Sports
Last reviewed
May 13, 2026
Sources
75 citations
Review status
Source-backed
Revision
v2 ยท 5,183 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 13, 2026
Sources
75 citations
Review status
Source-backed
Revision
v2 ยท 5,183 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: Sports ChatGPT Plugins
Artificial intelligence has reshaped almost every layer of professional and amateur sports, from how leagues officiate matches to how coaches design tactics, how broadcasters package highlights, and how doctors predict soft tissue injuries. The shift accelerated in the late 2010s when computer vision became cheap enough to install in every stadium, and again in the early 2020s when deep learning replaced rule-based heuristics in tracking, officiating, and content generation. By the mid-2020s most major leagues, including the NBA, NFL, MLB, Premier League, and Major League Soccer, ran some form of machine learning on every game.
AI in sports covers a wide territory. The most visible applications are the ones fans see during a broadcast: the goal-line technology graphic, the offside line drawn during a VAR review, the strike-zone overlay on a baseball telecast, the speed reading on a tennis serve. Behind those graphics sits a network of cameras, radar units, inertial measurement units, and machine learning models that produce structured data at frame rates of 25 to 100 hertz.
That data feeds back into many other parts of the sport. Front offices use it for scouting and contract decisions. Coaches use it for tactical analysis. Sports scientists use it to manage training load and predict injuries. Broadcasters use it for automated highlights and live commentary. Sportsbooks use it to price odds in real time. Fan engagement teams use it to drive chatbots, personalised recap clips, and second-screen experiences in league apps.
The table below lists categories that this article covers in detail.
| Category | Representative systems | Notable deployments |
|---|---|---|
| Officiating | Hawk-Eye, VAR, semi-automated offside | Tennis Grand Slams, Premier League, FIFA World Cup |
| Player tracking | Second Spectrum, SportVU, Hawk-Eye, Sportlogiq | NBA, NHL, MLB, NFL, La Liga |
| Wearables | Catapult, Whoop, Oura, Playermaker | Premier League, NFL, NCAA, Olympic teams |
| Scouting | Stats Perform Opta, Wyscout, SciSports, Olocip | Premier League clubs, La Liga, MLS |
| Health and injury prediction | Zone7, Kitman Labs, Digital Athlete | Liverpool FC, NFL, Premier League |
| Broadcasting | IBM Watson, WSC Sports, Pixellot, Veo | US Open, Wimbledon, NBA, college sports |
| Sports betting | DraftKings, FanDuel, Simplebet | US online sportsbooks |
| Research | TacticAI, Digital Athlete, DeepMind robotic football | DeepMind/Liverpool, NFL/AWS |
The statistical revolution that prepared sports for AI started in baseball. The term sabermetrics was coined by writer Bill James in 1980 from the acronym of the Society for American Baseball Research, founded in 1971. James started publishing his annual Baseball Abstract in 1977, arguing that traditional baseball statistics undervalued on-base percentage and walks. His ideas spread slowly inside the sport for two decades.
The Oakland Athletics, first under general manager Sandy Alderson in the early 1990s and then under Billy Beane from 1997, were the first front office to systematically apply sabermetric thinking to roster construction. Their 2002 season, in which they won 20 consecutive games on a payroll a third the size of the New York Yankees', became the subject of Michael Lewis's 2003 book Moneyball and the 2011 film of the same name.
The machine learning era began in the late 2000s when SportVU, an optical tracking system created in 2005 by Gal Oz and Miky Tamir, two engineers with backgrounds in Israeli Defence Forces missile tracking, was acquired by STATS LLC in 2008. STATS adapted the technology for basketball and signed four NBA teams for the 2010-11 season. By the 2013-14 season SportVU cameras were installed in every NBA arena.
The deep learning era arrived with two parallel shifts. First, optical tracking replaced the radar and RFID hybrids that had powered MLB Statcast since 2015. The 2020 season was the first in which Hawk-Eye cameras tracked every pitch, hit, and player at major league parks. Second, AI-generated content moved from research demos to live production. IBM Watson clipped highlights at the 2017 Masters Tournament, Wimbledon, and US Open. By the 2024 Paris Olympics, NBC Peacock was serving personalised recap videos narrated by an AI clone of Al Michaels.
