For a club scout, the question is almost limitless: who can we buy? For a national team scout, the question is stranger and much narrower: who is ours?
Finn McCallum, the player and coach I have been interviewing throughout this series, corrected me on that point the first time I asked how scouting works at the international level. At club level, money opens the map, but at World Cup level, your passport closes it. Norway cannot just decide it wants a French midfielder, Cape Verde cannot bid for a Brazilian winger, and the United States cannot buy a striker from Spain because he fits the way they want to play. Before you can evaluate a player, you have to know whether the player is eligible to wear the shirt.
This is the third post in our series on AI and machine learning at the 2026 World Cup. The first post laid out the two ways this technology is used in the tournament, on the pitch and off it, and the second asked whether a model can predict who wins. This post covers the on the pitch side, and it comes down to how a team decides who can even be considered before anyone asks whether they are good.
That’s the piece that shapes how AI and machine learning get used in World Cup recruiting. Most people jump straight to “who is good?” The harder, more useful question is “who is good, eligible, overlooked, and aligned with the way we want to play?” That question is much harder to answer. This is where machine learning earns its place in the process.
FIFA Runs the Tournament, but the Teams Do the Recruiting
FIFA runs the tournament and does not pick players. FIFA sets eligibility rules, builds technology infrastructure, and, in 2026, is making more AI enabled analysis available to all 48 teams through tools like Football AI Pro. Recruiting decisions are the federation’s, along with the coaching staff’s, not FIFA’s.
There is also the money. FIFA is spending 200 million dollars building elite youth academies around the world through its Talent Development Scheme, with a target of 75 academies by 2027. No squad gets picked by FIFA directly, but the pool that future squads are picked from is being shaped quietly, right now, through that investment.
“Recruiting” at the World Cup is not like recruiting in college football or signing players in the Premier League. A national team manager is working inside a fixed player pool, one defined by nationality, family history, association eligibility, and whether a player has already committed to another country.
And the pool is less fixed than it used to be. In 2020, FIFA changed its eligibility rules so that a player can switch national teams if he played no more than three matches for his first country, none at a World Cup or continental final, all before turning 21, and at least three years have passed since his last appearance. Players at this World Cup have already used it, including DR Congo defender Aaron Wan-Bissaka, who switched from England, and Australia midfielder Cristian Volpato, who switched from Italy. Eligibility changes over time, and somebody has to notice when it does. A player who looked locked away five years ago might be available now. Keeping track of that is the kind of tedious, rules based monitoring that software handles better than a scout’s memory.
So the first job AI does here is not glamorous. Nobody’s asking a model to spot the next Messi here. This is database work: identity resolution, eligibility mapping, finding the right Roberto Lopes before he disappears into the noise of global football.
Lopes is the perfect story because it sounds fake until you realize it is exactly how modern scouting works now. The Cape Verde defender, born in Ireland to a Cape Verdean father, was reportedly contacted by Cape Verde through LinkedIn after officials struggled to reach him through normal channels. He initially thought the message was spam, and years later, he was part of Cape Verde’s World Cup story.
There is no AI in that story, just something almost comically human: a message, a missed translation, a second follow up, a player who almost never answers. But it points toward the AI version of the same idea. Somewhere in the world, there is a player whose club career does not scream “World Cup,” whose domestic league does not attract global attention, whose name does not show up in the highlight economy, but whose passport, parentage, movement data, and position specific profile make him exactly right for a national team.
The old version of that discovery was a scout with a notebook and a network. The new version adds a video platform, tracking data, event data, social signals, eligibility databases, and an AI assistant that can narrow the search before a human ever gets on a plane.
The Best Scouting Question Is Not “Who Looks Good?”
One of Finn’s best points was that the “eye test” still counts for something, but it is not the whole story. Soccer gets interesting for machine learning right around that gap. A casual fan watches the ball, while a good scout watches everything around it: the run that pulls a defender away, the body shape before a pass arrives, the midfielder checking his shoulder before receiving under pressure, and the forward sprinting at a center back after losing possession, not because it will show up in the box score, but because panic is a tactic.
