Overview
- Spring AI CTO Yehonatan Dodeles published a Medium report claiming a reverse-engineering analysis found TikTok’s recommender amplifies pro–Zohran Mamdani content and suppresses support for Andrew Cuomo.
- The study describes an 'Excessive Publicity' model that flags content receiving more exposure than expected, using data from real and synthetic users to estimate non-organic boosts.
- Researchers report 17% of all videos showed elevated exposure, political content rose to 55%, and Mamdani and Cuomo keywords had the highest excessive-publicity rates at 74% and 79% respectively.
- After normalizing within politics, the analysis says pro-Mamdani and anti-Cuomo videos exceeded the already-elevated baseline, while pro-Cuomo content fell below it; the model reportedly identified known ads with 76.4% accuracy.
- TikTok issued an unequivocal denial calling the work a bogus study, the findings remain unverified by independent researchers, and the claims surface weeks before New York City’s election.