One reported Elo-based forecast examined 49,000 matches, reduced the influence of older results and rewarded stronger victories. It made Spain the leading contender at 27%, ahead of Argentina on 21%.
Another system, built around 1,200 data points and 25 variables, reportedly favoured France. The difference is revealing. AI does not produce one universal prediction; it produces an outcome shaped by the information and assumptions it receives.
A reported retrospective test also returned 52% correct results across the 2018 and 2022 World Cups after 50,000 simulations. That is a useful benchmark, but hardly a guarantee.
The leading names are familiar: Spain, France and Argentina appear regularly near the top of forecast lists, with England, Portugal and Brazil also prominent.
Yet less conventional outputs have included Morocco reaching the semi-finals and Norway emerging as a possible surprise package. Some systems prioritise recent form, while others give greater weight to World Cup history, squad value, coaching or tournament experience.
That explains the volatility. Change the prompt or the weighting, and the bracket can change with it.
Paul’s appeal was simplicity. He made a selection, the world watched and every correct choice strengthened the myth. His reported calls around Germany, Argentina and Spain during the 2010 tournament made him more than a novelty; he became football’s most famous forecaster.
AI is more sophisticated, but less charming. Its predictions can be revised endlessly and rarely offer a single memorable moment.
As the 2026 World Cup approaches, final squads, injuries and warm-up form will reshape the forecasts. The next storyline may be less about finding a perfect prediction than seeing whether any model - or one unlikely human pick - can match Paul’s legend.