Most prediction sites tell you how good they are. Almost none show you the receipts. So here they are — every single one of them.
Across all 104 matches of the 2026 World Cup, from the opener on 11 June to the final on 19 July, PredictoBets published a tip on every game from both our AI model and our human expert. Then we did the part nobody likes: we graded each pick against the final result and added up the profit and loss. No cherry-picking, no quietly deleting the losers. This is the full scorecard.
The headline results
Flat-staking one unit on every tip across the tournament, here is how the two approaches finished:
- Expert tips: 67 of 104 correct — 64.4% accuracy, +20.25 units profit, a +19.5% return on investment.
- AI model: 58 of 104 correct — 55.8% accuracy, roughly break-even at +1.25 units (+1.2% ROI).
- At least one of the two on target: 70.2% of matches (73 of 104).
- Both combined: backing every tip from both returned +21.5 units across 208 bets — a +10.3% ROI.
To put the expert number in real money: at £10 a unit, those tips turned £1,040 staked into £202.50 profit. At £100 a unit, £2,025. A 19.5% ROI across a full major tournament is a genuinely strong run — and it's a number we can only quote because we track every result.
How we grade every pick
Before each match we publish one clear tip: the market, the selection and the approximate odds. After the final whistle, an automated evaluation checks that selection against the actual score and marks it correct or incorrect — the same way a bookmaker settles a bet. Over/Under picks are checked against the total goals, Both Teams to Score against whether each side found the net, and so on. Nothing is graded by hand or by opinion, which is exactly why we're comfortable publishing the misses right next to the hits.
AI vs expert: the human won this one
The honest story of this World Cup is that our human expert comfortably out-performed the AI. The model landed a respectable 55.8% of its calls, but at average odds of 1.82 that only just cleared break-even. The expert hit 64.4% at similar prices — and in betting, the gap between 56% and 64% is the difference between treading water and a seriously profitable month.
We think that's a feature, not an embarrassment. The two work best together: the AI is relentless and unemotional across every fixture; the expert adds the context — squad news, motivation, tournament fatigue — that a model can miss. This time, the human read it better. Publishing the numbers every time is how we keep both honest.
Where the money was made: goals markets
The engine behind the expert's return was the Over/Under goals market. Of 99 Over/Under calls, 63% landed, banking +16.5 units on their own. At this World Cup, reading the tempo of a game — whether it would open up or stay tight — proved more reliable than predicting who would win it outright. That's a lesson the model is already learning from.
A few calls we're proud of
- USA 4–1 Paraguay — Over 2.5 goals, five in the game. Comfortable.
- South Korea 2–1 Czechia — both Both Teams to Score and Over 2.5 landed on the same match.
- Mexico 2–0 South Africa — Under 2.5 goals and "no BTTS", both correct as Mexico's defence held.
And the misses were just as public — Canada 1–1 Bosnia burned our "no both teams to score" call when Bosnia grabbed a late equaliser. That's the whole point: you get to see all of it.
What the World Cup taught us — and what's next
A tournament is a stress test. Unfamiliar teams, neutral venues and knockout tension make it one of the hardest events to predict, so a +19.5% expert ROI and a 70% "at least one right" rate is a run we're proud of. The bigger point is the habit: we grade everything, we publish everything, and we let the results speak.
The domestic season is where that edge really compounds — far more data, far more matches. Our AI and expert tips are live for every game across the top leagues, from the Premier League and Serie A to the Champions League. Same approach, same transparency: every pick tracked, every result shown.
Follow this season's tips — and hold us to the numbers.


