Research articles and in-depth analysis from the AI March Madness 2026 team. Topics include AI prediction methodology, source citation analysis, confidence calibration insights, prediction drift patterns, upset detection, prompt sensitivity testing, and tournament strategy.
This section contains 37 research articles covering how GPT-4o, Gemini 2.5, and Perplexity Sonar Pro approach NCAA Tournament predictions. Each article examines a specific aspect of AI forecasting with data from our automated collection pipeline.
PERPLEXITY VS GPT-4O: HOW MUCH SOURCE OVERLAP IS THERE?
GPT-4o with web search and Perplexity Sonar Pro are both real-time search-enabled models. You might expect their source domains to largely overlap. They don't.
IT
Intelligence Team
Source Intelligence
Mar 12, 20265 min readUpdated May 1, 2026
PerplexityGPT-4oSourcesEnsemble
THE OVERLAP ANALYSIS
Our source overlap analysis from initial collection runs shows less than 30% domain overlap between GPT-4o and Perplexity citations. GPT-4o concentrates on ESPN, CBS Sports, Sports Illustrated, and The Athletic. Perplexity surfaces KenPom, Barttorvik, HerHoopStats, and conference-specific fan sites.
WHY LOW OVERLAP CREATES ENSEMBLE VALUE
When two forecasters draw from different evidence bases, their errors are uncorrelated. An ensemble that averages their confidence scores outperforms either solo because correct picks reinforce each other while errors cancel out.
We're watching the Source Intelligence leaderboard throughout the tournament to see which domain types correlate with correct predictions. Results will shape how we weight ensemble picks by mid-tournament.
LIVE DATA
See this tracked in real-time as the tournament plays out.
