Researchers debunk TikTok censorship claims about ICE and Epstein : NPR

After a U.S.-led investor group took over the social media platform’s U.S. operations earlier this year, some users claimed political topics had been sidelined. New research contradicts their claims.

Riccardo Milani/AFP via Getty Images

hide caption

toggle caption

Riccardo Milani/AFP via Getty Images

As a consortium of investors led by Oracle’s Larry Ellison took control of TikTok’s U.S. operations, users accused the app of limiting videos about Immigration and Customs Enforcement raids, the late sex offender Jeffrey Epstein and posts related to the fatal shooting of Alex Pretti in Minneapolis.

The posts went viral on social media, blaming the perceived removal of content on TikTok’s new bosses. The hashtag #TikTokCensorship gained traction on X, masses of users downloaded alternatives to TikTok, and California Governor Gavin Newsom, as well as European Union lawmakers, called for investigations.

But a data center outage that wreaked havoc on the platform appears to have disrupted all categories of posts, rather than targeting political content, according to a new analysis in the publication Good Authority. It was conducted by eight academics who examined how videos evolved during ownership transition.

Using audience metrics on more than 100,000 videos, the researchers focused on videos about ICE, Pretti, Renee Good, the woman killed by an ICE agent last month, and the keywords “Trump” and “Epstein.” They compared how often TikTok recommended content versus non-political posts on topics like cooking recipes and the Oscars.

At the time of the TikTok server outage, “posts on all these topics fell to almost zero,” wrote Benjamin Guinaudeau, a professor at Laval University in Quebec, and his seven co-authors. “Total views dropped directly after TikTok went down, then started to rebound.”

Although cries of systemic, top-down political censorship do not appear to be supported by publicly available data, researchers say it’s still possible that TikTok’s new owners have begun reconfiguring content rules.

“It may be that a small number of messages were deleted or banned in a way that is not visible in overall trends,” wrote the researchers, who added that users who saw the word “Epstein” being blocked in private direct messages could not be studied because that data is not accessible.

Part of the challenge of studying TikTok, academics note, is that the platform does not grant the kind of access to researchers required to conduct comprehensive analyzes of how content moderation takes place — what is amplified, what is removed, and what priorities or policies may be driving these trends.

“Our position is that TikTok and other platforms should allow third-party researchers to study their recommendation systems and look for evidence of undue political influence,” the researchers wrote.

The timing of TikTok’s disruption struck a chord, as many users expressed wariness about how Ellison, a staunch ally of President Trump, could remake the app according to his vision, just as the Ellison family overhauled CBS in an effort to appeal to conservatives.

Besides Ellison’s Oracle, a cloud computing and data center giant, TikTok’s new investors include Silver Lake, a major private equity firm, and prominent investment firm MGX. ByteDance, TikTok’s Beijing parent company, will retain a minority stake in the new U.S. entity, while owning the powerful algorithm, which will be retrained using Americans’ data.

A TikTok spokeswoman said no changes had been made to its algorithm since the new investors took the reins of the social media company’s U.S. operations. The deal was reached to bring the app into compliance with a federal law requiring TikTok to distance itself from its Chinese parent company for national security reasons.

“Right now, TikTok can say pretty much anything about algorithm changes and we can’t verify it,” Guinaudeau told NPR.

He added: “Now we might see massive changes, like they suddenly stopped showing all political content, which was one of the accusations we look at in the article. But until they make more comprehensive data available to researchers, it’s almost impossible to detect subtle changes in their ‘For You’ recommendation system (“the algorithm”).”