Too many posts, too little time
Anyone who follows a few hundred accounts, joins some groups and receives posts from advertisers has far more material available than they could read in a day. A feed is the answer to a practical problem: out of everything that could be shown, what goes first, and what is left out?
Most large social networks answer with a ranking system. Instead of showing every post in the order it was published, the system scores posts and arranges them so that the ones it expects you to care about appear near the top. The details are private and change often, so this article describes the general idea rather than any one service’s formula.
The basic steps
Although every service does it differently, the process is commonly described in a few stages.
- Gathering candidates. The system collects posts that could be shown: from accounts you follow, groups you belong to, and sometimes accounts you do not follow but that are similar to ones you engage with.
- Filtering. Posts that break the service’s rules, that you have muted or hidden, or that are duplicates are usually removed at this point.
- Scoring. Each remaining post gets one or more predicted scores, such as how likely you are to stop and read it, react to it, reply, share it or hide it.
- Ordering. The scores are combined, often with other goals such as variety or limits on how many posts from one source appear in a row, and the result becomes your feed.
The word “predicted” matters. The system does not know what you will do; it estimates, based on patterns from your past behavior and the behavior of many other people.
Signals: the clues feeds use
A signal is a piece of information the system can measure. Commonly mentioned categories include:
- About the post: its format (text, photo, video, link), its length, how recent it is, and how other people have responded to it.
- About the author: how often you interact with them, whether you have looked at their profile, and whether they are a close contact or a distant one.
- About you: what you tend to watch, skip, save or comment on, how long you pause on certain kinds of posts, and the time of day you usually use the service.
- About the context: the device you are using, your connection speed, or which part of the app you are in.
Much of this relies on data collection, which is how a service learns patterns in the first place. Some signals are explicit, like a like or a “show me less of this” choice. Others are implicit, like pausing on a video without tapping anything. Because these systems also determine who sees paid posts, the same signals matter for audience targeting in advertising. Services often weigh a deliberate action more heavily than a passive one, but the exact weights are not public.
Why two people see different feeds
Even if two people follow exactly the same accounts, their feeds will usually differ, because the scores are personal. One person has often watched video from a certain account, so its videos rank high for them. The other tends to skip them, so the same posts sink. Timing matters too: a feed viewed at breakfast may be built from different candidates than one viewed at midnight.
Other differences come from choices outside your behavior. Your language, your region, the groups you have joined and the settings you have changed all alter the candidate pool. Experiments run by the services themselves also mean that some people see a slightly different version of the ranking than others, often without being told.
This is part of why public conversations about “what social media shows people” are hard to settle. There is no single feed to point at, only millions of personalized ones.
Ranked versus chronological
Many services let you switch between a ranked feed and a simple newest-first list, at least for accounts you follow. Each has trade-offs.
A chronological feed is easy to understand and predictable, but it favors people who post often and can bury a post that arrives while you were asleep. A ranked feed can surface things you would have missed, but it is harder to see why anything appears, and it can reward whatever the system measures best, which is not always what is most useful to you.
For creators, none of this makes a post’s success predictable, which is the point of the idea of viral luck. Neither is automatically better. Which suits you depends on how many accounts you follow, how often you check, and whether you value completeness or relevance more.
What ranking does not tell you
Because a feed is built to predict engagement, what appears near the top is not a measure of truth, importance or popularity in the wider world. A post may rank highly because it provokes strong reactions, which is a reason to keep in mind the idea of echo chambers: if you mostly respond to one kind of material, you may be shown more of it. Researchers continue to study how strong that effect really is, and findings differ by service and by person, so it is best treated as a possibility rather than a certainty. Strong reactions can also help false claims travel further than careful corrections, although how much depends on the topic and the audience.
Likewise, if you run an account, a drop in views does not by itself prove that you have been “penalized.” Audience interest, changes to the system and plain variation can all move numbers. The difference between organic reach and what you pay for is a separate topic, but it is a reminder that visibility is partly in the hands of the service.
Practical things you can do
- Use the controls that exist: mute, hide, “see less,” unfollow and list features all feed back into the ranking.
- Notice what you linger on. Passive signals count, so what you stop to watch may shape what you see next.
- Check the settings for an option to see posts in order of publication, if offered.
- Search directly for people and topics you care about instead of relying on the feed to deliver them.
- If you manage an account, treat reach as variable. Looking at social analytics over weeks, not single posts, gives a calmer picture than reacting to one result.
A system that keeps changing
Ranking systems are adjusted constantly, partly in response to user complaints, partly because of business goals and partly because of outside rules. Something that held true about a feed last year may not hold now. That is why the most durable advice is general: a feed is a prediction machine, it works from the signals your behavior provides, and it shows each person a version of the service that no one else sees in quite the same way.