Our Source Intelligence dashboard tracks every URL each model returns as a citation. After initial collection runs, the patterns are clear: GPT-4o concentrates on 5–6 major sports media domains. Perplexity spreads across 15–20 domains including niche analytics. Gemini sits between, mixing official statistics with national coverage.

Different sources have different biases. National sports media overweights storylines: the Cinderella team, the star player's injury narrative, the coach's Final Four history. Analytics domains underweight momentum and overweight efficiency metrics in small samples.

Models that cite narrower source sets also tend toward higher confidence scores - but not higher accuracy. This is a calibration red flag we track throughout the tournament.

The Source Intelligence page ranks every domain cited across all three models by total citation count and by per-model breakdown. As the tournament runs, we'll layer in win-rate data to show which domain types correlate with correct picks.