Bluesky was conceived as a rebuttal to the black-box logic of mainstream social media: instead of locking users into a single hidden ranking system, it offers a reverse-chronological default feed and a marketplace of custom feeds that let people “choose their algorithms.” That architecture matters because it turns Bluesky into something rare in platform research: a live social network where different ranking systems can be compared more transparently, rather than merely guessed at from the outside. (bsky.social)
The most striking evidence arrived in a Nature study published on May 27, 2026. Researchers randomly assigned 2,000 participants on Bluesky to different feed designs for eight weeks spanning the 2024 U.S. presidential election. Compared with a reverse-chronological feed, an engagement-based feed amplified “intergroup, moralized and emotional” content, increased exposure to toxic material, and most strongly boosted moral outrage and political content. Just as importantly, it distorted users’ perceptions of social norms and increased perceived partisan animosity. In other words, the algorithm did not merely show people politics; it made politics look angrier, harsher, and more combative than it otherwise would have appeared. (nature.com)
What makes this result especially revealing is that Bluesky is not, in broad ideological terms, a perfectly balanced miniature democracy. A 2025 PLOS One study of roughly five million users found the platform to be largely left-of-center, with very few users on the right, while a 2025 arXiv study reported that about 13% of posts were political and that several topics still displayed high structural polarization. So the lesson is subtler than “algorithms create division out of nothing.” Rather, they can seize on a relatively small amount of conflict, over-represent its most incendiary edges, and make that fringe intensity feel socially normal. (journals.plos.org)
The experiment also offered a hopeful counterexample. A redesigned “diversified extremity” feed reduced exposure to polarizing and toxic content, improved the accuracy of users’ norm perceptions, and did so without reducing platform enjoyment. That finding is crucial. Earlier research on Facebook and Instagram found little short-term effect from switching to chronological feeds, whereas a 2026 study on X found that algorithmic ranking could shift some political attitudes. The Bluesky experiment suggests a plausible mechanism linking these debates: algorithms may intensify polarization not only by changing what we believe, but by warping our sense of what everyone else believes—and how hostile public life supposedly is. (nature.com)
