Users taking back feeds
Princeton researchers tested what happens when social‑media users get more control over their feeds instead of leaving ranking entirely to engagement‑maximising algorithms. Their findings offer a sharper way to talk about platform mediation in interviews—showing that algorithms shape attention but user choices and experience also matter (engineering.princeton.edu).
Social media feeds are usually ranked by software that predicts what will keep people clicking. Princeton researchers tested what changed when users built those feeds themselves. (engineering.princeton.edu) The Princeton team built a tool called Bonsai and tested it on Bluesky, a decentralized social network with an application programming interface that outside developers can use. The study will be presented at the Association for Computing Machinery Conference on Human Factors in Computing Systems in Barcelona on April 16, 2026. (cs.princeton.edu) Bonsai starts with a plain-English prompt about the feed a person wants, then uses a natural language model to suggest accounts, hashtags and existing feeds. Users can edit those suggestions and add more sources before the system assembles a personalized timeline. (engineering.princeton.edu) The researchers recruited 15 Bluesky users and studied them over an average of 12 days. Most participants said they liked curating their own feeds, but they also said building one from scratch took work. (cs.princeton.edu) That trade-off sits inside a larger fight over recommendation systems, the software that selects and orders posts for each user. Arvind Narayanan of the Knight First Amendment Institute wrote in 2023 that these systems determine, “in large part,” who hears what people say online. (knightcolumbia.org) Recent research has found that feed ranking can change behavior and attention in measurable ways. A 2026 Nature field experiment on X found that moving active United States users from a chronological feed to an algorithmic one for seven weeks increased engagement and shifted some political attitudes in a more conservative direction. (nature.com) The Princeton study points to a different question: not only what platform algorithms do to users, but what users do when they get more agency over ranking. Marianne Aubin le Quéré, a Princeton postdoctoral researcher and co-author, said participants “felt liberated from the algorithm,” even as the system demanded more cognitive effort. (engineering.princeton.edu) The tool also depends on platform structure. The researchers said Bonsai can work on decentralized services such as Bluesky or Mastodon because those networks expose the data needed for third-party feed tools, while more closed platforms generally do not. (cs.princeton.edu) Princeton researchers said one next step is making feed-building easier for casual users who do not want to spend time tuning every source. For now, their test suggests that taking back the feed is possible, but not effortless. (engineering.princeton.edu)
