We Audited 165 Football Prediction Services: Only 23% Made Money

We tracked 165 football prediction services over eight months. 38 finished in profit — and the most accurate source in the sample was not one of them.

Soccerdino.com called 76.0% of its picks correctly across 870 predictions between January and August 2026 — the highest hit rate in the entire sample. It lost money anyway: ROI −1.94%.

Goal.pl got fewer than half its calls right — a 47.8% win rate across 178 predictions — and finished ahead anyway, one of four profitable services that won less than half their picks. ROI: +2.01%

Same eight months, same market, same flat-stake math applied to every row. The best accuracy score in the dataset belongs to a service that finished in the red; one of the worst belongs to a service that finished in the black.

That pairing isn’t a quirk of two rows picked to make a point. Tips.GG tracked 165 football prediction services with enough volume to qualify between 1 January and 31 August 2026. Thirty-eight finished the period with a positive return on investment.

That’s 23.0%.

More than three quarters of the services a bettor could have followed in the first eight months of 2026 — websites, tipster operations, YouTube and Twitter accounts — lost money for anyone backing every pick. And how often a service was right told you little about which side of that line it landed on.

Nobody Was Grading This

Hundreds of thousands of bettors follow prediction services every matchday, usually picking whoever posted the loudest winning streak that week. What none of them can easily check is the number that decides whether following along was worth it: the return across a season, not the last good weekend. Win rates get advertised. Returns rarely do. This index closes that gap — the tracker logs predictions across the industry, settles each at a recorded price, and lets eight months of results answer it.

How We Measured It

Every prediction logged between 1 January and 31 August 2026 is settled at a flat $10 stake, using the price Tips.GG records for that match. Those prices carry a margin — 8.19% on average across this period, 8.35% at the median — and we keep it in, because a service has to beat it to finish ahead.

The unit we measure is the thing a reader actually follows: a website, or a single channel or account on a platform. A YouTube channel is no less a separate outlet than a website is, so counting all of YouTube as one row would have been like counting “the web” as one row.

Two different thresholds follow from that. A website needs 50 graded predictions to qualify; a channel needs 200. The same number means different things: the median website published 1,126 predictions over the eight months, the median channel 221. Among channels with 50 to 199 predictions, 53% finished in profit; above 200, 26% did — the same rate as websites. That gap is the length of the run, not skill, and at a threshold of 50 it would have put an account with exactly 50 picks at the top of the table. Moving that threshold between 100 and 500 moves the index between 25.7% and 21.0%.

In the table below every qualifying service is named, ours included. A few rows are community-voting feeds — an aggregate of what a platform’s users picked — and they are labelled as such.

The full methodology is published at Tips.GG Accuracy Index: Methodology.

One Publisher, Several Names

Four groups of names were merged into single rows, on evidence of shared author accounts, near-identical picks, or a shared domain. One merge removed the apparent leader of the table: a service showing +26.6% ROI on 61 predictions published under the same author pool as four other names, and once the five were combined, the row finished at a loss.

The only merged row that ended in profit is tipsbet.co.uk — two labels whose picks diverged on two of 1,202 shared matches, combining to +0.47% across 1,637 predictions. Merging also strips out double-counted bets: where the same pick appeared under more than one name in a group, it counts once — 1,694 removed from a dataset of 504,562. One row was dropped entirely: our own Telegram channel republishes picks already counted under tips.gg, so keeping it would have counted the same predictions twice.

Four pairs with confusingly similar names were checked and kept apart: their picks agreed between 56.8% and 84% of the time, short of the 95% we require before treating two labels as one publisher.

The Money Doesn’t Follow the Accuracy

The two services that opened this piece aren’t an outlier pair we went looking for. Across the full table, accuracy and return move almost independently. The clearest case sits just below the accuracy leader: 100percentwins on Twitter called 75.9% of 582 picks correctly — within a tenth of a point of the most accurate service in the study — and returned +0.78%. Two services with effectively the same hit rate, separated by the difference between a rounding error and a loss.

