The European Commission has issued preliminary findings alleging that TikTok’s For You recommendation system breaches the EU’s Digital Services Act (DSA), focusing on the risk that a highly personalized feed can repeatedly serve a user more of the same potentially harmful material.
The concern is not simply that TikTok recommends videos. Recommendation is the core organizing function of a feed built for vast volumes of short-form content. The Commission’s case is instead about how the signals behind recommendations may reinforce a pattern of viewing—and whether TikTok sufficiently assessed and reduced the risks that pattern could pose to wellbeing.
Its findings highlight content concerning sadness, isolation and body image as examples where repeated exposure may be especially consequential. The regulator describes these escalating sequences as “rabbit holes”: a user engages with a theme, receives more videos resembling it, interacts again, and is then presented with still more of that material. The Commission says this can be detrimental to mental health, particularly for vulnerable adults and minors.
What the Commission is alleging
The preliminary finding says TikTok did not adequately evaluate the way its For You Page could lead people into these recommendation loops or sufficiently mitigate the resulting risks. That distinction matters. The DSA analysis described here is not an allegation that every difficult topic should disappear from a social platform. It is a challenge to the platform’s risk assessment and the safeguards surrounding the design of a personalized feed.
In practical terms, a recommendation system turns user signals into guesses about what to show next. Signals can include explicit actions, such as selecting an option that indicates the user is not interested in a video, as well as behavioral activity, such as interacting with posts or continuing to watch. The Commission’s concern is that the latter can create a powerful feedback loop when the platform interprets attention as a reason to continue recommending closely related content.
That is why viewing time is central to the preliminary findings. The Commission says TikTok should stop using viewing time as an indicator for recommending content. It also says the service should place more weight on a user’s explicit preferences—choices a person deliberately makes—rather than signals inferred from passive viewing behaviour.
The regulator also says that selecting “not interested” may not be enough to meaningfully alter what arrives in a person’s feed. That is an important usability issue. A control is most useful when people can locate it, understand what it does and see it reliably affect their experience. If a user believes they have rejected a category of content but continues encountering similar posts, the control may offer far less practical agency than its presence suggests.
What could change on the For You Page
The Commission has set out several broad changes it believes are needed. It says TikTok should:
- rely more heavily on clearly expressed user preferences;
- disable viewing time as a recommendation signal;
- make user controls easier to find and use; and
- regularly test the algorithm.
Those points aim at both the inputs and the oversight of the recommendation system. A feed can be personalized in many ways, but the Commission’s position suggests a model that treats time spent watching as an especially sensitive input when the result is repetitive, emotionally charged content.
Algorithm testing, in this context, should not be confused with a one-time technical check. The Commission’s request implies ongoing evaluation of whether recommendation behavior produces foreseeable harms, particularly for people who may be more susceptible to distressing or compulsive patterns of use. It is about examining the system’s outcomes, not only whether its software functions as intended.
This would be a meaningful design issue for any feed-based service. The challenge is not solved merely by adding a settings page. A person needs controls that are understandable at the moment they need them, and the choices made through those controls need to have a material effect on subsequent recommendations. The Commission’s comments indicate it believes TikTok has not yet cleared that bar.
A preliminary finding, not a final penalty
The Commission’s position is preliminary. TikTok has an opportunity to respond and address the concerns before any final conclusion is reached. If the EU ultimately confirms a breach and considers TikTok’s response insufficient, ByteDance, TikTok’s parent company, could face a fine of up to 6% of its annual worldwide revenue.
That maximum should not be treated as a declared fine or a guaranteed outcome. It describes the upper limit referenced in the process. The immediate significance is regulatory: TikTok is being told that the Commission sees the design and safety questions around its principal recommendation feed as serious enough to potentially violate the DSA.
TikTok disputes the allegations. The company says the Commission’s statements are inaccurate and mischaracterize the platform’s architecture and its approach to user safety. It also maintains that users have direct, intuitive controls over their content experience.
So the dispute turns on more than whether controls exist. It turns on their effectiveness, the importance assigned to particular recommendation signals, and whether TikTok’s measures adequately address the risks identified by regulators.
Why “rabbit hole” design is harder than a simple moderation problem
Content moderation usually brings to mind a decision about an individual post: whether it is permitted, removed or restricted. The Commission’s criticism is different. It is directed at the sequence created when many individual videos are arranged by recommendation logic.
A video may not be the central issue by itself, while the repeated delivery of similar videos can still be concerning. This is the difference between content-level policy and systems-level risk. The former asks what a specific item says or shows. The latter asks what experience the platform creates when it repeatedly selects, ranks and presents material to a particular person.
For users, this is a useful distinction because it explains why a “not interested” button is under scrutiny. A button may be an item-level response, but it must also work as a system-level instruction. When someone uses it, the feed should not merely hide one video while continually reconstructing the same theme through other accounts, sounds or formats.
The issue also carries relevance beyond social apps. Modern digital entertainment increasingly relies on personalization, retention metrics and recommendation systems. Discussions about the business incentives behind engagement-led design have appeared in games as well, including debates over free-to-play structures and loot boxes. The wider games-industry debate over access and engagement mechanics shows why clear user choice and meaningful safeguards remain important wherever product design is built around keeping people involved.
That comparison does not mean TikTok’s case is about video games, nor does the Commission’s preliminary finding establish a rule for every personalized service. But it underlines a wider question: when a system detects that someone is staying with a particular kind of experience, should continued attention automatically count as a request for more of it?
The wider investigation and pressure on platforms
The For You Page findings come from a broader EU investigation into TikTok that began in early 2024. The company has already faced other DSA-related scrutiny. In February, EU officials said TikTok breached the law over what they characterized as addictive design. In July, they alleged that the platform did not adequately protect the privacy of minors’ accounts.
Those separate actions do not automatically determine the result of the current recommendation-system case. They do, however, show that regulators are examining product design, youth protections and data-related safeguards together rather than as isolated concerns.
Scrutiny is also not limited to Europe. Google, Meta and TikTok have challenged a California law that would prevent social media companies from letting users under 16 access features characterized as potentially addictive. The companies’ legal challenge signals that the boundaries of age-based design rules and platform responsibility remain contested.
For ordinary TikTok users, the most immediate takeaway is not a confirmed redesign or a penalty. It is that features such as recommendation settings and “not interested” feedback are becoming part of a larger regulatory test: can users genuinely influence what a feed serves them, especially when their viewing habits may be read as a reason to keep escalating a theme?
For TikTok, the preliminary findings put the focus squarely on demonstrating that its controls work in practice and that its recommendation model is tested against foreseeable wellbeing risks. The Commission has outlined a direction—more explicit preference, less reliance on viewing time, more usable controls and regular testing. Whether that becomes a final legal requirement will depend on the next stages of the DSA process and TikTok’s response.