Every major league now runs an optical or hybrid tracking system that produces positional data for every player and the ball at 10 to 100 frames per second. The data is the substrate for almost every downstream AI application in the sport.
| System | Sport | Provider | Active in | Notes |
|---|---|---|---|---|
| SportVU | Basketball | STATS LLC | 2010-11 (4 teams), 2013-14 (all NBA), retired 2017 | Six cameras per arena, 25 Hz |
| Second Spectrum | Basketball | Genius Sports (acq. 2021) | 2017-18 NBA season onward | Replaced SportVU as official NBA optical provider |
| Statcast (Hawk-Eye era) | Baseball | Hawk-Eye Innovations and Google Cloud | 2020 onward | 12 cameras per park, 5 dedicated to pitch tracking at 100 fps; player tracking at 50 fps; +/- 0.1 inch accuracy |
| Next Gen Stats | American football | Zebra Technologies, Wilson, AWS | 2014 (RFID rollout), 2017 (AWS partnership) | RFID chips in shoulder pads (10 Hz) and ball (25 Hz); 20+ ultra-wideband receivers per stadium |
| Hawk-Eye Live | Tennis | Hawk-Eye Innovations (Sony) | 2020 US Open partial, 2021 Australian Open full, ATP-wide from 2025 | 18 cameras, 3 mm accuracy, 100 ms response |
| Sportlogiq | Hockey | Sportlogiq, acquired by Teamworks 2025 | 31 of 32 NHL clubs as clients | Computer vision on broadcast feeds |
| Stathletes | Hockey | Stathletes | NHL data partner | Manual and deep learning annotation; Waterloo collaboration reports 94.5% player tracking accuracy |
| Opta Vision | Football | Stats Perform | Premier League, EFL, La Liga, MLS | Synchronised event and tracking data, 25 Hz x/y positions |
| Catapult Vector | Multi-sport | Catapult Sports | 5,000+ teams in 40+ sports including NFL, EPL, MLB, NHL, AFL, NCAA | GNSS plus 9-axis IMU at 1000 Hz internal sampling |
| Playermaker | Football | Playermaker | 80+ clubs across Europe and North America | Boot-mounted 6-axis sensors, no infrastructure required |
| KinaTrax | Baseball | KinaTrax | 75+ stadiums and labs in MLB, MiLB, NCAA | Markerless 3D pose at 300 fps, 20 joint centres |
The NBA's transition from SportVU to Second Spectrum in the 2017-18 season was the first time a major North American league switched its full-league tracking provider. Second Spectrum's machine vision pipeline produced not just raw coordinates but derived event data: a label for every pick-and-roll, every drive, every defensive switch. Genius Sports acquired Second Spectrum in 2021 and extended its NBA contract into a multi-year agreement covering analytics, broadcast graphics, and the NBA League Pass second screen.
MLB Statcast's 2020 overhaul, built jointly by Hawk-Eye and Google Cloud, was the largest single jump in baseball data quality since pitch tracking arrived in 2008. The system measures spin axis and pitcher release point directly rather than inferring them from ball flight, and it generates pose estimates for every player at 30 Hz. The dataset has driven a wave of pitch design research at teams like the Los Angeles Dodgers and Tampa Bay Rays.
The NFL's Next Gen Stats has been jointly developed with AWS since 2017. AWS stores about 300 million data points per season and runs roughly 75 machine learning models against them. The 2024 season added two new derived stats, Tackle Probability and Offensive Shift and Motion Classification, both produced by models trained on years of tracking data. AWS also operates the Digital Athlete simulation discussed later in this article.
The most visible AI applications in sport happen on the field of play. Hawk-Eye, founded by Paul Hawkins in 2001, started as a television graphic for cricket and became an officiating standard within five years.
