Short answer. As of 18 November 2026, no football prediction site we measure beats the betting market's closing line. We read the public predictions of 18 sites every day before kick-off. 9 of them have been read for at least 14 days and have enough settled predictions for a verdict:
- beating the closing line with statistical confidence: 0;
- measurably below the closing line: 8;
- not distinguishable from the market: 1.
Verdicts that rest on under 14 days of matches, shown in a separate table and not counted above: 4. Still being measured: 5.
Hit rates look better than the verdicts. Among the 9 counted sites with enough priced picks they ran from 46.5% to 66.1%, and 5 of the 9 had a hit rate above the average implied probability of their odds. That is not the same as making money: flat-stake returns shown, 9; positive, 3. None of these sites beats the closing line with statistical confidence. And "below the closing line" does not mean worse than chance: see "What the numbers do not say" further down.
This is a dated study. The figures are frozen as of 18 November 2026 (UTC); the figures of the day are on the review hub and on each site's review page. TipsAudit has no affiliation with any site in the table or with any bookmaker. We publish our own record too; our value flags are measured against the same margin-free closing price. 18+.
The table
Sites read for at least 14 days that have a verdict, in the order of the lower end of the closing line value interval. The order is not a ranking of skill (see "What the numbers do not say").
| Site | Measured since | Settled picks | Hit rate vs implied probability | Closing line value (95% interval) | Flat-stake return | Verdict |
|---|---|---|---|---|---|---|
| Zulubet | 18 Sep 2026 | 416 | 52.7% vs 53.5% (n = 387) | +0.05% (-0.66% to +0.76%), n = 359 | -3.9% (n = 387) | Not distinguishable from the market |
| Forebet | 14 Oct 2026 | 1,404 | 48.5% vs 48.3% (n = 1,366) | -1.35% (-1.75% to -0.95%), n = 1,284 | -2.2% (n = 1,366) | Below the closing line |
| Vitibet | 4 Nov 2026 | 1,339 | 66.1% vs 64.6% (n = 955) | -1.62% (-2.08% to -1.15%), n = 924 | +1.9% (n = 955) | Below the closing line |
| PredictZ | 14 Oct 2026 | 1,644 | 46.5% vs 46.1% (n = 1,518) | -1.82% (-2.21% to -1.44%), n = 1,471 | -0.2% (n = 1,518) | Below the closing line |
| FreeSuperTips | 14 Oct 2026 | 498 | 52.5% vs 50.5% (n = 442) | -1.63% (-2.35% to -0.91%), n = 436 | +5.4% (n = 442) | Below the closing line |
| ProSoccer.gr | 5 Oct 2026 | 2,272 | 49.2% vs 49.3% (n = 2,199) | -2.02% (-2.39% to -1.66%), n = 2,143 | -1.0% (n = 2,199) | Below the closing line |
| Statarea | 15 Sep 2026 | 2,408 | 53.2% vs 52.4% (n = 2,194) | -2.09% (-2.39% to -1.80%), n = 2,109 | +1.5% (n = 2,194) | Below the closing line |
| WinDrawWin | 14 Oct 2026 | 2,554 | 50.2% vs 51.5% (n = 2,269) | -2.16% (-2.48% to -1.84%), n = 2,199 | -3.4% (n = 2,269) | Below the closing line |
| FootballPredictions.net | 14 Oct 2026 | 2,306 | 56.3% vs 58.8% (n = 1,405) | -2.98% (-3.27% to -2.69%), n = 1,395 | -3.8% (n = 1,405) | Below the closing line |
How to read it. Settled picks leaves out correct score. Hit rate vs implied probability is how often the picks won, next to the average implied probability of their odds (the average of 1 ÷ odds): below it, the picks won less often than their prices suggested. It is not an exact break-even point when the odds are mixed, so a site can sit above it and still show a negative return. The flat-stake return (one unit on every pick at the best odds we recorded, shown from 100 priced picks) is the direct measure. Closing line value (CLV) is the average gap between the odds available when we read a pick and the fair closing price; a negative number means the pick was, on average, a worse price than the market's final view. The verdict comes only from the CLV interval. A positive return next to a closing line value that is not above zero most likely reflects luck over this sample, not an edge.
