Seed and Conference Bias Analysis - AI March Madness 2026

Analysis of systematic biases in AI predictions for the 2026 NCAA Tournament. Measures whether GPT-4o, Gemini 2.5, and Perplexity Sonar Pro systematically favor higher-seeded teams or teams from power conferences (SEC, Big Ten, Big 12, ACC, Big East).

Seed bias occurs when a model picks the higher seed more often than historical win rates justify. Conference bias occurs when a model over-favors teams from certain conferences regardless of matchup-specific factors. Both biases can reduce accuracy in upset-prone rounds.

The bias register also flags analyst citations where the cited analyst has a documented connection (played, coached, family, hometown) to a team they predicted to win. This human-verified data reveals potential inherited bias from source materials.

Seed Bias

The tendency of an AI model to systematically favor higher-seeded (lower number) teams regardless of actual team quality metrics.

Conference Bias

The tendency to over-favor teams from specific conferences (e.g., SEC, Big Ten) beyond what performance data supports.

Analyst Affiliation

A documented connection between a cited sports analyst and a team they predicted to win.