Artificial intelligence in sports: uses, athlete data and the EU rules
Artificial intelligence in sports now reaches from the training ground to the broadcast: tracking systems that measure every sprint, video tools that tag a match automatically, offside and line technology, scouting models and personalised fan apps. This guide explains where AI is genuinely used, what data each use needs, what the rules on athlete data and the EU AI Act mean for clubs, federations and sports businesses, and how to start an artificial intelligence sports project that people in the club will actually use.
It is written for club and federation managers, sports technology companies and product owners. It covers sport as performance, officiating, media and business, and leaves aside gambling, which has its own licensing rules.
The short version
- Most AI in sport turns tracking data, wearable data and video into faster, more consistent analysis; very little of it makes decisions on its own.
- At the top level, officiating technology proposes and people decide; tennis line decisions are the clearest exception.
- Athlete data is personal data and often health data, with strict rules on purpose, access and retention.
- Under the EU AI Act, evaluating contracted players can be high-risk, emotion recognition in training is prohibited, and face recognition of crowds is tightly limited.
- Good projects start from one recurring decision and the data that already exists for it.
Where AI is used in sports
| Area | What AI does | Data it needs |
|---|---|---|
| Performance and load | Turns positions and sensor readings into distances, sprints, accelerations and load per player, and flags unusual patterns | Optical tracking or wearable GPS and heart rate, training plans |
| Video analysis | Detects players, ball and events in video, tags passes, shots and set pieces, finds tactical patterns and clips them | Match and training video, event data for training the models |
| Officiating | Tracks limbs and ball to support offside, goal-line and line decisions, and produces the graphics that explain them | Dedicated stadium cameras, ball sensors, calibrated pitch models |
| Scouting and recruitment | Finds players whose profile matches a role, compares leagues, and shortlists for scouts to watch | Event and tracking data across leagues, video, contract and age data |
| Medical and recovery | Highlights players whose load, sleep or recovery figures differ from their own normal, for staff to review | Wearables, questionnaires, medical records, held under health data rules |
| Fans and media | Automatic highlights, personalised feeds and notifications, translated content, chat assistants, ticket demand forecasts | Video, content metadata, app and ticketing data with consent where needed |
| Grassroots and fitness | Coaching apps that analyse technique from a phone camera with pose estimation | Phone video, user profiles |
Officiating: AI proposes, people decide
The best documented example is football's semi-automated offside technology. FIFA's description of the system used at the 2022 World Cup (read 11 November 2026) explains that 12 dedicated tracking cameras under the stadium roof follow the ball and up to 29 data points of each player 50 times per second, while a sensor in the ball sends data to the video operation room 500 times per second. Applying artificial intelligence to the combined limb and ball data, the system produces an automated offside alert, and the video match officials validate the proposed decision by checking the automatically selected kick point and offside line before the referee is informed.
Tennis has gone further on line decisions. The Associated Press reported on 9 October 2024 (read 11 November 2026) that the All England Club would use technology to give the out and fault decisions at Wimbledon from 2025, replacing human line judges. The common thread is narrow, measurable questions with very precise sensors, not general judgement. Decisions that need interpretation, such as fouls or intent, stay with people.
Wearables and tracking: the rules of the game
Sports law shapes what can be used in competition. In football, the IFAB Laws of the Game, Law 4 (read 11 November 2026), allow wearable technology as part of electronic performance and tracking systems, EPTS, in official competitions, provided the competition organiser ensures it is not dangerous and meets the requirements of the FIFA Quality Programme for EPTS. Where the organiser provides EPTS, it is responsible for the data sent to the technical area during matches being reliable and accurate. Other sports and leagues have their own equipment rules, and a product meant for match use has to be checked against each competition's rules before anything is built.
For clubs, the practical points are interoperability and ownership. Tracking data often comes from a league's official provider, wearables from another vendor and video from a third. Value appears when they are joined on one timeline per player, and that depends on what the data licences allow the club to store, combine and keep.
Athlete data is personal data, and often health data
Performance data relates to identifiable people, so the GDPR applies. Heart rate, injuries, recovery and fitness assessments are usually data concerning health, and face or body measurements used to identify someone are biometric data; Article 9 prohibits processing these special categories unless one of its listed conditions applies. For a club and its contracted players, consent is a weak basis: recital 43 explains that consent is not a valid ground where there is a clear imbalance between the person and the controller, and a player whose contract depends on the club is in that position.
- Purpose first. Data collected for training load is not automatically available for contract negotiations.
- Access by role. Medical data seen by medical staff, aggregated load figures by coaches, and nothing personal in sponsor or media products without a separate basis.
- Retention. A rule for what happens to a player's data when they leave, and a stricter one for academy players who are children.
- Vendors. Every wearable, tracking and video provider that processes the data is a processor and needs a processing agreement.
The EU AI Act and sport
The EU AI Act, Regulation (EU) 2024/1689, does not mention sport by name, but several of its categories reach it. It has applied in general since 2 August 2026.
- Emotion recognition at work. Article 5(1)(f) prohibits AI systems that infer emotions of people in the workplace, except for medical or safety reasons. For professional athletes the training ground is a workplace, so a tool that reads players' emotions from faces or voice for coaching purposes is off limits. This prohibition has applied since 2 February 2025.
