Which AI picks the World Cup winner? I asked five

I asked five AI assistants the same simple question on the same afternoon: who is going to win the 2026 World Cup? Four of them said France. One said Spain. Not one of them hedged, which for a question still being decided by twenty-two players and a ball takes a certain confidence.

There’s no trick here, and no grading. The tournament isn’t over. As I write this we’re at the quarter-final stage, eight teams left, France the bookies’ favourite, the final on Sunday 19 July at MetLife Stadium. So this is a prediction, not a test with a right answer. What I wanted to see was simpler: hand five AIs the identical question and watch whether they agree. They mostly did. The interesting part is the one that didn’t, and how it argued.

What I asked

The prompt, word for word, put to ChatGPT, Claude, Gemini, Perplexity and Grok:

Each on its default model, in a fresh chat, on 8 July 2026, web search on wherever the tool exposes the toggle. Nothing clever. The sort of thing you’d type on your phone at half-time.

The split

Four for France, one for Spain.

France was the safe money, and the models knew it: betting favourite, deepest squad still standing, clean sheets through the knockouts. Four different tools, four slightly different reasons, one name. Claude was the only one that stopped to reframe the question before answering it, pointing out that with the tournament this far along this was now “a live read rather than a preseason guess” rather than the open-ended punt the question sounds like. A small thing, but it was the only one that noticed the question had aged.

The one that showed its working

Then Perplexity turned up with a supercomputer and a reading list.

It picked Spain, and it was the only one of the five to put a real model behind the call rather than a paragraph of reasoning. Its number came from Opta’s supercomputer: Spain on a 16.02% chance to lift the trophy, ahead of France on 12.54% and England on 10.66%, with fourteen sources listed under the answer.

I’m not saying Perplexity is right. Nobody knows if Perplexity is right, which is rather the point of a World Cup. What’s worth noticing is the difference in how the answer was built. Four assistants told me France with a sentence of reasoning. One told me Spain and showed me the numbers, the model they came from, and where to check them.

There’s a quieter version of the same gap. Claude ran four web searches to answer this, then handed me France in a confident, well-argued paragraph with not a single link to click. Its working was real. I just couldn’t see it. Perplexity’s was laid out on the table. When the stakes are a bit of World Cup fun, that difference is a curiosity. When the question is which fund to buy or whether some rule still applies, “show me where that came from” stops being a nicety, which is exactly why I keep a running check on whether an AI’s cited sources actually say what it claims.

The honest bit

This is a prediction, and predictions are just confident guesses in a nice jacket. There is no right answer today. There will be one on 19 July, and it will settle this cleanly: either four-fifths of the assistants I asked look shrewd, or Perplexity’s supercomputer does, or the whole lot of them are watching Argentina, England or Morocco lift it and saying nothing at all.

That’s the whole fun of it, and the reason this one’s worth a bookmark. So, which AI predicts the World Cup winner correctly? I genuinely don’t know yet. Come back after the final and you can see who called it. I’ll update this post with the result, and I won’t quietly delete the ones who got it wrong.

Ask me on the 20th.