, making questionable outputs easier to challenge in real time.

For a network, the trade-off is delicate. Broadcast innovation can make a game feel modern and data-rich, but a shaky feature can also break trust with the very audience it wants to attract.

What are the limits of remote tell detection?

Remote tell detection is limited by what the camera can actually see and by what the model can infer from those images. A live stream usually shows only part of the table, from fixed angles, under lighting that may not be ideal for behavioral analysis.

Many live tells happen outside the camera frame: a player’s breathing pattern, foot movements, subtle tremors, off-camera conversation, a brief glance toward a stack, or a shift that only matters because of timing and table dynamics. Those details are often invisible on television.

Even when the camera catches a movement, it may not know whether the action happened because a player was bluffing, thinking, cold, tired, irritated, excited or simply adjusting in the chair.

• Broadcast footage is incomplete by design.
• Many tells are too small or too situational for cameras.
• Players can change behavior once they know they are being studied.
• Body language is not the same thing as hand strength.
• Machine learning still depends on the quality and quantity of data.

How do poker pros study opponents today?

How poker pros study opponents today depends on the event, the stakes and the amount of information available. In big televised tournaments, many players and coaches already review footage, track betting patterns and compare hand histories in order to spot habits that may repeat.

Deeb said he has worked with teams that include live tell specialists, and that the best version of that process still involves watching players in person rather than through a screen. According to him, teams often prefer a human observer sitting near the action to a remote analyst studying television footage later.

That distinction matters because live observation captures body language in real time, from a nearby angle, while also allowing the observer to note context that a camera may miss. In other words, the machine may be able to sort data, but the people doing the best work are still gathering better data to begin with.

Gagliano described a similar pattern. He reviewed streams during his run not because they told the whole story, but because they offered some clues. In poker, even partial clues can matter. The point is not certainty; it is incremental advantage.

Could AI become more useful in high-roller poker?

Could AI become more useful in high-roller poker? Yes, but likely in a narrower way than ESPN’s broadcast experiment suggests. The high-stakes circuit features a smaller set of elite players who appear on camera repeatedly, which means there is more footage to study and more opportunity for patterns to emerge.

Those conditions are far better for AI than a giant field like the Main Event. In high rollers, the same professionals often face each other in multiple events, and there may be hundreds or thousands of hours of archived broadcast footage available for analysis.

That creates a more plausible use case for machine learning: helping coaches and analysts sort through the enormous volume of film and identify areas worth watching more closely. It could become a filtering tool, a kind of assistant that points humans toward promising moments rather than claiming to solve the game outright.

Even so, the most valuable use may still be indirect. AI might help identify habits, but humans would still need to decide whether those habits matter, when they matter and how much to trust them.

What about devices at the table?

What about devices at the table is where the conversation shifts from analysis to security. Live poker rooms and tournaments generally prohibit electronic aids on the table, and some events go even further by tightening restrictions on what can be brought near the action.

Deeb predicted that wearable tech such as smart glasses will eventually face stricter bans because the line between study and assistance can blur quickly. If a player is able to receive real-time guidance, the concept of fair competition changes immediately.

That is why the broadcast tool and a live competitive tool are not the same thing. One is commentary. The other would be cheating.

Why this debate matters beyond poker

Why this debate matters beyond poker is that it mirrors a larger fight across sports and entertainment about what AI should be allowed to infer from human behavior. From broadcast enhancements to scouting tools, organizations increasingly want to use machine learning to transform visual data into prediction.

But the poker dispute shows a recurring limit: some human activities are defined by ambiguity, and ambiguity resists clean automation. The more a game depends on psychology, timing and hidden intent, the harder it is to reduce to a score generated by a camera.

That does not mean AI has no place in poker coverage. It may help explain tendencies, present statistics or make broadcasts more interactive. But there is a difference between enriching a viewing experience and claiming to decode the inner state of a person at the table.

For now, poker pros appear comfortable keeping that distinction intact.

The bottom line

ESPN’s AI tell-detection experiment has highlighted both the promise and the limits of applying machine learning to poker. The system may have offered viewers a sleek new way to watch the game, but the players who know poker best argue that camera-based analysis cannot replace the judgment, context and in-person observation that serious tell work requires.

For now, the strongest verdict is that the feature is interesting as a broadcast experiment, but not convincing as a true poker edge. In a game where a person’s hesitation can mean five different things, the old-fashioned human read still appears to hold the advantage.

Deeb summed up the skepticism in competitive terms, saying he would still back his own team over the AI if the two ever had to go head-to-head.