How to verify a football tipster's track record

Published 2026-08-18

Football tipping is one of the few businesses where the product — past performance — can be fabricated in five minutes with an image editor. So before you trust anyone’s picks, you need one skill above all: knowing how to verify a football tipster’s track record. This article is the checklist we’d want applied to us. It works on any tipster, any model, any “AI predictions” site — including this one.

Why screenshots and testimonials prove nothing

A screenshot of a winning bet slip is not evidence. Slips are trivial to forge, and even genuine ones tell you nothing about the losing slips that weren’t posted. The same goes for testimonial-heavy sales pages and “results” lists published as plain text that the site owner can edit at will.

The uglier tricks are structural, not graphical:

Backfilled picks. “Winners” published after the matches finished, timestamped to look pre-match.
Deleted losses. A results page that only ever grows on good weekends.
The many-groups play. Run several private channels, send different picks to each, then advertise whichever channel happened to win. Each individual subscriber saw a genuine, unedited stream of picks — and the operator manufactured a “hot” record from pure randomness.

None of these require the tipster to know anything about football. Which is why the burden of proof has to sit with the record itself, not with anyone’s word.

Win rate is the wrong number anyway

Even a genuine win rate is close to meaningless on its own. In our nine-league dataset, simply picking the home team wins about 44% of the time, and picking short-priced favourites wins far more often than that — while losing money. A tipster hitting 60% on odds of 1.30 is performing worse than the prices they’re betting into.

A record you can actually evaluate needs the odds attached to every pick, so you can compute return on investment, and ideally the stated probabilities, so you can score calibration with a proper metric like the Brier score. “Compared to what?” is the question a bare win rate never answers — the market’s closing prices are the benchmark that most tipsters quietly fail.

How to verify a track record: five checks

Published before kick-off, verifiably. Every pick needs a timestamp the tipster cannot rewrite — an independent proofing platform, an archived page, an append-only feed. If history can be edited after results are known, it isn’t a record; it’s marketing.
Complete, losses included. The record must contain every pick ever issued, not a curated “recent winners” reel. One missing losing month invalidates the whole thing.
A sample large enough to mean something. Luck is astonishingly persistent in small samples. A no-skill coin-flip picker betting at even odds hits 60% or better over 30 picks roughly 18% of the time — nearly one in five. Over 300 picks, that probability collapses to about 0.03%. Streaks are even cheaper: any single no-skill tipster goes 8-for-8 only once in 256 tries, but among a thousand of them, the chance that at least one can advertise a perfect 8-0 run is about 98%. The industry contains far more than a thousand — survivorship guarantees that someone, somewhere, is always “on fire”.
Odds recorded at the time of the tip. Without them you can’t compute returns, and you can’t check the tip against the closing price — the single strongest public benchmark a prediction can be measured against.
A stated method. Not the secret sauce — the shape of it. What data goes in, what the model does, what its known blind spots are. A tipster who can’t describe why their picks might work has given you no reason to expect the past to repeat.

A record that passes all five can still belong to someone whose edge has gone. But a record that fails any of them tells you nothing at all — and “nothing” is what most of the industry offers.

Hold us to the same standard

This checklist isn’t a stick to beat other sites with; it’s the standard FootInsights was built against. Every prediction we publish carries its probabilities and model version, is frozen before kick-off, and is settled and displayed — wins and misses alike — on our public track record, scored per league and per market with the Brier score rather than a bare win rate. The methodology page describes exactly what the model does and where it’s weak, promoted clubs and all.

We don’t quote performance numbers in articles, ours or anyone’s, because static text can’t be audited and can’t go stale honestly. The live, append-only page is the claim. Apply the five checks above to it — that’s precisely what it’s for. And whatever record you end up trusting, bet only what you can afford to lose: a verified edge is still a small one, and variance never signs a contract.