I build AI solutions for small businesses. I spend my days helping people understand what AI can and can't do. So when the World Cup 2026 rolled around and I built a fantasy football app for me and a couple of mates, I had an idea.
What if I entered AI as actual competitors? Not to help me pick my team. Not as an advisor. As independent players in the league, picking their own squads, making their own decisions, and competing head-to-head against real humans.
So that's exactly what I did.
The Experiment
I built a custom World Cup fantasy football web app from scratch — a full-stack application with squad selection, live scoring, and a leaderboard. Then I invited three humans and five AI models to compete in the same league.
The AI competitors: ChatGPT, Gemini Flash, Gemini Pro, Claude Sonnet, and Claude Opus. Each was given the identical prompt. Same rules, same player list with prices, same budget, same scoring system. No special instructions. No hints.
“Pick the best possible 11-player squad within a £100m budget.”
Each model made its choices independently, with no knowledge of what the others picked.
The Rules
Every competitor plays by the same rules: Pick exactly 11 players within a £100m budget. Choose a valid formation. Maximum 3 players from the same country. Name a captain (double points) and vice captain (double if captain doesn't play).
Points for goals, assists, clean sheets, player of the match. Minus points for yellow and red cards. Midfielders score 8 points per goal versus 6 for forwards and 10 for defenders — so loading up on forwards isn't automatically the best strategy.
A Quick Look at Each Squad
ChatGPT
4-4-2 · £100m
The most balanced squad. Mbappé captain, Wirtz vice captain. Forwards: Mbappé, Gyökeres. Midfield: Baena, Fernandes, Wirtz, McTominay. Defence: Hakimi, Cucurella, Mendes, Dumfries. GK: Martínez. The sensible, conservative choice.
Claude Sonnet
4-3-3 · £99.5m
The aggressive gambler. Three premium forwards. Mbappé captain, Vinícius Júnior vice captain. Forwards: Mbappé, Vinícius Júnior, Yamal. Midfield: Wirtz, Valverde, Allister. Defence: Hernandez, Cubarsí, Bensebaini, Dedić. GK: Maignan. High risk, high reward.
Gemini Pro
3-4-3 · £98.5m
The maverick. Only AI to captain a midfielder. Bellingham captain, Mbappé vice captain. Forwards: Saka, Pulisic, Mbappé. Midfield: Bellingham, Musiala, Fernandes, Aouar. Defence: Aït-Nouri, Dest, Hernandez. GK: Martínez. Left £1.5m unspent.
Claude Opus
3-4-3 · £100m
The household names squad. Mbappé captain, Kane vice captain. Forwards: Mbappé, Kane, Torres. Midfield: Bellingham, Fernandes, Allister, Reijnders. Defence: Cucurella, Magalhães, Romero. GK: Martínez. Every player a recognisable name.
Gemini Flash
3-5-2 · £99.5m
The wildcard. Five midfielders, three German. Mbappé captain, Haaland vice captain. Forwards: Haaland, Mbappé. Midfield: Bellingham, Wirtz, Musiala, Fernandes, Karl. Defence: Alaba, Lindelöf, Abdulhamid. GK: Samba. Will look genius or catastrophic.
What They All Agreed On
Every single AI picked Mbappé — five out of five, captain in four. Bruno Fernandes appeared in four out of five squads. Martínez in goal for three out of five. Bellingham appeared in three squads.
When five different models built by three different companies independently reach the same conclusions, it tells you something.
Where They Disagreed
Formations varied wildly — 4-4-2, 4-3-3, 3-4-3, 3-5-2. Only Claude Sonnet paired Vinícius Júnior with Yamal. Only Gemini Flash picked Haaland. Only Gemini Pro captained a midfielder. Defensive strategies ranged from cautious to frankly reckless.
The Human Factor
The three human managers have something the AIs don't: years of watching football, gut instinct, and irrational confidence. But the AIs have zero bias. They didn't pick a player because he plays for their favourite club. Every decision was based on data, value, and probability.
The humans know things data doesn't capture — quiet injuries, squad rotation, team togetherness. The AIs know things humans can't process — the mathematical value of every possible combination within £100m.
So who wins? Data or instinct?
What's Coming Next
Individual squad breakdowns with each AI's full reasoning for every pick. Regular updates once the tournament kicks off. And the final verdict after the last whistle.
Why This Matters Beyond Football
This experiment demonstrates how AI works in the real world. Five models, same information, same constraints, five completely different strategies. That's exactly what happens when businesses use AI. The tool matters, but how you set it up matters more.
This is Part 1 of an ongoing series. Pure Publishing builds custom AI solutions and websites for small and medium businesses.