Home› How the AI works
How the SeerPlay AI works
No black box, no mystique. Here is the model, in plain language, and the public record it stands on.
Step 1 · Rate every team
Each club carries an Elo-style strength rating. Win and the rating rises; lose and it falls. Beat a stronger side and it moves further. Ratings update after every result, so the model's view of a team is always its current form-weighted strength, not its reputation.
Step 2 · Turn ratings into goals
The gap between two ratings (plus home advantage) sets each side's expected goals, and a Poisson goals model turns those into the probability of every scoreline of 1-0, 2-1, 0-0 and so on. Sum the scorelines and you get the three numbers you see on every match page: home win %, draw %, away win %.
Step 3 · Respect the draw
Plain Poisson models under-price draws. Football punishes that. SeerPlay's model is draw-aware: a correction tuned on historical seasons lifts draw probabilities in tight games, which measurably improved the model's hit rate in back-testing. The AI also calls the draw when the draw is close behind the favourite, not only when it is outright top: a tight game is a draw risk, and pretending otherwise flatters the record.
Step 4 · Lock it, publish it, score it
That verdict rule (favourite, unless the draw is within a set margin of it) becomes the AI's published call, so the call is not always the single highest number on the card. It locks before kick-off, is never edited, and every settled match is scored on the accuracy page Including the misses. A prediction model you cannot audit is just an opinion.
Quick answers
How do AI football predictions work?
Most serious models, SeerPlay's included, work in two steps: a rating system estimates how strong each team is right now (updated after every result), and a goals model turns the gap between two teams into probabilities for a home win, draw and away win. SeerPlay publishes the favourite unless the draw is within a set margin of it, in which case it calls the draw. Either way the probability is published with it, not a certainty.
Why does the AI sometimes pick a draw?
Plain goals models systematically under-price draws, so SeerPlay's model is draw-aware: it applies a correction (tuned on historical seasons) that raises draw probabilities in evenly-matched games. When the draw becomes the most likely single outcome, the AI says so.
Can the AI's predictions be trusted?
Judge it on the public record: every pick locks before kick-off, none is ever edited, and every settled match is scored on the accuracy page, hits and misses alike. At the FIFA World Cup 2026 the model called 66% of scored matches correctly.
Can I beat the AI?
That is the whole game. SeerPlay is free: swipe your own call on every match before kick-off, score points when you are right, and compare your record against the AI on the leaderboard.
Free to play · picks lock at kick-off · your record vs the model, in public.
More: the AI's track record · this week's forecasts · the title race