President Donald Trump announced on September 19, 2025 a preliminary agreement for the sale of a majority participation in Tiktok of the Chinese technological giant bytedance to a group of US investors after Trump’s negotiation with Chinese leader Xi Jinping.
The agreement would create a new version of the application only for the US, putting it in accordance with a law signed by former president Joe Biden on April 23, 2024 and confirmed by the Supreme Court on January 17, 2025. The details of the agreement have not yet been completed, but some details are emerging. These include what will happen with the central algorithm of the application to share videos, and what that means for the millions of American users of Tiktok.
The Chinese government indicated that it will not allow Bytedance to sell the algorithm, because it is classified as an export of controlled technology, in accordance with Chinese law. Meanwhile, US technology industry executives and some legislators say that compliance with the law requires that the algorithm be under American control. The proposed agreement includes the algorithm license to continue to be Chinese intellectual property while the American version of the application continues to use technology.
Tiktok’s “For You” page algorithm is considered widely the most important part of the application. As an analyst said: “Buy Tiktok without the algorithm would be like buying a Ferrari without the engine.”
The value of the algorithm lies in its amazing ability to anticipate user content preferences. Many users claim that they know them better than they know themselves, a feeling that became a curious mixture of spiritual belief and conspiracy theory, as my colleagues and I have documented. Other academics indicated in a similar way that users feel more intimately seen and known by the Tiktok algorithm than those that drive other popular platforms.
I have studied the algorithms of social networks for almost a decade, exploring how our relationships with them evolve as they intertwine more and more with daily life. As a student of social networks and devotee from Tiktok, I want to shed some light on how the algorithm works and how it is likely to change the application following its sale.
How the Tiktok algorithm works
Somehow, Tiktok’s algorithm does not differ significantly from other social media algorithms. In essence, algorithms are simply a series of steps used to achieve a specific objective. They perform mathematical calculations to optimize production at the service of that goal.
There are two layers in the Tiktok algorithm. First, there is the abstract layer that defines the result that developers want to achieve. An internal document shared with The New York Times specified that the Tiktok algorithm is optimized for four objectives: “user value”, “value for the user in the long term”, “value for the creator” and “value for the platform”.
But how to turn these objectives into mathematics? What does an abstract concept mean as “user value”? It is not practical to ask users if they value their experience every time they visit the site. Instead, Tiktok is based on proxy signals that translate abstract results in quantifiable measures, specifically, I like it, comments, actions, followers, time spent on a certain video and other user behavior data.
These signals then become part of an equation to predict two specific specific results: “retention”, or the probability that a user will return to the site, and “time spent” to the application.
You may be interested: X of Musk loses the attempt to cancel the new content elimination system of India
The Tiktok algorithm “for you page” is based on automatic learning to predict retention and dedicated time. Automatic learning is a computational process in which an algorithm learns patterns in a set of data, with little or no human guide, to produce the best equation to predict a result. Through learning patterns, the algorithm determines how much individual data signals import to reach a precise prediction.
An investigation of the Wall Street Journal found that the amount of time that users spend seeing each video plays an important role in the way the algorithm chooses the videos that suggests to users. Using the equation that has generated to predict retention and time spent, the algorithm assigns a score to each video and classifies the possible videos that could be shown to the user by this score. The higher the score of an individual user, the more likely it will be that the video appears in its feed.
Of course, the characteristics of the content and other users also report the recommendations, and there are other subprocesses included in the equation. This step is where algorithmic moderation usually comes into play. If a video looks like a bait of participation or has an excess of blood, for example, the content score will be penalized.
Which is likely to change for US users
The sale was not over, but the destination of the algorithm is focusing. According to the reports, the algorithm version only for the United States will train only with the data of US users. Users will not need to download a new version of the application to work on the modified algorithm.
Although the algorithm itself is the same as before, it is quite sure that Tiktok will change. I see two key reasons for change.
First, the population of users of the application proposed only in the US will alter the composition of the underlying data set that informs the algorithmic recommendations continuously. As the types of content reflect US preferences, values and cultural behaviors, the algorithm can be slightly different as new patterns learn.
Although users are more likely to stay with the application because they do not need to download a new version, not all users will choose to do so, especially if it is considered under the control of Trump’s allies.
According to the current agreement, Oracle Corp. and the US government would supervise the resentment of the algorithm. This agreement suggests that the US government may have a significant influence on how the application works.
The agreement would give an 80% participation to US investors, including 50% to new Oracle, Silver Lake and Andreessen Horowitz investors. These investors have connections with Trump, and an apparent provision of the agreement allows the United States Government to select a member of the Board.
These influences pose the possibility of a boycott of users and creators of the left similar to the previous boycots of Target to reverse the measures of Dei and Disney after the reverted suspension of Jimmy Kimmel. This can result in a population of users, and data, which reflect a closer environment of interests and ideologies.
Secondly, it is possible that the majority shareholders of the new application decide to adjust the algorithm, particularly when it comes to moderation of content. It is possible that the new owners wish to modify the guidelines of the Tiktok community according to their opinion on the acceptable and unacceptable discourse.
For example, the current Tiktok community guidelines prohibit erroneous information and work with independent facts verifiers to evaluate content accuracy. While Meta used to follow a similar approach to Instagram and Facebook, in January 2025 Meta announced that it would end their relationships with independent facts verifiers and loosen content restrictions. YouTube similarly relaxed its moderation of content this year.
With the reports that the US government would supervise the resentment of the algorithm, there is the possibility that not only the new investors, but also the government itself can influence how the content is prioritized and moderated.
The conclusion is that algorithms are very sensitive to context. They reflect the interests, values and visions of the world of the people who build them, the preferences and behaviors of the people whose data informs their models and the legal and economic contexts in which they operate.
This means that, although it is difficult to predict exactly what a Tiktok will be only in the US. It is safe to assume that it will not be a perfect mirror image of the current application.
*Kelley Cotter is an assistant professor of Science and Information Technology at Penn State.
This article was originally published in The Conversation
Subscribe to our YouTube channel and do not miss our content
