A post can make you react without making you want another one like it. In a February 2023 experiment, Twitter's ranked political posts were more hostile than those from a time-ordered feed, and users were less likely to ask for more of them.
The researchers' account, published by the Knight First Amendment Institute in January 2024, tested the gap between a system designed around engagement—users' reactions—and what readers said they wanted to see.
Two ways of choosing the next ten posts
The team recruited active US Twitter users through an online crowd-work platform. For each participant, it collected the first ten posts the ranking system would have shown and the first ten from accounts that person followed, ordered by recency.
Participants rated the posts in a randomized order. There were 806 distinct people and 1,730 participations, because people could return in later study waves.
More hostility did not mean more satisfaction
Compared with the time-ordered feed, the system's political picks were more partisan and hostile toward the other political side. Readers also reported feeling worse about that side after seeing them. They were less likely to say they wanted more political posts like the ones the system had chosen.
An exploratory alternative, ranking posts by readers' stated preferences, reduced angry and hostile content. But it also risked reinforcing echo chambers: feeds that expose people mainly to views similar to their own.
This was a snapshot of Twitter, before X
The participants were younger and more Democratic than the national survey sample the authors used for comparison. The experiment examined small sets of posts, not the effects of using the service for months, and it predates Twitter's change to X.
It does not tell us how today's feed performs. It does show why a system's success at eliciting reactions should not be mistaken for evidence that readers want what it selects.