Hawk-Eye's first competitive deployment was at a Test match between Pakistan and England at Lord's on 21 April 2001. The system uses six or more high-speed cameras to triangulate ball position to within 2.6 mm and is now used to adjudicate leg-before-wicket reviews under the Decision Review System and in tennis line calls.
In tennis, Hawk-Eye debuted at the 2006 Hopman Cup in Perth and was approved for Grand Slam use later that year, with the US Open the first major to allow player challenges. The system evolved into Hawk-Eye Live, an automated line-calling product that does not require a chair umpire challenge. The 2020 US Open used Hawk-Eye Live in place of human line judges for all matches except those held at Arthur Ashe Stadium and Louis Armstrong Stadium. The 2021 Australian Open became the first Grand Slam to use electronic line judges on every court. The ATP Tour announced in April 2023 that all men's tour events would use Electronic Line Calling Live from the 2025 season.
The first sanctioned trial of camera-based goal-line technology in football was at Fulham's Craven Cottage in the 2006-07 season. After the controversial Frank Lampard non-goal at the 2010 World Cup, the IFAB approved goal-line technology in 2012. FIFA used GoalControl, a German rival to Hawk-Eye, for the 2014 World Cup. The Premier League chose Hawk-Eye and rolled it out for the 2013-14 season, starting with the Community Shield in August 2013.
The Royal Netherlands Football Association proposed video review under the name Refereeing 2.0 in 2010. The first live trial was a friendly between PSV and FC Eindhoven in July 2016. The first competitive use was an Ajax versus Willem II KNVB Cup tie on 21 September 2016. FIFA formally approved VAR for the 2018 World Cup at a Council meeting in Bogota on 16 March 2018, and Russia 2018 became the first World Cup to use the system.
Semi-Automated Offside Technology (SAOT) followed at the 2022 World Cup in Qatar. The system uses 12 dedicated tracking cameras under the stadium roof to record up to 29 data points per player at 50 Hz, paired with an inertial measurement unit inside the Adidas Al Rihla match ball that streams data at 500 Hz to the video operation room. When a player receives the ball in an offside position, the system flags the moment to the VAR team. The 3D animation is then pushed to stadium screens and broadcasters. UEFA introduced SAOT for the group stage of the 2022-23 Champions League, slightly ahead of Qatar 2022. The Premier League followed with SAOT on Matchweek 32 of the 2024-25 season, on 12 April 2025, after testing in the FA Cup. The English implementation, supplied by Genius Sports rather than Hawk-Eye, uses standard cameras without a chipped ball and reduces close offside decision times by about 30 seconds on average.
The production side of sport runs on an increasingly autonomous content pipeline.
IBM piloted Cognitive Highlights at the 2017 Masters Tournament, using computer vision, audio analysis, and TV graphic recognition to flag clip-worthy moments. The same workflow was used at Wimbledon 2017 and at the 2017 US Open, where Watson assembled clip reels within five minutes of the end of every match, cutting the lag from previous years by two to ten hours.
Israel-based WSC Sports has become the back-end provider for automated highlights at over 300 rights holders, including the NBA, La Liga, the ATP, Major League Soccer, the Bundesliga, the J.League, the PGA Tour, and Cricket Australia. The platform ingests a live broadcast feed, detects events using computer vision and graphics recognition, and packages clips for social, app, and broadcast destinations. In the first half of 2025 alone it produced over 8 million video clips, a 52 percent year-over-year increase, and La Liga has used the platform to publish more than 260,000 highlight assets per season. In 2024 WSC began offering generative AI features that can write commentary scripts and produce synthetic voiceovers, including multilingual NBA commentary.
For the 2024 Paris Olympics, NBC's streaming service Peacock launched a feature called Your Daily Olympic Recap that produced personalised highlight videos narrated by an AI clone of Hall of Fame announcer Al Michaels. Users picked sports and topics, and the system pulled hundreds of NBC Sports clips into a ten-minute package each day. NBC Universal said about 7 million possible playlists could be assembled. Michaels approved the use of his voice. The recaps greeted viewers by name and were among the first uses of a cloned commentator voice at a major live sporting event.