Verdicts on under 14 days of matches
| Site | Measured since | Settled picks | Hit rate vs implied probability | Closing line value (95% interval) | Flat-stake return | Verdict |
|---|---|---|---|---|---|---|
| SoccerVital † | 15 Nov 2026 | 198 | 59.9% vs 56.9% (n = 137) | -1.71% (-2.92% to -0.51%), n = 135 | +5.7% (n = 137) | Below the closing line |
| MyBets Today | 14 Nov 2026 | 606 | 54.2% vs 54.1% (n = 426) | -2.71% (-3.35% to -2.06%), n = 424 | -0.2% (n = 426) | Below the closing line |
| 4BetPredict | 14 Nov 2026 | 250 | 77.9% vs 70.7% (n = 104) | -2.44% (-3.39% to -1.50%), n = 104 | +8.9% (n = 104) | Below the closing line |
| Dicas Bet | 14 Nov 2026 | 94 | 78.3% vs 70.6% (n = 46) | -2.67% (-4.15% to -1.27%), n = 46 | — | Below the closing line |
These sites already have the number of settled picks our review pages ask for, so their pages show a verdict. Most of those picks come from a few match days, so we keep them out of the counts above. Their verdicts can still change.
Still being measured
| Site | Measured since | Picks read | On matches we cover | Status |
|---|---|---|---|---|
| Adibet | 14 Nov 2026 | 30 | 29 | Measuring: 29 of 30 settled |
| Il Veggente | 14 Nov 2026 | 34 | 28 | Measuring: 28 of 30 settled |
| Rue des Joueurs † | 15 Nov 2026 | 83 | 77 | Measuring: 25 of 30 settled |
| Cassandra Pronostici | 15 Nov 2026 | 17 | 16 | Measuring: 16 of 30 settled |
| Bookenemy | 15 Nov 2026 | 11 | 2 | Measuring: 2 of 30 settled |
A verdict needs 30 settled predictions outside correct score and 30 of them with a measurable closing price. In this study we give only the counts for these sites; their review pages show the running figures with the sample size.
† Measured with aggregate figures only (see the limits below). Vitibet publishes the same predictions as Vitisport (46 of 46 were identical when we compared them on 3 November 2026).
One prediction API, measured the same way
| Site | Measured since | Settled picks | Hit rate vs implied probability | Closing line value (95% interval) | Flat-stake return | Verdict |
|---|---|---|---|---|---|---|
| API-Football predictions | 15 Oct 2026 | 3,069 | 70.8% vs 72.1% (n = 2,773) | -2.51% (-2.72% to -2.30%), n = 2,772 | -2.3% (n = 2,773) | Below the closing line |
API-Football is a data service, not a site with public prediction pages: we read its prediction once per match through its API, so it is not counted in the totals above. Disclosure: it is also one of our paid data providers.
Download. The full table as CSV, one row per site and, from 30 settled predictions, per market (match result, double chance, over/under, both teams to score, correct score). Free to reuse under CC BY 4.0.
Method
What we read, and when. The public, free prediction pages of each site, once a day, before kick-off: at most three pages per site per day, with a declared bot where the site accepts it and an ordinary browser through a residential proxy where it does not. We never solve bot checks. A prediction we read after the match's real kick-off time does not count, and the first version we read stays: if a site changes a pick later, we do not follow it.
Linking a pick to a match. Each prediction is matched to a fixture we cover by both team names, within the day's window: exact names first, then known aliases, then close name matches that fit only one fixture. One site gives fixture IDs, which we use. A pick that cannot be matched on both teams is left out, not guessed.
Settlement. On the result after 90 minutes plus stoppage time. Extra time and penalties do not count.
Hit rate against the odds. We record the best bookmaker odds available on the same selection when we read the pick (betting exchanges excluded). Next to each hit rate we give the average implied probability of those odds: the average of 1 ÷ odds over the picks that won or lost. A site that is right 70% of the time on picks priced around 1.40 has won less often than the 71.4% those prices imply, so a hit rate means little without the odds (why a 67% win rate loses money).
