Why your feed shows you what makes you angry

Published 6 October 2026 · Updated 6 October 2026

Your feed is not a window onto the people you follow. It is a selection, and the selection leans angry.

Three studies show how the loop works. Posts about the other side get shared most. A feed ranked by engagement picks more of them than a plain, time-ordered one, though people say they want fewer. And the likes on an angry post nudge its author to write another.^{1, 2, 3}

Anger at the other side travels

Researchers looked at 2.7 million posts by US news outlets and members of Congress. Posts about the other political side were shared about twice as often as posts about their own side. They also drew more of Facebook’s “angry” reactions.¹

The ranking picks it

In 2023, 806 Twitter users let researchers compare two versions of their timeline. One was the first ten tweets the ranked feed would show them; the other, the ten most recent from accounts they follow. The ranked feed picked angrier tweets, more hostile to the other side. Readers felt worse about the other side after reading them.²

Asked tweet by tweet whether they wanted to see more like it, people rated the ranked feed’s picks slightly higher overall. For political tweets, they rated them lower. The authors conclude that the ranking seems to give people what they engage with, not what they say they want, “particularly when it comes to political content”.²

The likes reward it

When a Twitter user’s outraged posts drew more likes and retweets than usual, they posted more outrage the next day. Doubling the feedback went with only 2 to 3% more outrage. But the authors note that an effect like this “can easily scale” across a network.³

One caveat. These are studies of US politics on Twitter and Facebook, and the effects are modest. The second study’s authors also point to a 2023 study in which Facebook’s and Instagram’s ranking made no measurable difference to how polarised users felt.²

Try this

For the people you most want to hear from, use something where nothing is ranked and nothing is counted. Then a note from your sister is not competing with a stranger’s argument.

The group letters in the directory are a good place to start: no ranked feed, and no likes to count. Every card in the directory has a Feed row and a Likes row that say how each product handles this. We make one of them, DearAll, and it is marked as ours.

Sources

Steve Rathje, Jay J. Van Bavel and Sander van der Linden, "Out-group animosity drives engagement on social media", Proceedings of the National Academy of Sciences 118(26), 2021, e2024292118, doi:10.1073/pnas.2024292118 (open access; full text read on Europe PMC, PMC8256037). An analysis of 2,730,215 Facebook and Twitter posts by US news outlets and members of Congress, mostly from 2016 to 2020. Each word about the other side raised the odds of a share by about 67%. Correlational, about elite accounts rather than ordinary users, and from the years before the 2020 US election; the authors say it is unclear how the platforms' algorithms contributed. https://doi.org/10.1073/pnas.2024292118
Smitha Milli, Micah Carroll, Yike Wang, Sashrika Pandey, Sebastian Zhao and Anca D. Dragan, "Engagement, user satisfaction, and the amplification of divisive content on social media", PNAS Nexus 4(3), 2025, pgaf062, doi:10.1093/pnasnexus/pgaf062 (open access, preregistered; full text read on Europe PMC, PMC11894805). An audit of 806 US Twitter users in February 2023, comparing the first 10 tweets of each person's ranked timeline with the 10 most recent from accounts they follow. The ranked picks were more partisan and more hostile to the other side (both by 0.24 of a standard deviation), and for political tweets anger was the emotion most amplified. Users rated the ranked picks slightly higher overall (0.06) and lower for political tweets (−0.18). One moment in time, the first 10 tweets only, and a sample younger and more Democratic than Twitter's users. The authors note that a 2023 study of Facebook's and Instagram's ranking found no significant change in how polarised users felt after two weeks to three months. https://doi.org/10.1093/pnasnexus/pgaf062
William J. Brady, Killian McLoughlin, Tuan N. Doan and Molly J. Crockett, "How social learning amplifies moral outrage expression in online social networks", Science Advances 7(33), 2021, eabe5641, doi:10.1126/sciadv.abe5641 (open access, preregistered; full text read on Europe PMC, PMC8363141). Two observational studies of 7,331 Twitter users and 12.7 million tweets, with outrage spotted by a machine-learning classifier, plus two online experiments with 240 people. Doubling the likes and retweets on outraged posts went with 2 to 3% more outrage the next day, an effect the authors call small but able to "easily scale". The Twitter studies are observational: the authors "cannot draw causal inferences". https://doi.org/10.1126/sciadv.abe5641