At the grassroots and college level, AI cameras have replaced cameramen for tens of thousands of teams. Pixellot, founded in 2013, runs more than 30,000 camera installations and is on track to stream 1.5 million games across 14 sports this year. Its largest partnership, with the youth baseball app GameChanger, started in 2025 after a pilot the previous summer. Veo, a Danish company, sells dual-lens 4K cameras tuned for amateur football and is used by more than 40,000 clubs across 100 countries, recording over 4 million matches as of 2026.
Professional clubs combine event data, tracking data, and biometric data to rank players the way a hedge fund ranks stocks. The dominant data providers in football are Wyscout and Stats Perform's Opta. Stats Perform covers more than 1,000 leagues and 500,000 matches per year and supplies AI-enriched metrics through its Opta Vision platform, which fuses event and tracking data into a single synchronised feed.
SciSports, founded in the Netherlands in 2012, sits on top of Wyscout's data and adds machine learning features such as player role detection, expected goals, and quality-of-action ratings. Olocip, founded in 2016 by former Real Madrid midfielder Esteban Granero, sells predictive scouting and tactical analysis to clubs in Spain, the Premier League, and the MLS. Real Madrid's analytics group has used Olocip's models for player evaluation and contract decisions.
In baseball, scouting has been heavily quantitative since the Moneyball era, but the introduction of pose-tracking data from Hawk-Eye and KinaTrax has shifted attention toward mechanics. KinaTrax's 8-camera markerless system is deployed in over 75 MLB and minor-league parks and produces 20-joint pose estimates at over 300 frames per second. Driveline Baseball, a private training facility founded by Kyle Boddy in 2008, has built much of its competitive edge on combining Rapsodo pitch trackers, KinaTrax pose data, and machine learning models for pitch design.
In hockey, Sportlogiq's broadcast-feed analysis is used by 31 of the NHL's 32 clubs, by 42 NCAA Division I programmes, and by IIHF federations. Stathletes is the other major Canadian hockey analytics vendor, and a 2024 University of Waterloo collaboration reported deep learning models that track players from broadcast video with 94.5 percent accuracy and identify individual players with 83 percent accuracy.
The single largest financial loss in elite team sport is time lost to injury. Premier League clubs lost roughly 21,000 player-days to injury in the 2020-21 season, which Zone7 estimates was worth up to 197 million pounds in contract value. AI-driven injury risk models are now a routine part of medical and performance departments.
Zone7, an Israeli company, sells an injury prediction service that ingests training load, GPS, biometric, and medical data, then forecasts elevated injury risk one to seven days in advance. The company reports that across 11 teams its models would have alerted clinicians 72.4 percent of the time that a real injury occurred. Liverpool FC has used Zone7 since the start of the 2021-22 season, and the system was adopted by Leeds United in June 2022.
Kitman Labs, based in Dublin, sells a broader performance intelligence platform that combines injury risk modelling with workload management, readiness testing, and medical record management. In October 2023 the company signed a contract with the Premier League to deploy a centralised Football Intelligence Platform across all Premier League academies. Its client list includes teams in the NFL, NBA, MLS, NWSL, Premiership Rugby, and the United Soccer League, as well as governing bodies including the Rugby Football Union.
The NFL Digital Athlete, jointly developed with AWS since 2019, is a simulated digital twin of an NFL player. The system uses 38 calibrated cameras at each stadium, combined with sensor data from equipment, to run millions of biomechanical simulations of game scenarios and identify positions, players, and play types associated with elevated injury risk. Risk Mitigation Modelling produces training load recommendations, and 3D Pose Estimation analyses real movement against simulated ideals. The NFL reported that the 2024 season had its lowest recorded concussion rate, a 17 percent year-over-year decline that the league attributed in part to Digital Athlete-informed equipment and rule changes.
