AI predicts the World Cup? Or just telling us what we want to hear?

The World Cup group stage has finished, and I’ve seen countless posts using AI to predict match outcomes. On the surface, it looks like straightforward prediction: you input teams, statistics, and match context, and it outputs a win or loss result.

Using different AIs and different prompts produces different results, yet each one feels hard to dismiss because the reasoning behind them is so coherent.

I can’t help but wonder: is AI actually predicting outcomes, or is it simply generating explanations that we find convincing, while the predictions themselves remain unreliable?

On SportEval AI, nine different AI models make simultaneous predictions for World Cup matches. In Curaçao vs Côte d’Ivoire's match, most models converged on the same prediction. But in Ecuador vs Germany, all of them produced different outcomes.After reading their analyses, I noticed something interesting: Claude feels like an economist, interpreting things through market-style data and trends, and appears relatively accurate; DeepSeek feels like a seasoned football fan, weighing squad value and recent performance before giving an answer; GPT comes across as a hybrid, with deeper analysis that even considers future fixture implications. Each model constructs its own internally consistent reasoning framework.

So what are we actually looking at: prediction, or explanation? Football is inherently a highly uncertain sport.Before a match, no one truly knows what will happen. Yet every analytical approach can construct a coherent logic from available information. After the match, correct predictions are seen as insightful, while incorrect ones are dismissed as having missed key factors. But before kickoff, most of these analyses still appear reasonable.

Often, the real discussion is not about who can accurately predict the future, but about who can provide the most convincing explanatory framework.