How Our AI Did: 10 of 12 World Cup 2026 Predictions Correct (4 Exact Scores)
By Ali Ammar · goals2026.org
It is easy to publish predictions and quietly forget the ones that miss. We would rather show our work. Every time our model locks in a pre-match prediction, we snapshot it before kickoff so it can be graded honestly afterwards — no hindsight edits. Here is the full scorecard from the group-stage finale through the semi-finals.
The headline numbers
- Correct outcomes: 10 of 12 (83%) — win, draw or lose, including ties settled on penalties.
- Exact scorelines: 4 of 12 (33%) — the precise result, not just the winner.
- Quarter-finals: 4 from 4 outcomes, 3 exact— the model's best round.
Predicting an exact scoreline in knockout football roughly one time in three is well above chance, and an 83% outcome hit-rate across a full month of matches is a result we are happy to be judged on.
The full scorecard
| Match | Predicted | Actual | Result |
|---|---|---|---|
| Algeria–Austria | 1–1 | 3–3 | ✓ outcome |
| Jordan–Argentina | 0–2 | 1–3 | ✓ outcome |
| Portugal–Spain | 1–2 | 0–1 | ✓ outcome |
| USA–Belgium | 1–1 | 1–4 | ✗ miss |
| Argentina–Egypt | 2–0 | 3–2 | ✓ outcome |
| Switzerland–Colombia | 1–0 | 0–0 (pens) | ✓ outcome |
| France–Morocco | 2–0 | 2–0 | ✓✓ exact |
| Spain–Belgium | 2–1 | 2–1 | ✓✓ exact |
| Norway–England | 1–2 | 1–2 | ✓✓ exact |
| Argentina–Switzerland | 2–0 | 3–1 | ✓ outcome |
| France–Spain | 2–1 | 0–2 | ✗ miss |
| England–Argentina | 1–2 | 1–2 | ✓✓ exact |
The two it got wrong
Honesty means owning the misses. The model called USA–Belgium a 1–1 draw; Belgium won 4–1. And it backed France to beat Spain 2–1 in the semi-final — Spain won 2–0 and never looked like losing. Both misses share a theme: the model under-rated how completely one side would dominate. It reads probabilities well; blowouts are where it is humblest.
How the grading works
There is no cherry-picking here. The prediction our model publishes on each match page— win probabilities and a most-likely scoreline — is written to the database the moment it is generated, before the match starts. That snapshot is frozen; when the final whistle goes, we compare it to the real result. An "outcome" hit means we called the right winner (or a draw, with knockout ties judged on the penalty result). An "exact" hit means the scoreline matched to the goal.
A pattern is already visible across these twelve games. The model is strong at reading a match's shape— who is likely to win, and roughly how comfortably — which is why the outcome rate sits at 83%. It is weakest at the extremes: it tends to hedge toward tighter scorelines, so genuine blowouts like Belgium's 4–1 or Spain's semi-final control slip past it. That is a reasonable trade-off for a model built to estimate probabilities rather than chase dramatic results, but it is exactly the kind of thing that only shows up when you grade yourself in public.
What it means for the final
For the record, before the Spain–Argentina final our model predicts Argentina 2–1 (43% Argentina, 31% Spain, 26% draw). Given its semi-final wobble against Spain, that is a prediction we will be watching closely — and grading publicly, win or lose. See the live final prediction and the full tournament odds.
Graded from pre-match prediction snapshots in goals2026.org's database, group finale through the semi-finals, correct as of July 17, 2026. Snapshots began at the group stage's end, so earlier rounds are not included.