Finn described scanning as building “a mental map of the field” before the ball arrives. Modern computer vision and tracking data are trying to quantify exactly that. Did the midfielder check his shoulder before receiving? Was his first touch open to the field or closed toward his own goal? Some of this comes down to a single half second decision. Did he move the ball into danger, or did he preserve a pretty pass completion number by recycling it backward? Did the winger take on his defender, or did he just avoid losing the ball? Did the striker help the team win possession high up the field, or did he only exist when the ball came to him?
Soccer is continuous, with no downs, no set possessions, and no tidy pause after every play, so most of the value happens in motion, often away from the ball. AI has taken on a bigger role in scouting than a lot of fans might expect, and that continuity is a big part of why.
The first post in this series mentioned the bizarre fact that even elite players spend only a few minutes of a 90 minute match touching the ball. The rest of the game is positioning, timing, pressure, deception, anticipation, and restraint. AI earns its keep in that leftover space, not because it understands soccer better than people do, but because it can help people see the parts of soccer that are too constant, too subtle, or too far away from the ball for a normal viewer to track.
The Kante Problem
N’Golo Kante is the cleanest example of why data can change who gets noticed. Before Kante became Kante, he was not the obvious superstar archetype. He was not huge or flashy, and he did not play like the player a casual fan would build in a video game. But his defensive output was absurd: tackles, interceptions, ground covered, problems erased before they became problems.
In 2015, Leicester City signed him from Caen. In 2016, Leicester won the Premier League at 5000 to 1 odds. In 2018, Kante helped France win the World Cup. Really, it’s a story about data helping scouts trust what effectiveness looks like when it doesn’t look glamorous, not spreadsheets replacing anyone.
The danger is that fans often talk about analytics like it is a war between numbers and eyes, but that is the wrong frame. The best version of modern scouting uses numbers to pressure test judgment, not replace it.
Finn gave the simplest version of the problem: a player can complete 90 percent of his passes and still not help his team progress. If every pass is safe and backward, the stat looks clean while the player does very little damage. Better models don’t stop at completion rate for that reason. They ask what kind of pass it was. Did it break a line? Did it move the ball into a more dangerous area? Did it change the defense? Did it create a future shot even if it was not the assist?
Good scouting sounds less like “this player has good stats” and more like “this player does a specific useful thing, repeatedly, against real pressure, in a way that fits what we need.”
Why Smaller Countries May Benefit Most
The romantic version of AI in soccer is that it levels the playing field, but the realistic version is messier.
FIFA’s Football AI Pro, which the first post in this series covered in more detail, is meant to give all 48 World Cup teams access to advanced analysis. WIRED reported that FIFA is tracking around 150 million data points per match at this World Cup, and that the new tool gives teams a chat style interface for querying match data and reviewing 3D reconstructions. Tom’s Guide reported that FIFA AI Pro has been distributed to analysts from all 48 teams. A small federation can now access kinds of analysis that once required a much larger budget and staff, though access alone doesn’t guarantee they’ll use it as well.
Finn put it bluntly in our conversation: the tool can be the same, but richer countries may still have better people using it. France, England, Spain, Brazil, and Germany can afford analysts, data scientists, consultants, scouts, and software engineers, while smaller federations may get the same platform without the same human layer around it. This, more than anything else, is the real AI divide in soccer. Increasingly everyone has data. What separates teams is who can turn it into one clear decision a manager will use.
Underneath that sits a quieter problem: the data itself is unevenly distributed. Tracking data and detailed event data are thickest in the big European leagues and thinnest in exactly the places where overlooked eligible players tend to be: lower profile domestic leagues, second divisions, youth academies outside the traditional pipeline. AI helps most where the data is richest, but national teams need it most where the data is poorest.
For smaller countries, the upside is still enormous. If your domestic player pool is small, you cannot afford to miss eligible talent abroad. If your federation has a large diaspora, then scouting becomes partly a geography problem and partly an identity problem. Who has a parent or grandparent from your country? Who played youth football somewhere else but is not cap tied? Who is good enough to help now? Who is good enough to build around four years from now?
