16/06/2026
Part 2 of the AI vs Humans World Cup Fantasy experiment is live.
Same prompt. Same rules. Same budget. Five completely different strategies — and one moment that perfectly sums up the biggest risk with AI.
Each model explained its reasoning in full. ChatGPT played it safe and sensible. Claude Sonnet actively tried to outsmart the other AIs — it predicted what its competitors would pick and deliberately went the other way. Claude Opus produced the most human-like reasoning of the lot, referencing real injury news and fixture difficulty. Gemini Pro went purely mathematical. Gemini Flash fielded five midfielders, three of them German.
But the standout moment? One model quoted a player's price completely wrong — and stated it as fact with total confidence. No hesitation, no uncertainty. Just a made-up number presented as reality.
It's a perfect example of why AI always needs proper setup and supervision. It doesn't know what it doesn't know.
The wider point beyond the football — five models, same information, same rules, five completely different strategies. That's exactly what happens when businesses use AI. The approach matters as much as the tool.
Part 3 coming soon — why did every single AI ignore Messi and Ronaldo?
Full breakdown here 👇
purepublishing.com/blog/world-cup-fantasy-how-ai-picked