⚡ Short answer
In soccer, the match winner: 61.9% hit rate and +4.0% over 386 picks, the best combination of volume and result. Outside soccer, NHL props (+23.1%) and the MLB winner (+20.5%).
The market that loses the most with a large sample is “Over 2.5 goals” (−9.8%). And the draw hits only 31.0% but earns +19.6%: hitting is not the same as winning.
🧩 What a market is and why it changes everything
A market is the type of bet: who wins, how many goals there are, whether both teams score, how many shots a player takes. The same AI can hit a lot in one market and little in another, because each one depends on different things.
That is why asking “is the AI right?” without saying in which market is an incomplete question. What is useful is knowing where. To answer it we use the 1,864 settled predictions in the verified record (March 21 to November 14, 2026), grouped by type of bet, with the hit rate and profit (yield) of each group.
📖 Markets in 30 seconds (for anyone just starting)
- Match winner (1X2): you bet that the home team wins (1), the match is a draw (X) or the away team wins (2).
- Double chance: you cover two of the three outcomes. 1X = home win or draw; X2 = draw or away win.
- Over or under N goals: “Over 2.5” wins if there are 3 goals or more; “Over 1.5,” if there are 2 or more.
- Both teams to score: wins if both teams score at least one goal.
- Handicap: you give one team a head start or a deficit in goals or points before the match begins.
- Player props: bets on what a player does, such as shots, points or runs.
🧮 The math that explains everything: fair odds
To understand why one market pays off and another does not, there is one calculation worth having clear: fair odds = 1 divided by the probability. If something happens 50% of the time, the fair odds are 2.00. If it happens 25% of the time, they are 4.00. If your bookmaker pays more than the fair odds, there is value; if it pays less, there is not.
Illustrative example (not a real pick): the AI calculates 60% for the home team, which gives fair odds of 1.67. If the bookmaker pays 1.80, there is value (the bet returns +8% on average). If it pays 1.55, there is not: even if the home team wins often, at that price you lose money in the long run.
Each market has its own relationship between probability and price, and that is why profit does not follow hit rate. It is the core idea behind value betting and the expected value calculator.
📊 The soccer market ranking
| Market (soccer) | Picks | Hit rate | Profit (yield) | Sample |
|---|---|---|---|---|
| Draw | 58 | 31.0% ±12 | +19.6% | acceptable |
| Cards | 11 | 63.6% ±28 | +15.3% | very small |
| Double chance 1X (home win or draw) | 64 | 64.1% ±12 | +6.3% | acceptable |
| Over 1.5 goals | 62 | 67.7% ±12 | +4.4% | acceptable |
| Match winner (1X2) | 386 | 61.9% ±5 | +4.0% | large |
| Others (first half, handicap…) | 59 | 50.8% ±13 | +2.7% | acceptable |
| Both teams to score | 73 | 65.8% ±11 | +1.3% | acceptable |
| Double chance X2 (draw or away win) | 112 | 53.6% ±9 | −6.7% | acceptable |
| Over 2.5 goals | 265 | 54.7% ±6 | −9.8% | large |
| Both teams not to score | 33 | 45.5% ±17 | −12.7% | small |
| Under 3.5 goals | 18 | 38.9% ±23 | −23.9% | very small |
| Corners (full match) | 31 | 41.9% ±17 | −26.7% | small |
Soccer picks settled through November 14, 2026, sorted from highest to lowest profit. Sample: large = 150 picks or more; acceptable = 50 to 149; small = 20 to 49; very small = fewer than 20. “±” is the margin of error on the hit rate (95% confidence).
The soccer markets where the AI does win with an acceptable sample are the match winner and double chance 1X; the one that loses the most with a large sample is Over 2.5 goals.