- Evaluating employed players. Annex III, point 4, lists as high-risk AI used for recruitment and selection, for decisions on promotion or termination of work relationships, and for monitoring and evaluating the performance and behaviour of people in those relationships. A model that ranks a club's own contracted players for contract or selection decisions can fall here; obligations for Annex III systems apply from 2 December 2027.
- Face recognition in stadiums. Remote biometric identification systems are high-risk under Annex III, point 1, and Article 5(1)(h) prohibits real-time remote biometric identification in publicly accessible spaces for law enforcement, apart from narrow, authorised exceptions.
- Generated media. Under Article 50(4), deployers of AI that generates or manipulates images, audio or video constituting a deep fake must disclose it, which matters for synthetic voices, generated commentary and edited highlight clips. For generative systems already on the market, the marking duty in Article 50(2) applies from 2 December 2026.
Most performance and video analysis tools that support staff, without deciding about employment, are outside the high-risk list. Keeping it that way is a design choice: the tool informs, and the coach, analyst or sporting director decides and records why.
The data comes before the model
Most failed AI projects in sport fail on data, not algorithms. Before any model is built, check four things:
- Volume and history. Enough matches, sessions or seasons for the question. A recruitment model needs data on many players across leagues; a load model needs months of one squad's sessions.
- Consistency. The same definitions over time, since an event provider that changed how it codes a pressure or a sprint breaks the comparison.
- Rights. Whether licences from leagues and providers such as Stats Perform, Hudl StatsBomb or Second Spectrum allow the club to store, combine and model the data, and for how long.
- Labels. For video models, footage tagged by people who know the sport, which is often the slowest part.
Many questions do not need machine learning at all. Clear descriptive figures and good dashboards answer a large share of what coaches ask, as our guide to the types of analytics explains. Computer vision techniques behind video analysis are covered in our guide to computer vision in retail, which describes the same building blocks in another setting.
How to start an AI project in a club or sports business
- Pick one recurring decision. Training load per player, opponent set-piece preparation, a recruitment shortlist, which highlights to push to which fans.
- Write down how it is made today. Who decides, with what information, how long it takes, and how you would know a better decision.
- Check the data and the rules. What exists, who owns it, whether it is health data, and which AI Act category the use falls into.
- Build a first version small. A dashboard, an alert or a ranked list, used by the people who make the decision, for a defined period.
- Measure against the baseline. Time saved, consistency, and outcomes where they can be measured honestly, then decide whether to extend it.
What we build for sport
We build the software around the models as well as the models: athlete and coach apps, data platforms that join tracking, wearable and video data per player, dashboards for staff, video pipelines that clip and tag, and fan apps with personalised content. Data stays in an EU region in the client's own account, and the code belongs to the client. See AI consulting for choosing and scoping a first use case, fitness app development for training and coaching apps, data analytics consulting and dashboard development for performance reporting, and sports scheduling software for fixtures, venues and officials.
Questions people ask
How is artificial intelligence used in sports?
Mainly in five areas: performance analysis from tracking and wearable data, video analysis that tags events and patterns automatically, officiating aids such as offside and line technology, scouting and recruitment models, and fan and media products such as automatic highlights and personalised content. Clubs and federations also use it behind the scenes for operations, from crowd flow to scheduling.
Does AI replace referees?
Not in the systems in use at the top level. FIFA describes its semi-automated offside technology as producing an automated offside alert that video match officials validate by checking the selected kick point and offside line before the referee is informed. Tennis is the exception for line decisions: Wimbledon announced in October 2024 that electronic line calling would make the out and fault decisions from 2025.
Is athlete performance data personal data?
Yes, whenever it relates to an identifiable person, and data on heart rate, injuries, recovery or fitness is usually data concerning health, a special category under Article 9 of the GDPR. Its processing needs a legal basis and one of the Article 9 conditions, a clear purpose, limited access and a retention period. Consent from a contracted player is often a weak basis because of the imbalance between club and player.
Does the EU AI Act apply to sports clubs?
It can. A club that uses AI to evaluate its own contracted players for decisions on selection, contracts or termination may fall under the high-risk employment category in Annex III, point 4, whose obligations apply from 2 December 2027. Emotion recognition in the workplace, which for a professional athlete includes training, has been prohibited since 2 February 2025 except for medical or safety reasons. Face recognition of spectators is high-risk at least, and real-time use by police in public spaces is prohibited apart from narrow exceptions.
Where should a club start with AI?
With one decision that is made often and matters, such as training load, opponent preparation or recruitment shortlists, and with an honest check of the data available for it. A first version is measured against how the decision is made today, and the coaches or analysts who make it keep the final say.
Sources
- FIFA, Semi-automated offside technology (FIFA World Cup 2022) (read 11 November 2026)
- Associated Press via NBC Sports, Wimbledon replaces line judges with technology, 9 October 2024 (read 11 November 2026)
- IFAB, Laws of the Game, Law 4: The Players' Equipment (read 11 November 2026)
- Regulation (EU) 2016/679, GDPR, Article 9 and recital 43 (read 11 November 2026)
- Regulation (EU) 2024/1689, AI Act, Articles 5, 50 and 113, Annex III (read 11 November 2026)