If your TikTok views fell this year and someone told you it was the new American algorithm, they were guessing. Eight and a half months after TikTok USDS Joint Venture took over the app's US operations in January, nobody outside the venture has shown that the For You feed's recommendations, creators' reach, or their payouts changed because of it. The last public word I could find from the company on the algorithm came in February, when NPR reported a TikTok spokeswoman saying no changes had been made. Since then its own newsroom hasn't said whether it has carried out its most specific promise, and nobody outside can check, partly because the terms of the creator program it runs keep participating creators from publishing their dashboard metrics.
That promise sits in the venture's founding announcement. Owned 15 percent each by Oracle, Silver Lake and the Abu Dhabi investment firm MGX, with ByteDance, TikTok's Chinese parent, keeping 19.9 percent, the venture said it "will retrain, test, and update the content recommendation algorithm on U.S. user data." It gave no date, so there is no missed deadline. It also took "decision-making authority for trust and safety policies and content moderation" and promised "transparency reporting."
I searched all 13 posts on the venture's own newsroom, January to October. "Retrain" appears once, in the founding announcement. Outside that announcement, "algorithm" appears twice, both in the stock phrase "algorithm security." That covers its newsroom, not everything it may have told reporters.
The January dip hit recipes too
The loudest charge came early. On January 26, California Governor Gavin Newsom, a Democrat, announced a review into whether TikTok was censoring content critical of President Trump, and his office wrote on social media, CBS reported, that it had "independently confirmed instances" of suppressed posts, alongside a screenshot that appeared to show a blocked message reading "epstein." In a search on October 10, I found no published result of that review.
The broader charge, that political posts were being suppressed, is what Benjamin Guinaudeau of Université Laval, a Canadian university, and seven co-authors tested in an analysis published in February. They couldn't test private messages, since they "lack access to users' private chats," but they tracked posts about ICE, about Alex Pretti, a protester shot in Minneapolis, and other keywords; NPR reported the work drew on viewership data for more than 100,000 videos. Political posts did crater on January 25, but posts about recipes and the Oscars "dropped to almost zero" too, along with every other topic the researchers tracked, then recovered. The researchers called the outage "the more likely culprit." The venture blamed a power failure at a data center site it later said was "operated by Oracle," one of its three 15 percent owners, and told creators that zero view counts and missing earnings were "a display error caused by server timeouts."
That finding cuts both ways, and the authors say so. They "cannot definitively reject the possibility that the new owners changed TikTok's core recommender system," and they note that the data TikTok makes available isn't enough for a closer look at that system. "Right now, TikTok can say just about anything related to algorithm changes and we can't verify it," Guinaudeau told NPR.
A senator asked. In May, Senator Ed Markey (D-Mass.) wrote to the venture and to Oracle, and CNN reported that he said the group had not released enough information about how it is retraining the algorithm, asking for details by June 18. My searches on October 10 turned up no published reply.
The audience mostly stayed. By mid-February, Sensor Tower, the app-analytics firm, put US daily users at around 95 percent of their level in the week of January 19 to 25, a week that already included deal day and the outage, with daily time spent back to about 80 minutes. CNBC, which reported the figures, found "little sign of a mass exodus," though deletions spiked briefly after the disruptions.
The people with the numbers can't publish them
Creators would notice a change first. One reach-drop figure going around comes from a blog run by a company that sells creator tools. It says many US creators reported reach drops of 20 to 40 percent in February and March, and that the drops "were the retrained model going live." The post gives no survey, sample, or method.
One reason better numbers haven't surfaced is in the program's terms. Creator Rewards is TikTok's program that pays eligible creators based on views and engagement; to join, you must be 18 or older with at least 10,000 followers and 100,000 views in the past 30 days. I read its US terms, last updated July 20 and now issued by the venture. The metrics that decide rewards are "based solely on TikTok data," calculated at the venture's "sole discretion." The reporting metrics it gives participants, and other non-public program information, "shall be kept strictly confidential by Creator and not published or shared with any third party," except with legal or professional advisors. Public view counts aren't covered; the researchers used them. The participating creators best placed to see a change have agreed not to publish their dashboard metrics. I couldn't check whether that clause predates the deal, because the Internet Archive was offline when I tried, so read it as how the program works now, not as something the new owners added.
Keep your own screenshots
If you earn on TikTok, keep your own records. A monthly screenshot of your analytics, kept privately, is the only before-and-after you control, and the terms bar publishing those metrics, not keeping them. Don't pay for a course that promises to beat a new algorithm; nobody has shown there is one. And if most of your income runs through a dashboard you can't compare in public, keep a second platform warm.
The one feed change the venture has announced came in February: a Local Feed tab that shows posts based on location, topic, and recency, with an optional precise-location setting that is off by default. If you run a restaurant or a shop with a local following, it's worth testing.
The venture's first transparency reports, published in July, cover content moderation and platform security. I searched both, and neither mentions retraining, recommendations, or reach. As of October 10, a report on what the feed actually recommends is the one it hasn't published on its reports page.
