Published on 2026-07-25
How does your feed actually decide what to show you first?
Feeds rank posts by predicted engagement, not chronology or importance. Here is what the score actually measures, and why that matters.
Open any app and scroll for ten seconds. Something stops your thumb: a video mid-argument, a stranger's rant that somehow feels aimed at you. Meanwhile a friend's ordinary update from yesterday sits three screens down, unseen. Nobody at the company decided your friend's life mattered less. A ranking system did that, and it finished the job in a fraction of a second, before you even opened the app.
Chronology lost its job years ago
Feeds used to work the way a diary works: newest entry on top, oldest at the bottom, no judgment involved. Most major platforms dropped that model because a plain timeline is boring compared with a feed sorted for you. What replaced it is a score. For every candidate post, a model estimates how likely you are to do something with it: pause on it, watch it through, comment, share it, come back to it. Posts with a high predicted score go first. Posts with a low one might never surface at all.
The model learns from you specifically. Every scroll you speed past, every video you rewatch, every comment you leave becomes a data point about what holds your attention. The next batch of posts gets sorted using that pattern. It isn't reading your mind. It's running a very fast, very personal experiment on you, over and over, and updating its guess each time you react.
What the score is actually chasing
This is where it gets interesting: the score was never built to measure truth, or importance, or even how much you'll enjoy something afterward. It measures action in the moment. Researchers at Oxford's Reuters Institute, reviewing how platforms like Facebook rank posts, found the system favours updates that can be shared and liked, often content built to produce, in their words, either a sense of oversize delight or righteous outrage. Calm, accurate, unremarkable information does not perform nearly as well against that yardstick, and the ranking doesn't care.
- how long you paused on it before scrolling on
- whether you replayed, reread, or zoomed in
- whether you commented, reacted, or sent it to someone
- how closely it resembles whatever already held you last week
None of those four signals asks whether a post is accurate. They ask whether it works on you. A well sourced update about a policy change and a wildly overstated rumour about the same policy can carry the same weight in the model's eyes, right up until one of them makes people linger, argue in the comments, or pass it on, and the other doesn't. Guess what gets shown to more people next, and guess what gets buried before most of your friends ever scroll that far.
What happens when you turn the sorting off
In 2023, a research team led by Andrew Guess at Princeton, working with Neil Malhotra and Jennifer Pan at Stanford, ran a rare real world test of this. With Meta's cooperation, they gave a group of consenting Facebook and Instagram users a plain chronological feed for three months around the 2020 US election, instead of the usual ranked one, and compared what happened to a group left on the standard algorithm. People kept on the ranked feed spent 73 percent more time per day on the platform than a typical monthly user; the chronological group's extra time was 37 percent, roughly half as much. The sorting itself, not just being on the app, was doing real work to keep people scrolling.
The chronological switch changed what people saw, too: on Facebook it increased the share of political content in the feed and noticeably increased content from lower quality sources, while also reducing uncivil posts. It's one study, on two platforms, during one election season, and the authors themselves are careful about how far the findings travel. But the core contrast holds up against everything else we know about how these systems are built: an engagement based sort keeps people around longer than a neutral one does, because that is the exact thing it was tuned to do.
A feed that holds your attention has done its job. Whether it told you anything true was never the metric it was graded on.
Reading a feed that isn't reading for you
None of this makes the feed a villain. It's a sorting tool doing exactly what it was built to do, and most of what it shows you is perfectly ordinary and true. The useful shift is simpler than that: separate the fact that something held your attention from the question of whether it deserves your trust. Those are two different scores, judged by two different systems. One of them is the app's. The other one has to be yours, because the feed was never built to run it for you.