Are AI football predictions accurate?

A good football model gets the match result right slightly more often than a coin weighted toward the home team. That sounds modest, but it's what the sport allows. The more useful question is whether a site's stated probabilities mean anything.

Last updated
17 October 2026
Reading time
6 min read

How accurate can a football prediction actually be?

Across the major European leagues, the favourite wins a little over half of all matches and about a quarter end in a draw. A model that picked the home team every time would score around 44-47% on match result without knowing anything at all. Closing market prices get the winner right about 55-58% of the time in top divisions. They are the best-informed forecast available for any football match, because they combine the views of a very large number of participants.

Nobody has beaten that number. It's the ceiling. For a well-built statistical model on 1X2, the honest range is the low-to-mid fifties, and a few points better than always picking the home team is a genuinely good result. Markets with two outcomes, like both teams to score and over/under goals, score higher simply because there is one fewer way to be wrong. That's why a single accuracy figure blended across markets tells you almost nothing.

Why 85% accuracy claims are not credible

Prediction sites advertising 85%, 87% or 90% accuracy are common in this category. Against the numbers above, those figures are implausible. Beating the closing market by thirty percentage points would be one of the bigger results in applied statistics, and nobody would announce it on a subscription landing page.

There are a few ordinary explanations, and none of them involves a better model. The figure may come from a hand-picked window when the model happened to run hot. It may count only the selections the site chose to publish and quietly drop the rest. It may blend easy markets with hard ones until the average looks good. Or it may be measured on double chance, where covering two of three outcomes gets you past 70% almost by construction, and then presented without saying so.

The giveaway is rarely the number itself. It's everything missing around it: no sample size, no date range, no breakdown by market, no way to see the individual predictions behind the figure.

Accuracy is the wrong thing to measure

Hit rate treats every prediction as simply right or wrong, which throws away the part that carries the information. A model that says 51% and a model that says 94% have made very different claims, and scoring them the same when the outcome lands hides exactly what you want to know.

Calibration keeps that information. A model is calibrated when its stated probabilities match what happens: of all the matches it called at 70%, close to 70% should go that way. Calibration can be checked, it can't be gamed by cherry-picking, and it is what makes a probability useful instead of decorative.

The standard summary statistic is the Brier score. It squares the distance between the stated probability and the outcome, so a confident wrong answer costs more than a cautious one. It has a handy reference point: saying 50% to everything scores 0.25, so a model that can't beat 0.25 is worse than no model at all.

How to check whether a probability means anything

The test needs one table. Group every published forecast by its stated probability (all the 60-65% calls together, all the 65-70% calls together, and so on), then compare each group's stated probability with how often those outcomes actually happened. If the 70% group lands near 70%, the number means what it says. If it lands at 55%, the model is overconfident and every figure it shows you is inflated.

Two things keep the test honest. Each group needs a stated sample size, because a group of twenty matches can be ten points off through chance alone. And it needs an interval around the observed figure, so a reader can tell a real miss from ordinary noise.

Almost nobody in this category publishes that table. It is the most informative thing a prediction site can show, and leaving it out says more than any headline accuracy figure.

How ScorePredicts measures itself

Every forecast is graded against the final result once the match ends, and the forecast is never changed afterwards. The record is public and needs no account: aggregate figures on the accuracy page, and the graded predictions themselves on every league and team page.

The accuracy page publishes a hit rate for each prediction type with its own sample count, the mix of markets those predictions came from, a comparison with a naive home-win baseline on match result, and the calibration table described above. That table covers the match-result market, all three outcomes, with sample sizes and 95% intervals per group. Where the data doesn't support a claim, the page says so instead of filling the space. Calibration isn't published for other markets yet; for both teams to score it is held back on purpose, because the model's forecasts are currently too concentrated to report honestly. Every hit rate is printed next to the number of graded predictions behind it, so a small sample is visible, and a calibration bin needs at least 50 fixtures before it is plotted. Fixtures that are never completed (abandoned, cancelled or awarded) are excluded instead of being counted as right or wrong, and the page shows how many were excluded.

The same figures are available as a JSON feed under a CC BY 4.0 licence, so anyone can check the numbers independently.

What a good model cannot do

A calibrated model tells you how likely something is. It can't tell you what will happen in a single match. A 75% forecast failing doesn't mean the model is broken: a model whose 75% calls always landed would be badly calibrated, not brilliant.

Football is uncertain in a way that caps how much any method can know. A red card in the ninth minute, a goalkeeping error, a rotated squad before a European tie: these aren't signals a model missed. They are noise the sport produces. The gap between a strong model and a weak one is real but narrow, and it shows up over hundreds of matches, not over any given weekend.

Anyone promising certainty in this category is describing something football does not contain.

Checking any prediction site's claim yourself

Four questions separate a claim you can measure from a marketing one, and none of them need statistical training. How many graded predictions is the figure based on, and over what dates? Is it broken down by market, or blended into one number? Can you see the individual predictions behind it, including the ones that lost? And is there any calibration reporting at all?

A site that answers all four is making a claim you can check. A site that answers none is asking you to take a number on faith, and in a category where the realistic ceiling is around 58%, that is a lot of faith.