📊 The market ranking outside soccer
| Other sports | Picks | Hit rate | Profit (yield) | Sample |
|---|---|---|---|---|
| 🏒 NHL · player props (shots, points) | 95 | 60.0% ±10 | +23.1% | acceptable |
| ⚾ MLB · match winner | 112 | 58.9% ±9 | +20.5% | acceptable |
| 🏀 NBA · player props | 133 | 54.9% ±8 | +3.7% | acceptable |
| 🎾 Tennis · winner | 125 | 56.8% ±9 | +1.4% | acceptable |
| ⚾ MLB · props and totals | 36 | 47.2% ±16 | +4.0% | small |
| 🏀 NBA · handicap | 62 | 48.4% ±12 | −10.9% | acceptable |
| 🏀 NBA · point totals | 49 | 42.9% ±14 | −15.5% | small |
| 🏈 NFL · props, handicap and totals | 63 | 41.3% ±12 | −26.3% | acceptable |
Same data and same criteria. In hockey, “winner” picks are only 4 and are not shown separately.
🥇 The markets where the AI does best
⚽ Match winner: soccer’s star market
It is the market with the most data and the most reliable: 386 picks, 61.9% hit rate and +4.0% profit. It is the one that best combines volume and result, and the one that depends least on the luck of a single goal.
🧱 Double chance 1X and “Over 1.5 goals”: they hit a lot and earn a little
Double chance 1X (the home team wins or draws) has a 64.1% hit rate and +6.3%. “Over 1.5 goals” posts 67.7% and +4.4%. These are bets with a high hit rate and low odds: they earn little per pick, but often. With 64 and 62 picks, the sample is fair: the margin of error on the hit rate is ±12 points.
Watch out for the mirror image: double chance X2 (the away team wins or draws) does the opposite: 53.6% hit rate and −6.7% over 112 picks. Same idea, opposite result, and we have no proven explanation. The mechanics of this market are in the double chance guide.
⚾ Baseball winner and 🏒 hockey props: the biggest earners
The MLB winner delivers +20.5% (112 picks, 58.9% hit rate). NHL player props, +23.1% (95 picks, 60.0%). They are the two best numbers in the table. In hockey’s case it is barely two months of history, and in both the sample is still only medium-sized. The details by sport are in the best sport article.
🧭 The cross-check: the match winner, league by league
The match winner is the market with the most data, so it works for crossing market and league. In soccer, 1X2 performance changes quite a bit depending on the competition:
| League (match winner only) | Picks | Hit rate | Profit (yield) |
|---|---|---|---|
| Liga Profesional (Argentina) | 22 | 68.2% ±19 | +25.0% |
| Serie A | 42 | 71.4% ±14 | +16.6% |
| World Cup 2026 (tournament already played) | 27 | 77.8% ±16 | +16.3% |
| Copa Libertadores | 28 | 67.9% ±17 | +14.3% |
| Premier League | 32 | 62.5% ±17 | +6.9% |
| Brasileirão | 27 | 55.6% ±19 | −0.1% |
| Ligue 1 | 23 | 43.5% ±20 | −11.2% |
| Copa Sudamericana | 26 | 50.0% ±19 | −29.2% |
Only leagues with 20 or more match-winner picks: these are small samples, so the margin of error is large (from ±12 to ±20 points). Read it as a clue, not a conclusion.
What this cross-check shows: the same market performs differently depending on the league, and the samples are so small that it is best not to draw strong conclusions. For the full league ranking, it is in the first article of the series.
🔭 The markets we cannot judge yet
Some markets have so few picks that any number would be noise: soccer cards (11 picks), the NHL winner (4 picks) or baseball totals (5). We leave them out of the conclusions on purpose. A percentage over 10 picks can be 70% or 30% by pure luck. The honest thing is to wait for more data before claiming anything.
🔁 The draw case: low hit rate, yet it wins
The draw is the best proof that hitting is not winning: it hits only 31.0% of the time, and still delivers +19.6% profit (58 picks). It is simple math: the average odds on those picks are 3.22. At odds of that size, hitting a little more than 1 in 3 is enough to make money, and it did. But watch the sample: 58 picks, margin of error of ±12 points. It is an interesting clue, not a recipe. More on related bets in double chance vs draw no bet.
⚽ “Over 2.5 goals”: the market that loses the most, and what changed
“Over 2.5 goals” is the market with the biggest loss and a large sample: 265 picks, 54.7% hit rate and −9.8% profit. Almost all of them are from before October 14: 264 picks at −10.0%. That day the criteria changed: since then, the over or under 2.5 goals market is only published when there is an edge against Pinnacle’s fair price (a very sharp bookmaker used as a reference).