The reason isn’t mysterious once it’s said out loud. What a prediction is worth depends on the price it was made at, not only on whether it landed. Call nothing but heavy favourites and you can post a win rate in the seventies while barely covering your stakes, because the price already assumed those results. Be right less often, at prices that underrated the pick, and a coin-flip record turns a profit. Accuracy is the number that markets well. It isn’t the number that pays.

“Accuracy measures how often you were right. ROI measures whether being right at those particular odds was worth anything. They’re two different things, and our table shows why: the source with the highest accuracy in the sample finished in the red, while one of the profitable sources was right less than half the time.”

— Anton Malyutin, CEO of Tips.GG

A Steep Slope, Not a Plateau

The profitable quarter is not a flat plateau. Five rows out of 165 cleared 10% ROI, led by bets.co.za at +17.40%. The remaining 33 profitable services have a median return of 2.47%, fourteen sit below 2%, and seven cleared the line by less than a single percentage point.

So the honest shape of this dataset isn’t “a quarter of services are good bets.” A handful returned something a bettor would notice, the next thirty-odd finished marginally ahead, everyone else paid for the privilege. Profitability isn’t a category here but a slope — and most services on the right side of it stand close to the edge.

Quantity Isn’t a Strategy

Reach and returns pull in opposite directions. Of the 73 services that published 1,000 or more predictions, four finished in profit — 5%. Among the smallest qualifying services, those with fewer than 100 predictions, half finished ahead, though with only eight rows in that bracket it’s a hint rather than a finding. The middle band is where most of the industry sits: 84 services between 100 and 999 predictions, 30 of them profitable.

None of that proves publishing less makes anyone sharper: a service grading a hundred matches has far more room for a lucky run than one grading ten thousand. But at the top of the volume range the pattern is hard to miss — publishing at industrial scale bought reach, not returns.

The Leaderboard

The top of the table, by ROI:

Three of those eight are individual channels, not publishing operations: one person picking 223 matches is a different kind of record from an editorial desk grading thousands.

The leader’s 95 predictions is a small sample, and a good run at good prices is more fragile over 95 picks than over 900 — worth remembering before crowning a champion. It doesn’t make the result less real: every pick behind it is graded, priced and in the table below.

Where We Stand

Tips.GG’s own predictions are in this table, on the same rules, the same flat stake and the same threshold as every other website, with no separate scorecard for ourselves. Our row: 5,264 predictions, 53.7% accuracy, ROI −3.96%.

That sample deserves a moment: it’s more than 55 times larger than the 95 predictions behind the service at the top. A loss across five thousand graded picks is much harder to dismiss as variance than a profit built on fewer than a hundred — which is why both sit in the same table, sorted by the same column.

What This Index Doesn’t Tell You

Some things this study does not claim. Every pick is settled at the same flat stake: real stakes are unknown to us, so a fixed one is the only fair basis for comparison, not a claim about how anyone bet. The price behind every ROI figure is the single price Tips.GG records for a match, identical for every prediction on that fixture, so no service gains an edge from the price it was given — but not necessarily a price any reader could have taken.

The timestamps in our data aren’t reliable enough to say when a prediction became public. A small share — 2.9% — carry a time later than kick-off; their accuracy is 49.6%, in line with everything else, and removing them leaves the index unchanged. Services below their threshold don’t appear at all, which on the channel side means most of a platform’s smaller accounts. Our merges rest on what the data shows: two brands with the same owner but no overlap in authors, picks or domains would pass through as separate rows.

Using This Data

The full table — all 165 services, with predictions, accuracy and ROI — is published inside this study below, free to cite and reuse with attribution to Tips.GG and a link back to the original research. For methodology questions, comment or the data in another format, write to [email protected].

This is the first of four campaigns on the same index. The next ones put different questions to the same tracking data, and these numbers become the baseline they’re measured against. The interesting part won’t be this edition’s profitable quarter. It’ll be which of them are still in it next time.

The Full Table

All 165 services that cleared their threshold, ranked by ROI. Football match-result predictions, 1 January to 31 August 2026, settled at a flat $10 stake. Accuracy is the share of predictions that won. ROI is total profit as a percentage of total stakes. Rows marked (YouTube) or (Twitter) are individual channels; websites needed 50 predictions to qualify, channels 200.

You can view all TipsGG research here.