A recent study indicates that TikTok in China may be using its content algorithm to significantly promote posts favoring Socialist Party mayoral candidate Zoran Mamdani in New York City, while also hindering visibility for former Mayor Andrew Cuomo. The researchers suggest this behavior transcends simple bias and presents a form of “algorithmic steering.”
“Our findings show that TikTok’s algorithm is preferentially boosting pro-Mamdani and anti-Cuomo content in the current mayoral election,” stated Yehonatan Doderes, CTO of Spring AI, in a recent post on X. The report is available on Medium.
“This analysis was carried out by Spring AI,” Doderes continued, emphasizing the company’s mission to uncover the hidden intricacies of social media algorithms to defend democracies against manipulative AI.
“Through a rigorous reverse engineering of TikTok’s recommendation system and analyzing both real and synthetic user data, we developed an AI model to determine instances where a video gains more visibility than it should. We call these ‘overhyped’ videos,” he clarified in a follow-up post on X.
Doderes noted that roughly 17% of all videos displayed this sort of unnatural increase in visibility and mentioned that their model effectively identified 76.4% of known paid ads, effectively validating its functionality.
“In summary, this effort is aimed at uncovering hidden promotions,” he explained.
“Subsequently, we searched for any topics that strayed from the typical patterns,” Doderes added, noting that “most topics did not show significant deviation.”
“Even popular hashtags were around the standard 17% level, which indicates our model can differentiate between mere popularity and actual promotion,” said the CTO. “However, one topic stood out: politics, where 55% of political videos were considered overhyped, which is three times the usual rate.”
Doderes mentioned that the names “Mamdani” and “Cuomo” prominently featured in the analysis, and his team created another AI model specifically to classify “pro” and “anti” videos and validate its results manually.
He remarked that when normalizing within the political category—which was already at 55% promotion—videos supporting Mamdani and opposing Cuomo both surpassed the baseline.
In contrast, he indicated that “pro-Cuomo” content was “being suppressed,” which he described as a form of “algorithmic steering” rather than mere “bias.”
“These are initial findings, and more research will be necessary, but the potential implications are critical,” the CTO of Spring AI noted.
“The algorithms that influence our view can also shape our beliefs,” Doderes concluded, linking to his team’s article for a detailed exploration of these findings.
TikTok, a platform extremely popular with American youth, is already recognized for its endless supply of brief, often provocative videos, contributing to its reputation as a leading source of persuasive messaging.
Peter Schweitzer, a noted author and senior contributor at Breitbart News, pointed out that a military academic in China suggested that the actual conflict between the U.S. and China revolves around “information-driven psychological warfare.”
