Breaking the Loop: Exploring User Agency in Escaping Algorithmic Rabbit Holes on Social Media

This brief presents early-stage results from a research collaboration between Panoptykon and Piotr Sapieżyński (Northeastern University), building on Panoptykon's earlier "Algorithms of Trauma 2" case study, which had found that clicking Facebook's "Hide post – See fewer posts like this" button 122 times over two months did not reduce a user's exposure to distressing suggested content.

A first single-participant pilot in August 2025 replicated that finding: over 100 clicks on Facebook's "Not interested" button produced no meaningful or lasting reduction in unwanted suggested content. A second, larger study with seven participants in December 2025 used a refined methodology, clustering each participant's emerging content "rabbit holes" before they chose what to mark as unwanted, and found a real but highly uneven effect: unwanted content fell by 38% on average, with a statistically significant drop for five of seven participants but an increase for one. The brief also cites a forthcoming, not-yet-published TikTok study describing a "relapse effect," where content a user tried to suppress gradually creeps back into their feed.

Panoptykon argues that feedback tools which are this inconsistent or slow to take effect may put platforms in breach of several DSA provisions: Article 27(3) on recommender-system transparency, read together with Articles 25 and 28 on dark patterns and minors' protection, and the systemic-risk mitigation duty under Article 35, especially since platforms themselves cite these same tools as risk-mitigation measures in their DSA risk assessments without providing data on whether they work.