Why the closing line. Results are noisy: a few hundred picks can look good or bad by luck. The closing price of a sharp bookmaker, with its margin removed, is the market's best estimate of each outcome, and consistently getting better odds than that price is the earliest reliable sign of skill (closing line value explained). We compare each pick's odds with Pinnacle's closing price on the same selection, margin removed.
Confidence intervals. Each CLV comes with a 95% interval (bootstrap; a normal interval above 2,000 picks) and its sample size. "Below the closing line" means the whole interval is below zero. "Beats the closing line" would mean the whole interval is above zero. "Not distinguishable from the market" means the interval contains zero: on this sample we cannot tell the site's picks from the market's own price. It is not a finding that the site matches the market.
Limits
- Short history. The first site has been read since 15 September 2026, the newest since 15 November 2026; the tables give each site's start date. A verdict enters the counts of this study only after 14 days of reading. Not counted yet (4): SoccerVital (since 15 Nov 2026, 198 settled), MyBets Today (since 14 Nov 2026, 606 settled), 4BetPredict (since 14 Nov 2026, 250 settled) and Dicas Bet (since 14 Nov 2026, 94 settled). Every history here is short for football results.
- Coverage. Of 39,076 predictions read, 25,458 (65.1%) are on matches we cover. The other 34.9% are not, or not yet, matched to a match we cover and are not measured. Of the 11,547 unmatched predictions we looked up in a fixture catalogue, 7,838 (67.9%) were real matches in competitions we do not cover. If a site is better or worse there, we do not see it.
- Markets. Correct score has no closing price, so it is scored separately (in the CSV, against the estimated fair chance of the scores picked) and is not part of any verdict. Combination bets, bet builders and picks written only in sentences are not measured. Some sites publish only a few plain picks a day, so they reach a verdict slowly.
- Odds. The best price across bookmakers is a price a bettor would have to shop for. It makes the comparison easier for the sites, not harder.
- Aggregate figures only (2). SoccerVital and Rue des Joueurs: their terms of use do not allow reproducing their content, so we publish our measurements of them and never their picks.
- Free pages only. Paid tips, Telegram products and premium packages of these sites are not measured.
- Sites we do not measure. SportyTrader, FootyStats and Betensured: when we tried (October and November 2026), their pages refused automated reading or showed a "verify you are human" check, and we never solve those. SoccerVista, OLBG, Wettbasis, BettingClosed, LeagueLane and Opta Analyst: not added. When we looked at their terms of use and robots rules (29 October and 14 November 2026) we were not confident that they allow this kind of daily reading, so we left them out rather than guess. The absence of a site says nothing about how accurate it is.
What the numbers do not say
- Below the closing line does not mean worse than chance. Bookmakers keep a margin, so even the best bookmaker price is, on average, below the fair price at any moment. On the 11,954 picks where we also recorded the fair price when we read them, the odds were on average 2.1% below the fair price at reading and 2.0% below it at the close. Picks chosen at random at the same prices would get about the same verdict. What the verdict says is that these picks did not find prices good enough to overcome that margin, which is what beating the closing line takes.
- Closing line value should not be compared between sites. It depends on the markets and leagues a site picks, because bookmaker margins differ between them. On the same picks, the odds ended below the fair closing price by: match result 1.7%, over/under 2.4%, double chance 3.1% and both teams to score 3.6%. Two sites with the same skill and different markets would show different numbers.
- A site below the closing line is not a scam. Several of these sites say themselves that their tips guarantee nothing.
- Nobody is named the most accurate. Hit rates differ mostly because sites pick different markets and different odds; ranking them on that would be ranking noise.
- A positive return over a few hundred picks is not evidence of an edge, and a negative one is not proof of a bad site.
- Nothing here is betting advice.
How to cite
TipsAudit (18 November 2026). Football prediction sites accuracy study, November 2026. https://tipsaudit.com/blog/prediction-sites-accuracy-study-november-2026
The table is the CSV on this page (CC BY 4.0). The DOI of our prediction dataset covers that dataset, not this table.
Right of reply and corrections
If a site in the table disputes a figure, write to us: we will publish the response next to the figure on its review page and note it here. If a site asks us to stop reading it, we stop and note the date. If we find an error in this study we correct it with a dated note on this page; the corrections page lists the corrections we announced.