I like getting random video recommendations. That's probably why I keep opening the app. But I've had evenings where I meant to watch one thing and somehow lost an hour. The EU thinks TikTok's recommendation feed can pul…
Wing Li@wingli
I chose Option A: limit recommendations by default. I am not asking to turn off personalized recommendations entirely. I would start with stronger limits on repeated recommendations around sensitive themes, while preserving a clear route for people to opt into a broader, more personalized feed. The principle I want is simple: a user's explicit choice to see less should carry more weight than the system's inference from watch time.
The documented case for changing the default is serious, although still preliminary. On 6 November 2026, the European Commission preliminarily found that TikTok's recommender system breached the Digital Services Act because of its "rabbit hole effect." The Commission says the "For You" feed continuously serves personalized videos and recommends more similar content as users interact with a topic. It identified particular concern around mental-health struggles, sadness, isolation and body image, including the possibility of increased sadness or anxiety, lower self-esteem, and worsening eating disorders or mental conditions. These are preliminary findings, not a final ruling, but they establish a concrete risk rather than a hypothetical one.
The existing controls do not adequately answer the problem I am raising. In its preliminary assessment, the Commission found that TikTok's mitigation measures were not effective, including diversified-content measures that did not sufficiently reduce the frequency of similar recommendations. It also found that using the "not interested" button may not change the feed because the algorithm prioritizes viewing time over the user's choice to see less. My view is that a control is not meaningful if the underlying recommendation system can continue overriding it. Optional controls should accompany stronger defaults, not substitute for them.
TikTok disputes that assessment. According to The Guardian, the company says its policies prohibit content promoting disordered eating and risky weight-management behaviours, and that more than 79 million EU users used its "not interested" feature or keyword filters during the first six months of 2026. A TikTok spokesperson also defended the platform's controls as direct and intuitive. These are the company's claims and should be weighed, but usage figures alone do not show that repeated recommendation loops are being broken. If users must actively filter material while the feed still reacts primarily to continued viewing, availability of a control is not the same as control over the recommendation.
My proposal is to make the safer version of personalization the starting point. When repeated engagement with a sensitive theme is detected, the feed should reduce repetition and surface alternatives rather than interpret every additional view as a request for more. It should also give an explicit "not interested" decision priority over passive engagement and make opting into broader recommendations a deliberate action. This is my recommendation, not a Commission ruling, although it reflects the Commission's examples of relying more on explicit preferences and reducing the role of viewing time. The burden should be on the system to prove that continued escalation reflects genuine intent, rather than asking users to escape a loop one report at a time.
I would rather personalized feeds begin with restraint and let people request more than begin with maximum engagement and require people to request less. Personalization can remain useful without treating persistence, discomfort or repeated exposure as affirmative consent.
Note: this post was drafted with AI assistance; I checked each claim against the linked sources and take responsibility for the content.