On Nov. 6, 2026, the European Commission issued a preliminary finding that TikTok’s For You recommendation system breached the EU’s Digital Services Act (DSA). The Commission focused on TikTok’s assessment and mitigation of risks from repeated recommendations, which it said could draw users into harmful “rabbit holes.” The finding concerned the For You feed, not algorithmic feeds generally. The Commission’s notice described its position as preliminary.
Why the Commission raised concerns about repeated recommendations
The Commission said TikTok may recommend more videos similar to a type of content after a user interacts with it. It identified repeated exposure to material about sadness, isolation or body image as a potential risk, especially for vulnerable adults and minors. Its preliminary finding covered risk assessment and mitigation measures.
The concern is about a possible pattern: as similar recommendations accumulate, a user may keep seeing related material. The Commission’s examples describe risks it identified, not a claim that every user experiences harm.
Changes the Commission reportedly sought
Euronews reported that the Commission sought to give users’ explicit preferences more weight, stop using viewing time as a recommendation signal, make feed controls easier to access and test the algorithm regularly. These details were reported as proposed changes; they were not listed in the Commission notice’s summary of its preliminary finding. Euronews’ report also said the Commission considered existing diversification measures inadequate and questioned whether “Not interested” sufficiently changes the feed.
TikTok rejected the Commission’s assessment
A TikTok spokesperson told Euronews that the Commission’s statements were inaccurate and misrepresented the platform’s architecture and safety commitments. The Commission’s published finding remained preliminary, rather than a final adjudication.
What a 2024 audit observed about personalization
A separate study published by auditing TikTok on Aug. 26, 2024, examined feeds for researcher-created profiles with predefined interests. After about an hour of scrolling, up to 80% of videos matched those interests. Videos appearing across multiple profiles were most numerous in the first 15 minutes and became less common later. The study’s interest classifier was trained on 688,000 videos, and its authors reported 95% precision. The audit’s account describes personalization in that test setup; the Commission’s concern addressed potential risks from repeated exposure to particular themes.
A measure of personalization can show how closely a feed follows preset interests; it does not, on its own, answer whether repeated recommendations create a risk.
