TikTok Stacking Algorithms in Chinese Government’s Favor, Study Claims

A study published on Thursday asserts TikTok’s algorithms promote Chinese Communist Party narratives and suppress content critical of those narratives, a claim the embattled company forcefully denied to KQED.

Titled “The CCP’s Digital Charm Offensive,” the study by the Rutgers University-based Network Contagion Research Institute argues that much of the pro-China content originates from state-linked entities. ByteDance, a Chinese technology company, owns TikTok.

Institute co-founder Joel Finkelstein wrote that includes media outlets and influencers, such as travel vloggers who post toothlessly about Chinese regions like Xinxiang, where the government has imprisoned more than 1 million Uyghurs and other mostly Muslim minorities.

“This manipulation is not just about content availability; it extends to psychological manipulation, particularly affecting Gen Z users,” Finkelstein wrote.

While American teenagers do not typically make geo-political decisions, Finkelstein argued a positively skewed view of China shaped by manipulation “undermines critical thinking and democratic discourse.”

NCRI said in its report that “our findings, which, while not definitive proof of state orchestration, present compelling and strong circumstantial evidence of TikTok’s covert content manipulation.”

TikTok has repeatedly said the Chinese government has no influence over its U.S. app, and proving otherwise would be difficult — something that the Department of Justice has acknowledged in discussions over a law that could ban the app.

NCRI’s report could add to a surge of concern in Washington, D.C. Other research outfits have urged TikTok to give them better access to data to study things like the mental health of young users.

“Trying to understand how social media works from the outside is difficult and getting more difficult because social media is tightening their grip on access,” said Hany Farid of UC Berkeley’s School of Information.

An NCRI analysis published in December looked at the volume of posts with certain hashtags — like “Uyghur,” “Xinjiang,” “Tibet” and “Tiananmen” — across TikTok, Instagram and YouTube. That report found anomalies in TikTok content based on its alignment with the interests of the Chinese government. For example, researchers wrote, hashtags about Tibet, Hong Kong protests and the Uyghur population appeared to be underrepresented on TikTok compared with Instagram.

NCRI claims that following the first report’s publication, TikTok disabled its hashtag measurement functionality in a move that made it impossible for the researchers to replicate their findings.

For its new report, NCRI researchers turned to “user journey” data, setting up 24 “sock puppet” accounts to mimic the new-to-the-platform experience of 16-year-old Americans. The accounts searched for each of four target keywords — Uyghur, Xinjiang, Tibet and Tiananmen — on TikTok, YouTube and Instagram, clicked on the first post that appeared and scrolled through subsequent videos fed by the algorithm.

Researchers also said they saw evidence of hashtag/keyword “hijacking,” attaching trending hashtags for topics considered highly sensitive or provocative by the Chinese government to unrelated posts in an attempt to muddy the waters and drown out genuine content. The researchers wrote, “60.3% of all content served by TikTok’s search algorithm was flagged as irrelevant in contrast to <5% for Instagram and YouTube, demonstrating the CCP influencing tactic of hashtag/keyword hijacking.”

In a statement, a TikTok spokesperson dismissed NCRI’s study. The spokesperson said many TikTokers use unrelated hashtags to drive traffic to their content and that it’s a practice common across all social media platforms. All platforms algorithmically promote engagement, so if a video is proving popular, TikTok’s algorithm is designed to elevate it, not bury it, the spokesperson added.

“This non-peer reviewed, flawed experiment was clearly engineered to reach a false, predetermined conclusion,” he wrote. “Creating fake accounts that interact with the app in a prescribed manner does not reflect real users’ experience, just as this so-called study does not reflect facts or reality.”