Can AI Read a Poker Face? ESPN’s WSOP Bluff Detector Sparks a Backlash

Can artificial intelligence really tell when a poker player is bluffing? ESPN’s 2026 World Series of Poker coverage tried to answer that question with an experimental AI poker tells system—and the result has become one of the most debated poker technology stories of the year.

The broadcast tool analyzed visible behavior such as eye movement, blinking, posture, chip handling and hand fidgeting. It then displayed a model estimating whether a player might hold a strong made hand, a draw or a bluff.

For viewers, the graphic offered something irresistible: a machine apparently attempting to read a poker face in real time. For professionals, however, the experiment raised much harder questions. Was the model working with enough hands? Could it distinguish genuine confidence from a player who merely misunderstood the strength of a hand? What happened to the physical signals the cameras could not see?

The strongest criticism is not that artificial intelligence can never detect poker tells. It is that a polished probability graphic can look far more certain than the underlying evidence deserves.

This article explains how ESPN’s WSOP bluff detector worked, why players including Michael Gagliano and Shaun Deeb questioned it, why it was removed from the final-table broadcast, and how AI-based tell detection could eventually affect live poker study, security and tournament rules.

What Was ESPN’s AI Poker Tells Detector?

During portions of the 2026 WSOP Main Event broadcast, ESPN displayed an “AI tells detection” overlay. The system was created by Luke Geel, an AI engineer identified in reporting as working for the United States Air Force.

The model used footage from televised Main Event hands to build behavioral profiles. Its inputs reportedly included:

• Eye movements
• Blink rate
• Изменение осанки
• Chip-handling movements
• Hand fidgeting
• Other visible behavioral patterns

After comparing those signals with known hand outcomes, the system attempted to estimate the type of holding a player might have. The television graphic could place probability around categories such as a strong made hand, a drawing hand or a bluff.

That makes the tool different from a normal poker solver. A solver studies actions, ranges, bet sizes and game theory. This system tried to study the person.

Readers unfamiliar with the difference should compare our guide to the best poker analysis tools, calculators and solvers with our article on reading physical tells in poker. One analyzes the decision tree; the other searches for information leaking from human behavior.

Why the Feature Became a Poker Trend

Poker broadcasting has always needed a way to make hidden information visible. The hole-card camera transformed televised poker because viewers could see the cards while the players could not. Real-time equity percentages later made close all-ins easier to understand.

The AI detector attempted the next step: not merely showing what a player held, but suggesting what the player’s body was revealing before the hand ended.

That is a powerful television idea. A viewer sees a player make a large river bet, then watches an AI graphic suggest elevated bluff indicators. The hand becomes a mystery with an on-screen suspect.

The feature also arrived during the WSOP’s major return to ESPN under a multi-year broadcast agreement. The network and Omaha Productions were trying to present poker with the pacing, graphics and storytelling associated with mainstream sports. Our earlier analysis of WSOP streaming and ESPN’s return to poker explains why new broadcast features matter far beyond one tournament.

But poker is not football. A speed graphic can measure how fast a ball travels. A behavioral model cannot directly measure whether someone is bluffing. It estimates a hidden intention from incomplete evidence—and that difference is where the controversy begins.

The Biggest Problem: The AI Did Not Have Enough Data

Machine-learning systems become more useful when they can study large, consistent and correctly labeled samples. The WSOP Main Event produced thousands of players, but only a small fraction appeared repeatedly on televised tables.

The model reportedly used footage from three featured tables. Even a player who reached the late stages might have appeared in only a limited number of broadcast hands. That creates several weaknesses:

• A single player may have too few recorded bluffs.
• Strong hands may occur in different emotional situations.
• Camera angles and lighting may change.
• The player may behave differently early and late in the tournament.
• Known showdowns may represent only a biased portion of all hands.

Michael Gagliano, one of the 2026 Main Event finalists, reviewed the available ESPN footage during the break before the final table. Even after studying the streams closely, he reportedly questioned how much actionable information the limited sample could provide.

If an experienced professional cannot build a confident read from the footage, an AI faces the same basic problem. It may process visual details faster, but it cannot create observations that were never recorded.

A Tell Does Not Automatically Reveal Hand Strength

The deeper problem is conceptual.

Suppose the AI correctly detects that a player appears relaxed. What does that mean?

• The player may hold the nuts.
• The player may be bluffing but has already accepted the risk.
• The player may incorrectly believe a medium-strength hand is unbeatable.
• The player may simply be comfortable on television.
• The player may behave calmly in every large pot.

Physical behavior can reveal emotion, but emotion is not identical to objective hand strength.