Catapult Sports, founded in Melbourne in 2006, sells the Vector S7, T6, and T7 GNSS-plus-IMU units that are worn by athletes in over 5,000 teams across 40 sports, including most NFL, EPL, and NRL clubs. Whoop and Oura have become standard recovery and readiness wearables for individual athletes. Both run AI models that combine heart rate variability, resting heart rate, and sleep stage data into daily readiness scores. The NBA briefly distributed Whoop straps to all players during the 2020 Orlando bubble, although the league's collective bargaining agreement restricts how teams can use the data.
The rise of legal sports betting in the United States, which began in earnest after the Supreme Court's 2018 Murphy v. NCAA decision struck down the federal sports betting ban, coincided with the maturation of machine learning models for pricing odds. DraftKings and FanDuel, the two largest US operators, both use machine learning across pricing, parlay generation, fraud detection, and customer risk.
DraftKings integrated Simplebet's machine learning models into its trading platform to expand the catalog of in-play markets, particularly for micro-bets such as the outcome of the next pitch or the next play. The company has publicly described testing new models in shadow mode against its production pricing engine. FanDuel uses anomaly detection to flag deposit patterns that differ from a customer's history and to surface suspicious betting activity to compliance teams.
The global market for AI-driven sports betting was estimated at 1.2 billion dollars in 2024, growing at a forecast 14.7 percent compound annual growth rate. Regulators including the UK Gambling Commission and the Massachusetts Gaming Commission have asked operators to document how their models price markets and how they intervene in problem gambling cases.
In March 2024 Google DeepMind and Liverpool FC published TacticAI: An AI Assistant for Football Tactics in Nature Communications. The paper, by Petar Velickovic and co-authors, describes a geometric deep learning model trained on 7,176 corner kicks from the 2020-21 Premier League season, with later versions trained on 9,693 corners from three seasons. Each corner was encoded as a graph in which players were nodes carrying position, velocity, and physical attribute features. The model can predict which player will receive the cross, predict whether a shot will result, and propose alternative attacking and defensive setups.
In a blinded study, Liverpool FC's analysts and coaching staff judged TacticAI's suggested arrangements indistinguishable from real Premier League corners, and preferred TacticAI's suggestions to the original setup 90 percent of the time. The work followed an earlier 2022 Nature paper from the same DeepMind sports analytics group titled Game Plan: What AI can do for Football, and What Football can do for AI, which laid out the research agenda for the Liverpool collaboration.
DeepMind also published Learning Agile Soccer Skills for a Bipedal Robot with Deep Reinforcement Learning in Science Robotics in April 2024. The team trained policies for a 20-joint humanoid robot to play one-versus-one football using deep reinforcement learning in simulation, then transferred them zero-shot to physical hardware. Compared to a scripted baseline, the trained robots walked 181 percent faster, turned 302 percent faster, and kicked 34 percent faster. They also learned tactical behaviours such as anticipating shots and recovering from falls in 63 percent less time.
The Robocup federation has run a humanoid football world cup since 1997 with the long-term goal of beating a human team by 2050. Pose estimation research from MIT, Carnegie Mellon, and several Chinese universities now feeds back into all major commercial tracking products. Multi-person pose estimation and graph neural networks for player interaction are the two areas of academic AI research with the deepest commercial overlap into sports analytics.
The NBA has been the most aggressive league in pushing AI into the fan-facing app. NBA Commissioner Adam Silver demonstrated NB-AI, a chatbot trained on NBA data and built on Microsoft's Azure OpenAI Service, at the 2024 NBA All-Star Tech Summit. The demo, voiced through a virtual Victor Wembanyama, showed live translation, automated stat overlays, and the ability to redraw a play in cinematic style. The 2024-25 NBA App release added a personalised Following tab and AI-assisted multiview streams.
WSC Sports' generative voice features have been used by NBA partners to produce multilingual commentary tracks for international markets. Football clubs including Manchester City, Bayern Munich, and Liverpool have launched AI-powered chatbots that answer fan questions about scores, ticket availability, and merchandise. The NFL Fantasy app uses AI projections for player performance and waiver wire recommendations.