Since the change, only one “Over 2.5” pick has come out. In other words, there is no sample to judge the new criteria. The honest thing is to say both: the old method lost, and the new one has no data yet.
❌ The markets to be wary of
- Pre-match corners: 31 picks, 41.9% and −26.7%. Small sample, but with a clearly negative sign. (The corners bot works differently, live, and is not part of this count.)
- Under 3.5 goals and “both teams not to score”: −23.9% and −12.7%. These are bets that things will not happen, and one extra goal breaks them.
- NBA (handicap and totals): −10.9% and −15.5%. NBA player props do better (+3.7%).
- NFL: −26.3% over 63 picks.
None of these numbers is a verdict: the samples are small. But if you have to cut bets, cut here.
🚫 Common myths about markets
“Goals markets are the easiest”: in our record, “Over 2.5” was the biggest loser.
“Bets with the highest hit rate are the best”: the draw hits rarely and won; both teams to score hits a lot and barely earns.
“An exotic market has more value”: sometimes it does, but with small samples it is impossible to know. Always check how many picks are behind it.
Choosing the right market is worth as much as choosing the match.
🎯 How to use this ranking
- Prioritize the match winner and markets with positive profit and an acceptable or large sample.
- Treat small-sample markets as tests, with a small stake.
- Cut the markets in the red before cutting good leagues.
- Cross both filters: market and league. A good market in a bad league loses anyway: see the league ranking.
- If you are looking for which bets are most stable, read the AI’s safest bets.
🚀 How to see the market of each pick
Every Gambeta prediction shows the market, the probability the model calculates, the minimum odds and the suggested stake. You can filter the record by market and repeat these calculations with the public CSV.
❓ Frequently asked questions
In which market does the AI hit most?
Among soccer markets with a large sample, the match winner (61.9% and +4.0% over 386 picks). “Over 1.5 goals” hits 67.7% and “both teams to score,” 65.8%, with smaller samples.
Which market earns the most profit?
In soccer, the draw (+19.6%), but with 58 picks and only a 31.0% hit rate. Outside soccer, NHL props (+23.1%) and the MLB winner (+20.5%).
Why does “Over 2.5 goals” lose so much?
Over 265 picks it ended at −9.8%. Almost all of them are from before October 14, when the criteria changed: now it is only published with an edge against Pinnacle’s price. There is no sample yet for the new method.
Are corners worth betting on with AI?
In full-match corners predictions, no: 31 picks and −26.7%. It is a small sample, but with a clear sign. The corners bot works live and is measured separately.
Why does the draw win if it almost never hits?
Because it pays a lot: the average odds on those picks are 3.22. At those odds, hitting 1 in 3 is enough to win. But with 58 picks the margin of error is large.
What is the difference between double chance 1X and X2?
1X covers home win or draw; X2, draw or away win. In the record, 1X delivers +6.3% and X2, −6.7%. More in the double chance guide.
👉 If you want to see how AI, bots and automation fit together, read technology in sports betting. And if you are starting from scratch, the starting guide is AI for sports betting: what it is and how to use it.
🧭 AI for betting, sport by sport
Gambeta's guides by sport and by league: how the AI reads each match, which markets it covers and what track record stands behind it.
🤖 Corners bot
It tells you when a match is ending and a corner is still missing. You bet "at least one more corner" and follow the alert.
🤖 Get the bot🧮 Tools
- Best AI for Sports Betting: ChatGPT, Gemini, Claude & More
- Build Your Own AI for Sports Betting: Step-by-Step Guide
- Automatic Betting Alerts: Build Your Software, No Code
- Free AI for Sports Betting: 7 Questions Before You Trust It
- Correct Score Prediction AI: Why It Almost Never Pays (2026)
- Free AI Prediction Bot: What It Really Is and a 7-Day Test
- 📐 Kelly Criterion calculator
- 🧮 Expected value (EV) calculator
- 🔄 Odds converter
- 🛠️ ChatGPT prompt builder
