Transparency

How to spot a fake football tipster: five tests you can run

By FootInsights · Published 6 Sep 2026 · 6 min read

Most advice on how to spot a fake football tipster is a list of vibes: flashy cars, urgent language, too many emoji. That advice is not wrong, but it is useless against a competent operator, who simply removes the vibes. The useful tests are arithmetic. A record either survives a calculation or it doesn’t, and the calculations take a few minutes with numbers the tipster has already published about themselves.

This post sets out five of them, in the order that eliminates the most services fastest.

What “fake” actually means

Three failure modes get lumped together, and separating them matters because different tests catch each one.

• Fabricated. The picks never existed at the stated price or the stated time, or never existed at all.
• Selected. Every pick was real, but the losing ones were deleted, hidden behind a “retired strategy”, or issued to a different channel.
• Real but worthless. Nothing was faked or hidden — the record is simply too short to distinguish from luck, and is being sold as if it weren’t.

The third is the most common by a wide margin and the only one with no villain. It still costs subscribers exactly as much money.

Test 1: is the claimed win rate mathematically reachable?

A win rate means nothing without the odds attached to it. At average odds of 1.90 the market’s own implied probability is 1 ÷ 1.90 = 52.6%, which is also roughly where a bettor breaks even. So a claim only becomes remarkable relative to that baseline.

Take a service advertising 80% winners over its last 50 bets at around 1.90. Model a tipster with no edge whatsoever, whose true win probability is the market’s 52.6%, and ask how often 50 bets produce 40 or more winners:

Now compare that with what a genuinely excellent tipster looks like. A real 56% strike rate at 1.90 is a +6.4% return on investment — a serious, career-grade edge. Over 50 bets that tipster averages 28 winners, with a standard deviation of 3.5, so their honest record lands somewhere between roughly 49% and 63%.

The 80% claim is not “better than excellent”. It sits far outside the range that skill produces, which leaves luck (1 in 18,000, so it happens to someone) or selection. Meanwhile a 60% run over 50 bets — the number that impresses nobody — happens to a coin-flipper one time in five.

The test: convert the claim to ROI at the stated odds. If the implied ROI is above about 10% over a few hundred bets, the record is being selected, not achieved.

Test 2: were the advised odds ever available?

This is the test fakes fail most often, because it checks a fact outside the tipster’s control.

Every record quotes a price. That price determines the entire profit figure — the same set of results at 2.00 instead of 1.90 turns a losing record into a winning one. So the question is whether the advised price existed at the moment of advice, in reasonable size, at a bookmaker a subscriber could actually use.

Two patterns give it away. The first is odds that consistently beat the best price in the market at post time; nobody’s price is systematically the market maximum by accident. The second is a record priced at the peak of the day rather than at the time of the post — trivial to do retrospectively, impossible to do live.

The honest version of this measurement is closing line value: compare the advised price against the price at kick-off, after the market has absorbed all information. A tipster beating the closing line over hundreds of bets has demonstrated something real, whether or not the bets won. One whose advised prices are unreachable has demonstrated the opposite. The gap between what was advised and what a subscriber could actually get is itself a published metric on serious services.

Test 3: who decides when a bet is settled?

Settlement rules are where a record gets quietly improved without deleting anything.

Ask what happens to a postponed or abandoned match, and whether it leaves the record entirely or counts as a void at stake returned. Ask whether a “half-won” Asian handicap is logged at its actual return or rounded to a win. Ask whether the stake was fixed in advance or assigned afterwards — variable staking applied retrospectively lets someone weight their winners at three units and their losers at one, producing a superb profit curve from a mediocre strike rate.

A record without a published, fixed staking plan and a written settlement rule is not a record. It is a narrative with numbers in it.

Test 4: check the denominator, not the numerator

Everything in tests 1 to 3 assumes you can see all the picks. Usually you cannot, and the missing ones are never a random sample.

The mechanisms are well documented: deleted losses, a relaunch that starts the count immediately after a drawdown, several strategies run in parallel with only the best one advertised, or the same pick issued to different channels in opposite directions. All of them produce a genuine, unedited, correctly timestamped stream of winners — which is why screenshots prove nothing. The full arithmetic of how badly this distorts a leaderboard is in survivorship bias in betting.

Two concrete checks. Are the picks numbered sequentially, with no gaps? Does the record start at inception and continue through every bad run, or does it begin at a suspiciously convenient date? A record that can only improve isn’t measuring anything.

Test 5: how long is long enough?

Even an entirely honest record needs a sample. That 50-bet history at 60% is consistent with a coin flip one time in five, so it cannot distinguish skill from noise no matter how sincere the person publishing it.

Roughly speaking, separating a real 6% ROI edge from zero at conventional confidence needs on the order of a thousand settled bets — years of output for most services. This is the uncomfortable part, and it is why an honest tipster’s own claims should sound tentative. Anyone certain about their edge after 50 bets has misunderstood their own data.

What a record that passes all five looks like

The tests share a structure: each one removes a freedom the faker depends on. Take away retrospective pricing, retrospective staking, retrospective deletion and retrospective start dates, and there is nothing left to manipulate.

That is why publishing predictions before kick-off, keeping them append-only with the model version attached, and scoring them with a proper metric rather than a headline win rate is the standard worth demanding. A Brier score grades the probability that was actually stated, so a forecaster who says 55% and is wrong is penalised less than one who says 95% and is wrong — which makes confident nonsense expensive rather than impressive.

FootInsights publishes its full record, losses included, on the track record page, and the method behind the numbers on how it works. Run these five tests against it, and against anyone else asking for your money. The point is not that any single service passes; it is that the tests are cheap, the answers are public, and a service that makes them hard to run has told you something already.