A recreational player may look confident with two pair on a board where a professional knows two pair is only a bluff catcher. A professional may appear nervous with a very strong hand because the pot is worth millions of dollars. The body can reflect the player’s interpretation of the situation rather than the actual strength of the cards.

This is why serious чтение рук в покере begins with position, preflop range, board texture, sizing and previous action. A physical tell should adjust that analysis. It should not replace it.

What the Cameras Could Not See

Shaun Deeb argued that physical tells are far broader than facial movement and chip handling. Experienced observers may watch breathing, voice, pulse, leg movement, the way cards are checked, changes in conversation and how a player reaches for chips.

A broadcast camera normally shows only part of the table and only part of the body. It may miss:

• Movement below the table
• Changes in breathing depth
• A visible pulse in the neck
• Foot and leg tension
• The exact way a player first looked at the cards
• Subtle interactions before the televised action began

Audio also creates problems. A microphone may capture a sentence but miss changes in volume, timing or tone clearly enough for reliable comparison. Background noise, commentary and production edits can further reduce consistency.

Human tell specialists do not merely stare at a face. They observe a player over time, establish a baseline and compare behavior in similar situations. A camera-based model with limited hands may notice movement without understanding whether that movement is unusual for the individual.

Why One “Bluff Signal” Is Usually Meaningless

Popular poker culture encourages the idea of one magical tell: touching the nose means weakness, shaking hands mean strength, looking away means a bluff.

Real tell reading is rarely that simple.

A useful read normally requires a cluster of information:

1. Baseline: How does this player normally sit, speak and handle chips?
2. Change: What is different in this specific hand?
3. Синхронизация: Did the change occur before or after the bet?
4. Context: Is the player facing elimination, a small cash-game pot or a televised final table?
5. Range support: Does the betting line contain enough logical bluffs?

An AI overlay can tempt viewers to reverse that order. They see “bluff probability” first and then interpret every movement as proof.

Strong players do the opposite. They build a range first, calculate the price and then use physical information to decide whether a close call moves slightly toward call or fold. Our guide to мышление в покерных диапазонах provides the foundation that no tell-detection graphic can replace.

Why ESPN Did Not Use the Tool at the Final Table

The AI tell detector appeared during portions of the Main Event coverage in July, but Omaha Productions confirmed that it would not be used during the final table. No public explanation was provided in the report announcing the decision.

Several possible reasons are plausible, although they remain interpretations rather than confirmed explanations:

• The model may not have been accurate enough for the biggest broadcast.
• The graphic may have distracted from the players and cards.
• Producers may have wanted to avoid implying that an experimental estimate was reliable.
• Player criticism may have reduced confidence in the feature.
• The final-table production may already have had enough graphics and storylines.

Removing the tool does not prove it failed. Experimental broadcast features are often tested and revised. But the decision matters because the final table would have offered the largest audience and the most repeated footage of the same players.

Is AI Tell Detection Cheating?

As a television graphic shown to viewers, the tool is not the same as a player receiving real-time assistance at the table. The finalists cannot see the broadcast feed while making decisions, and televised poker normally operates with a delay to protect hole-card information.

The ethical problem changes completely if a player or assistant can access similar analysis during live action.

Imagine smart glasses that identify an opponent, compare current behavior with thousands of recorded hands and privately display a bluff estimate. That would move the technology from entertainment into real-time assistance.

Most serious events already restrict electronic devices, outside communication and tools that can provide strategic help. The broader integrity problem is connected to the issues examined in our report on live poker cheating, RFID cards and hidden-camera risks.

Online poker faced a similar boundary with bots and real-time assistance. Studying after a session is different from receiving decisions while the hand is active. Our guide to Покерные боты и RTA explains why timing of assistance matters as much as the technology itself.

Could AI Become Better Than Human Tell Readers?

It is possible, but the system would need far better data and a more careful definition of success.

A future model could study thousands of hours featuring the same high-stakes professionals. It could compare:

• Behavior before checking versus betting
• Movement in bluffs versus value bets
• Changes by stack depth and tournament stage
• Speech patterns in small and large pots
• Individual baselines rather than universal tells
• How behavior changes against specific opponents

AI has an advantage in memory. A human observer may forget how a player handled chips six months ago; a model can compare thousands of tagged clips.

However, more footage does not automatically solve the labeling problem. Many poker hands end without showdown, so the true cards remain unknown. Even when cards are revealed, the system must distinguish value bets from thin value, semi-bluffs from pure bluffs and confident mistakes from correct assessments.

The best future tool may not say, “This player is bluffing.” It may say something narrower and more honest: “This player’s blink rate and posture differ significantly from their personal baseline in similar river situations.” The poker player would still need to decide what that change means.

