Social networks worked for decades according to a simple principle under which you saw what you had subscribed to. Now most platforms have switched to a different model, and your feed is formed not from your choice, but from a prediction about what will hold your attention. The difference between these two approaches is described as a transition from the social graph to the interest graph. In the first case, the system knows who you are connected to, while in the second it knows what interests you, and the second has turned out to be more accurate.

What Platforms Really Reveal

TikTok in its own materials describes three groups of signals that determine the content of the feed. The first and most important is your interaction with the content, that is, videos that you watched to the end, watched again, liked, shared, or skipped. The second is information about the video itself, including captions, hashtags, sound, and text on the screen, and it is this information that allows the system to understand what the video is about. The third is device and account settings, that is, language, country, and device type, and the platform directly states that this group weighs less, since users do not consciously express their preferences through it. The key wording from the company itself is that a strong indicator of interest, for example watching a long video in full, receives more weight than a weak one.

The Main Principle That Explains Everything Else

From this follows the most important thing that is worth understanding about any algorithmic feed.

The system believes your behavior, not your beliefs. You may think that you are interested in science, but if you stop at scandalous videos and watch them to the end, the algorithm will draw a conclusion from your actions, not from your self-assessment. That is why likes weigh less than watch time. It is easy to like something and it is a conscious gesture, while a person does not force themselves artificially to watch a video to the end, so this signal is considered more honest. The same principle explains the most unpleasant effect of such systems. Stopping on content that annoys you is read by the algorithm as interest, since it measures attention, not satisfaction.

About the Numbers That Are Circulating Online

Numerous publications give exact weights, for example forty percent for engagement signals, thirty-five for video content, and twenty-five for settings. Platforms have never published such figures, and the sources themselves are accompanied by a mention that these are estimates based on testing by content creators. Claims about specific thresholds after which a video becomes viral, or about the exact number of views in the first test group, are also widespread. These should be treated as guesses, not facts, since the exact parameters are a trade secret and change constantly.

The Closed Loop

The most interesting thing about these systems is that they do not simply reflect your interests, but form them. The mechanism works like this. You stay on a certain type of content, the system shows you more of it, you see it more often and begin to consider this topic more important than you did before, after which you stay on it even more willingly. After several weeks, your feed describes not the person you were at the beginning, but the person the feed has made you. Here it is worth making an honest qualification, since research on so-called filter bubbles gives mixed results, and some studies show that people online encounter different views more often than in offline life. That is, the effect is real, but it is weaker and more complicated than it is usually described.

Why Emotional Content Always Wins

Since the system optimizes for attention retention, it naturally gives preference to what holds that attention best. Emotionally charged material that causes outrage, fear, or strong surprise holds attention better than calm and balanced material, and this is not the platform's intention, but a consequence of the chosen metric. For Ukraine, this mechanism has a specific dimension, since it is precisely what the spread of disinformation is built on. Material created for manipulation receives amplification from the algorithm automatically because it causes a reaction, and the reaction is the same signal of interest. I wrote separately about how to recognize such content in a material about checking information.

How to Take Back Control

Use the button indicating that you are not interested, since this is a direct signal that the system takes into account quickly. Simply scrolling past is not enough, because the very fact of stopping is already being read. Do not linger on content that annoys you, even to write an angry comment, because this is the strongest signal of interest of all possible ones. Review and clear your viewing and search history, since it is the basis of the profile. Most platforms have separate settings where the history can be deleted or its recording can be disabled. Reset recommendations completely where such a function exists, and then the system will start building the profile again. Switch to the subscriptions feed instead of the recommended one, since most platforms retain this option, although they hide it further away.

Subscribe consciously to what you want to see, and unsubscribe from what you consume automatically.

Use search instead of the feed when you need specific information, because this moves you from consumption mode to intent mode. Review the advertising preference settings in your accounts, where you can see which interests have been assigned to you, and this list can be edited.

About Time

Optimization for attention retention means that the system has no incentive to let you leave the application, since its effectiveness is measured precisely by how long you stayed. This does not make platforms evil, but it means that you will have to set the boundaries yourself, because there are no boundaries inside the system by design. The simplest tool here is screen-time limits built into both mobile systems, and they work better than self-control precisely because they do not depend on your state in the evening.


Sources: materials from the TikTok Transparency Center on recommendations, industry publications on content ranking