How to choose a football prediction site

Every site in this category claims to be accurate, so the claim itself tells you nothing. What separates a model worth reading from a marketing page is whether its claims can be checked at all. Six questions settle that, and none of them need statistical training.

Last updated
30 October 2026
Reading time
6 min read

1. Is there a graded record, and is it complete?

Start with whether a site publishes the outcome of every prediction it made, including the failures. How accurate it says it is comes second. A record that only holds the selections a site chose to keep is a portfolio, not a record.

What to look for: a page listing individual past predictions with their outcomes, reachable without an account, covering a continuous period instead of a curated selection. What to be wary of: one headline percentage with no list behind it, results that only appear as screenshots or testimonials, or a history that starts conveniently recently.

Most sites fail this check, and it's a cheap one to pass: publishing a complete record costs nothing technically. A site that hasn't done it has usually decided not to.

2. How large is the sample, and over what dates?

A percentage without a denominator isn't a statistic. Forty predictions can produce almost any hit rate through chance alone, and a figure from one hot month tells you about that month, not about the model.

Any published accuracy figure should say how many graded predictions it covers and over what dates. If a site reports by market, each market should have its own count, because a model can be good at one market and poor at another, and a figure blended across markets is where that difference hides.

Two-outcome markets like both teams to score sit structurally higher than three-outcome markets like the match result. A site quoting one number across all of them is publishing a figure shaped by which markets it covers most, whether it means to or not.

3. Is calibration reported at all?

Calibration is whether stated probabilities match reality: of everything a model called at 70%, close to 70% should happen. It is what makes a probability useful instead of decorative, and it's the one thing a site can't fake by picking favourable examples.

Reporting it takes one table: predictions grouped by stated probability, each group's claim set against what actually happened, with a sample size per group. A site that shows this is handing you the tools to audit it. A site that talks about accuracy but never about calibration may never have checked whether its percentages mean anything.

This is the strongest single signal available, because it is easy to verify and rare. Even if a site published nothing else useful, a calibration table would tell you more than a homepage full of figures.

4. Is the method described in checkable terms?

"Advanced AI" and "proprietary algorithm" describe nothing. A site that has built something can say what kind of model it uses, what data goes in, how often it retrains and how probabilities are produced, and none of that gives away anything a competitor could copy.

A real description names model families instead of using adjectives, gives a specific count of input features instead of "millions of data points", and explains how raw model output becomes a stated percentage. That last step is the one most often skipped, and it decides whether a number means anything.

Vagueness isn't always dishonest. Some sites are genuinely thin and have nothing to describe. Either way, it tells you what you are reading.

5. How does the site make money?

A prediction site that gets paid for sending visitors to a particular third-party operator has an interest in which predictions it shows you and where they point. That's no accusation against any one site. It's an ordinary incentive that exists whether or not anyone acts on it, and knowing it's there changes how you read the output.

You can answer this in a minute. Are outbound links tracked affiliate or referral links? Does the site name commercial partners? Is it funded by subscriptions, advertising, referral fees or a mix? A site that states its revenue model plainly has removed one reason for you to wonder.

Subscription funding isn't automatically better; it creates its own pressure to look impressive. But that pressure at least points at the quality of the predictions, not their direction.

6. What does it say about being wrong?

Football is uncertain in a way that caps how much anything can know. A red card in the ninth minute is not a signal a model missed; it is noise the sport produces. Any method built on the game inherits that ceiling.

So the way a site talks about its own limits tells you a lot. Does it name the markets where it does worse? Does it give sample sizes, so a thin figure can't pass for a solid one? Does it explain that a 75% forecast should fail about a quarter of the time, instead of treating that as an anomaly?

The opposite is the clearest warning sign in the category. Language that promises outcomes or certainty describes something the sport doesn't contain, and a site willing to say that about football is telling you how carefully it says everything else.

Where ScorePredicts stands on all six

A framework applied to everyone except yourself would be worth nothing, so here is ours. Every forecast is graded against the final result and never changed afterwards, and the record is public with no account needed: aggregate figures on the accuracy page, individual graded predictions on every league and team page. Headline accuracy is reported separately for each prediction type, each with its own sample count, and the mix of markets behind those predictions is published next to it, so the figure can be taken apart and checked.

Calibration is published for the match-result market as a bin-by-bin table with sample sizes and 95% intervals, next to a comparison with a naive home-win baseline on that market. Where the data can't support a claim, the page says so: calibration for other markets isn't published yet, and for both teams to score it is held back on purpose because the forecasts are too concentrated to report honestly. Every hit rate is shown with the number of graded predictions behind it, so a small sample can be seen for what it is. A calibration bin needs at least 50 fixtures before it is plotted, and fixtures that were never completed are excluded, with the count shown. The same numbers can be downloaded as JSON under a CC BY 4.0 licence.

The method is described part by part: how many models, how they are grouped by competition, how many input features, and which calibration method turns model output into a stated percentage. Revenue comes from subscriptions only: no advertising, no affiliate or referral links, no commercial partnerships with third-party operators, no commission from anyone, and no links sending visitors to any third-party operator. And nothing here is presented as certain, because a probability is not a promise, and a site claiming otherwise would fail its own sixth test.