AI Poker Tells vs Poker Solvers

These tools answer different questions. A solver asks what a theoretically strong strategy looks like. A tracker asks how often an opponent takes an action. A tell detector asks whether visible behavior changes when hidden information changes.

Confusing those purposes creates bad strategy. An AI model can identify tension without proving that a call is profitable. A solver can recommend a bluff without knowing that a specific opponent never folds. Human judgment remains necessary.

Our article AI Can Talk Poker—But Can It Actually Play? examines the same gap between generating convincing analysis and producing reliable decisions.

What Poker Players Can Learn From the ESPN Experiment

The controversy is useful even if the current model is not accurate enough to beat professional observers.

1. Build Individual Baselines

Do not search for universal body-language rules. Watch how one opponent behaves in ordinary hands, then notice meaningful changes.

2. Compare Similar Situations

A player’s behavior in a $50 pot should not be compared directly with behavior while facing Main Event elimination. Pressure changes the baseline.

3. Use Tells Only for Close Decisions

If the pot odds and range analysis create an obvious fold, one eye movement should not turn it into a call.

4. Watch Your Own Information Leakage

Players often spend all their energy reading opponents and ignore their own repeated patterns. Recording a practice session can reveal timing, breathing or chip-handling habits.

5. Do Not Let Technology Replace Observation

AI can organize footage and highlight patterns, but it may also reinforce false correlations. Review the original hands and ask whether the pattern makes strategic sense.

Used responsibly after play, this kind of technology could become a training aid. Our guide to using AI for poker study without cheating explains the line between legitimate analysis and prohibited assistance.

Could This Change Poker Broadcasts?

Yes, even if the first version disappears.

Future broadcasts may use AI to:

• Identify unusual changes in player behavior
• Search old footage for comparable hands
• Measure decision time by street and bet size
• Build player-specific tendency profiles
• Suggest hands for commentators to review
• Create interactive predictions for viewers

The danger is presentation. A probability bar can look scientific even when the sample is tiny. Broadcasters should show uncertainty clearly, explain what data was used and avoid presenting correlation as proof of a bluff.

The best version would support commentary rather than replace it. A commentator could say that the model noticed a behavioral change, then explain why the betting line still matters more.

AI Poker Tells FAQs

What is the ESPN AI poker tells detector?

It is an experimental broadcast tool used during portions of the 2026 WSOP Main Event coverage. It analyzed visible player behavior and estimated possible hand categories such as strong hands, draws or bluffs.

What behavior did the AI analyze?

Reported inputs included eye movement, blink rate, posture, chip handling and hand fidgeting captured by broadcast cameras.

Who created the WSOP tell-detection tool?

The system was developed by Luke Geel, an AI engineer identified in reporting as working for the United States Air Force.

Was the AI used at the WSOP final table?

No. Omaha Productions said the feature would not be used during the final table, although no public reason was given in the report.

Why were poker professionals skeptical?

The main concerns were the limited number of televised hands, incomplete camera coverage and the difficulty of connecting visible emotion directly to objective hand strength.

Can AI know when someone is bluffing?

AI can identify repeated behavioral patterns, but a pattern does not prove intent. Reliable bluff detection still requires context, ranges, bet sizing and a sufficient individual sample.

Is using AI to study poker footage allowed?

Post-session study may be allowed, depending on the event and platform. Receiving real-time automated assistance during a hand is a different issue and may violate tournament or poker-room rules.

Could smart glasses detect poker tells?

Future wearable systems could theoretically analyze behavior in real time, which is why electronic-device and outside-assistance rules may become more important.

Are physical tells more important than poker ranges?

No. Range construction, pot odds and betting action should lead the decision. Physical tells are most useful as a secondary adjustment in close situations.

Will ESPN use AI tells again?

No future plan was confirmed in the reporting available at the time of writing. The feature may be revised, tested again or abandoned.

Final Verdict: Impressive Graphic, Unproven Read

ESPN’s AI poker tells detector created exactly what a new television feature is designed to create: curiosity, argument and attention.

The concept is credible. Human behavior contains patterns, and machines can compare more footage than any individual observer. With enough high-quality data, AI may eventually become useful for identifying changes that humans miss.

The current claim must remain narrower. Limited camera coverage, small player samples and the difference between emotion and actual hand strength make confident bluff predictions extremely difficult. The system may detect movement, but movement still requires interpretation.

The most important lesson for poker players is not to fear a machine that can read every bluff. It is to recognize how easily a professional-looking probability can create false certainty.

Use tells as evidence, not answers. Build the range, calculate the price, understand the player and then decide whether the behavior changes the conclusion. That process remains more reliable than any single graphic—human or artificial.