Skip to main content
Case Study

World Cup 2026 Fantasy Football

Can AI beat humans at fantasy football? We built an app to find out.

AI ExperimentWeb AppFull-Stack
5AI models competingChatGPT, Claude Sonnet, Claude Opus, Gemini Pro, Gemini Flash
3human competitorsReal people with real football knowledge
8blog posts and countingDocumenting the entire experiment

The Idea

What if AI could compete against humans at something everyone understands?

I wanted to run an experiment that would demonstrate how AI thinks — its strengths, its limitations, and where it differs from human decision-making. Fantasy football was the perfect test. It requires data analysis, risk assessment, budget management, and gut instinct. It's something anyone can follow, and the results are measurable.

So I built an app, entered five AI models as independent competitors alongside three humans, and documented everything.

What We Built

A full-stack fantasy football web application.

The app was built from scratch using Next.js, TypeScript, Supabase, and Tailwind CSS — the same stack we use for our client projects. It features a complete player database with pricing, squad selection with budget and formation constraints, live scoring powered by an ESPN API pipeline, an automated points calculation system, and a real-time leaderboard tracking all eight competitors.

Each AI model was given the same prompt with the full player list and asked to independently select its squad. No special treatment, no hints — just raw AI decision-making.

What we built

  • Next.js 14 full-stack application
  • Supabase database with player pricing
  • ESPN API integration for live scores
  • Automated points pipeline
  • Squad validation (budget, formation, country limits)
  • Real-time leaderboard
  • 5 AI model integrations
  • Built and deployed in under a week

What We Learned

AI is brilliant at data. Humans have instinct. The jury's still out.

The experiment revealed fascinating differences in how each AI model approaches the same problem. All five picked Mbappé — consensus on the obvious choice. But their strategies varied wildly, from conservative to reckless. One model captained a midfielder for mathematical reasons. Another hallucinated a player's price. A third based a key decision on injury news that turned out to be wrong.

The humans, meanwhile, made emotional decisions they later rationalised — including one manager who reluctantly dropped his favourite player. The experiment is ongoing and being documented in a blog series that's already generating interest and conversation.

This is a private project — the app is not publicly accessible.

Let's Talk

Want to see what AI can do for your business?

Fantasy football is fun. But AI can do serious things for your business too — from handling customer enquiries to automating your admin.