Manchester City Football Group entered a research partnership with Google Research in 2019 to launch the Google Research Football competition on Kaggle, a 3D football simulation where machine learning agents controlled a single player on the pitch. City Football Group also partnered with SAP as official cloud software provider, and in 2025 named Publicis Sapient as its digital transformation partner for AI-driven analytics on club operations.
Multiplayer esports titles use AI for two main moderation tasks: voice and text toxicity detection, and cheat detection.
Riot Games updated the Valorant privacy notice in April 2021 to allow recording of voice communications, started training a language model on player audio in 2022, and began enforcement actions based on AI-flagged voice incidents in 2023. The text-side equivalent, introduced in Valorant patch 5.10, automatically mutes players who send flagged messages and reportedly resulted in over 20 times more enforcement actions compared with manual moderation.
For cheat detection, FACEIT runs an internal system called Minerva that monitors about 3 million matches per month and 200,000 hours of voice chat. Valve has described an AI-driven anti-cheat for Counter-Strike that classifies suspicious aim behaviour using features such as crosshair placement before and after each shot. Tencent's Anti-Cheat Expert framework applies behavioural fingerprinting to several popular mobile titles. Academic work on transformer-based cheat detection, including the 2025 AntiCheatPT paper, has shown that sequence models trained on input traces can identify aimbots and wallhacks with high precision.
Wimbledon eliminated its 300 human line judges in 2025, replacing them with the Hawk-Eye Live electronic line-calling system across all 18 courts. The change ended a 147-year tradition. About 80 former line judges were kept on as on-court assistants but no longer adjudicated calls. The first weeks of the 2025 tournament drew mixed feedback. World number one Jannik Sinner praised the system's precision, but Chinese player Yuan Yue and others complained that the audio out call was too quiet in noisy moments. The French Open is the only Grand Slam that still uses human line judges, citing the ball mark left on clay.
Wearable and tracking data sit in a regulatory grey zone. The 2022 MLB collective bargaining agreement made it illegal for any club to sell or license confidential medical or biometric data, and the NBA's CBA bars use of wearable data in contract or trade negotiations, with fines of up to 250,000 dollars per violation. At the college level, NCAA athletes have far weaker protections. There is no federal biometric privacy law in the United States, and state laws such as Illinois's BIPA were not written with athletes in mind. Several academic papers have argued that wearable data sharing arrangements between universities and vendors create power imbalances similar to those that the NIL reforms were designed to address.
Generative AI has produced a new category of athlete-targeted harms. During the 2024 Paris Olympics, multiple US Olympians were targeted with AI-generated explicit imagery, and at least one White House social media account posted an AI-generated video showing US hockey player Brady Tkachuk mocking Canada after a gold-medal win. NFL Hall of Fame coach Jimmy Johnson publicly condemned an AI-generated deepfake video circulating under his name. AI voice cloning has also fuelled phone scams targeting athletes' family members and fan-facing romance scams. YouTube opened its AI likeness detection tool to athletes, creators, and musicians in 2025.
Sportsbook risk models trade on the same machine learning that powers price discovery. Critics argue that the same models that flag a suspicious betting pattern can also identify high-value problem gamblers and price them more aggressively. Regulators in the UK and Massachusetts have begun requiring AI risk model audits as a condition of operating licences. The arrival of AI-generated betting tipster content on social media has prompted leagues including the NFL and NBA to issue guidance on impersonation and unauthorised use of player likenesses.
VAR remains the most controversial AI-adjacent technology in football. Studies including a 2024 Sports Engineering paper documented that VAR reduced clear and obvious errors but introduced new ones related to subjective decisions and time delays. Semi-automated offside has shortened review times but has not eliminated complaints about cold lines and millimetre-sized infractions. The Premier League's 2025 SAOT rollout, supplied by Genius Sports rather than Hawk-Eye, was delayed by several months because of validation problems and technology transition.