Forebet publishes probability-based predictions at large scale, but it does not publish a complete, season-by-season accuracy record you can independently verify — so its true accuracy is unauditable. You cannot confirm any headline accuracy figure from published, timestamped data. That does not mean the model is bad; it means you have no way to check, and unverifiable accuracy should be treated as unknown.

How accurate are football prediction sites in general?

For match results (home/draw/away), well-built models tend to call the outcome right somewhere around half the time, because draws and upsets are common — the exact figure depends heavily on the leagues and markets covered. Any site quoting a much higher number is usually cherry-picking a market or a short window. The only meaningful accuracy figure is one you can recompute from a complete public record.

How can I check if a prediction site is accurate?

Look for four things: a complete public record of past picks, losing predictions kept rather than deleted, picks timestamped before kickoff, and a method that is explained. If you cannot add up the hits and misses yourself, the accuracy claim is unverifiable. Bawler publishes all four — its main "Most likely" call has landed 75.8% of the time across 2,241 settled matches, recomputable from the public ledger.

Prediction guide · Updated October 2026 · site details as of July 2026

Is Forebet accurate? An honest look

The short answer

Nobody can say — including Forebet's own users. It publishes no complete, timestamped record of its past predictions, so there is nothing to add up and no figure to check. Unverifiable accuracy should be treated as unknown.

This is not a claim that Forebet is inaccurate. It is that no one is in a position to tell.

Forebet is one of the best-known math-based football prediction sites, with probability predictions for most leagues in the world. So the question deserves a straight answer, and the four checks below are how you get one for any site, not just this one.

How Forebet works

Forebet runs a statistical model that assigns probabilities to outcomes — home win, draw, away win, plus markets like both teams to score and over/under — using historical results and team form. That is a sound basic approach, and it is why the site produces a prediction for almost any fixture you can think of. The workings of the model are not documented publicly, so you are trusting a black box.

Why the accuracy cannot be checked

There are no tables showing how often a "60% home win" actually lands 60% of the time, and no permanent, timestamped list of every past pick with its result. Without those, any accuracy figure — quoted by the site or repeated elsewhere — cannot be traced back to raw data, at least not from anything we could find published.

How to judge any prediction site

Context sets the ceiling. For match results, calling the outcome right around half the time is genuinely decent, because draws and upsets are common. A site claiming 85–90% is almost always cherry-picking a market or a short window.
Being right a lot is not the same as being right about how likely something was. A good model's probabilities match reality, so things it calls 30% happen about 30% of the time. That is only checkable with published results.
Sample size is everything. Anyone can look sharp over one weekend. Judge accuracy over hundreds of settled picks, not a rolling 7-day window.
Kept misses are the real test. If losing predictions vanish — or are never recorded — the headline number is meaningless.

The same four checks, applied

Check	Undocumented model site	Bawler
Complete public record	Not published	Every pick, kept
Accuracy shown by confidence level	No	Yes, band by band
Losing predictions kept	Unclear	Kept on the record
You can recompute the figure	No	From public data

Based on publicly available information as of July 2026. Sites change — always check the current state of any tool yourself.

Bawler logs every pick before kickoff, settles it automatically from the full-time result, and keeps it on the record; since 28 May 2026 a snapshot of every published pick has been anchored to Bitcoin each morning, so those snapshots cannot be quietly edited. Because the misses stay on the page, the figures can be checked: across 2,241 settled matches the main Most likely call has landed 75.8% of the time, the Close call 53.6%, and the least-likely Outside shot 29.2%, with the track record showing how each compares with the confidence it was given.

Bottom line

Forebet might well be accurate — but you cannot verify it, and unverifiable accuracy is unknown accuracy. The better question for any site is whether it lets you check. You can add up Bawler's record yourself, confirm nothing has been altered, and read how the model works, all from here.

Track record Verify a pick How the model works Today's free picks Compare the main tools

Model estimates are not guaranteed outcomes. More on the disclosure page.