This is not a niche strategy anymore. Almost a quarter of the 1,248 players at this World Cup represent a country other than the one they were born in, up from under 9 percent in 2006. Morocco has taken the idea furthest, with nineteen of their 26 players born outside Morocco, and during this tournament they became the first national team ever to field a starting eleven made up entirely of players born abroad.
The mechanics behind that are exactly what this post has been describing. One federation executive told Front Office Sports that when they learn about a player with a relevant parent or grandparent, they put him “into our system” and track and assess him from there through video and live scouting. Database work, basically, described plainly by someone who does it for a living.
Before the tournament even starts, this is where AI can help. It can help federations map the diaspora, flag eligible players in lower profile leagues, and compare a center back in Ireland, a midfielder in France, and a fullback in the Netherlands by role rather than reputation, the same kind of feature engineering that turns a pile of raw attributes into something a scout can compare. In model terms, that means describing every player with the same set of measurable behaviors, things like pressure regains, progressive carries, and aerial duels won, so a defender in Dublin and a defender in Rotterdam can sit in the same table and be compared line by line. It can tell a scout, “This player is not famous, but he fits the thing you are missing.”
The data gap is starting to close from the bottom, too. An app called aiScout lets any player record himself doing standardized drills on a phone, and computer vision scores the footage so club and federation scouts can filter the results. During its beta, players from 125 countries used it, and 135 of them earned trials or contracts with professional clubs and, in some cases, national teams. The purest version of the player everyone else missed is a kid with a phone in a country with no tracking cameras at all.
Then a human still has to send the message. Sometimes that message is on LinkedIn.
The Human Part Is Still the Hard Part
The more I learn about AI in soccer, the more it looks like machines are changing what humans can notice, not taking anything over. A model can tell you that a teenage midfielder scans constantly, receives on the half turn, advances the ball under pressure, and wins it back quickly after losing it, that a defender’s progressive passing is unusual for his age, or that a forward who does not score much still creates chaos by pressing.
What it cannot do is decide whether that player can handle a World Cup locker room, whether a dual national player feels enough connection to choose one country over another, or whether a player who did not help during qualification will feel comfortable taking the place of someone who did.
That was one of the most human moments in the interview. Finn brought up Odsonne Edouard, who turned down a spot on Haiti’s World Cup squad this year, saying he didn’t feel legitimate playing after not helping the team qualify. A question of identity, loyalty, and dressing room chemistry, not something a dataset was ever going to answer.
World Cup recruiting lives in that tension. The model can find the player, but the manager has to choose him, the player has to choose the country, the teammates have to accept him, and the fans have to believe he belongs. If you ignore the data you miss players, and if you ignore the human part you build a spreadsheet instead of a team.
What AI Really Changes
So, how is AI changing how World Cup teams recruit? Start with the search itself, and how the evidence gets built from there. Instead of relying only on who plays for Barcelona, Real Madrid, Bayern Munich, or Manchester City, teams can search by role, movement, pressure, eligibility, development curve, and tactical fit. Instead of asking whether a player “looked good,” they can ask whether he repeatedly did the things that predict future usefulness: scanning, line breaking passes, pressure regains, off ball runs, defensive coverage, shot creation before the assist.
Timing shifts too. Young players are hard for models because there is less data on them, which makes them risky but also interesting. The best scouts and analysts aren’t waiting until everyone agrees. They’re trying to find the player one tournament before consensus catches up.
Fairness changes. Or at least it gets more complicated. If every team gets a version of the same AI tool, that sounds democratic, but soccer has never been decided by tools alone. It is decided by who knows what to ask, who can interpret the answer, and who has the courage to act on it.
I keep coming back to the LinkedIn story because it’s so undramatic. It’s funny because it feels so low tech: a national team trying to reach a player through the same platform where people endorse each other for Excel. But it’s also the whole future of scouting in miniature.
Somewhere, in a database or a video clip or a social graph or a forgotten message request, there is a player who belongs to a country he has not played for yet. AI can help find him faster, explain why he is worth the call, and help a small federation see what a bigger federation missed. But somebody still has to believe enough to reach out, and somebody still has to answer.
This post is part of our ML in the Wild series on AI at the 2026 World Cup. Next up: how teams are using computer vision and biometric data to keep players on the pitch across a tournament that has now grown to 104 matches.
