A social feed is not a list of what was posted, but a ranking of a small selection predicted to hold one particular person. Understanding the prediction explains most of what feels strange about feeds.
Ranking replaces chronology
Early feeds showed posts in the order they arrived, which failed as the volume of available material grew far beyond what anyone could read.
Ranking systems select a candidate set and then score each item by how likely this viewer is to watch, react or share, ordering by that score.
Nothing in the process evaluates whether an item is accurate or worthwhile. The score estimates behavior, and behavior is what gets optimized, which is the source of most complaints about feeds.
The first seconds carry most of the weight
For video, the strongest early signal is whether viewers keep watching past the opening moments, since that is available almost immediately after posting.
Creators respond by front-loading the most arresting element, which is why so many clips open mid-action and explain themselves afterward.
Completion and rewatch rates then refine the estimate, rewarding short items that are watched twice over long ones abandoned halfway. Length is therefore a strategic choice rather than an editorial one.
Distribution expands in stages
A new post is typically shown to a small audience first. If it performs against expectations for that audience, it is shown to a larger one.
Each stage is a test, and most items stop advancing at an early stage. The rare item that keeps clearing thresholds is what appears to spread suddenly.
Because the earliest audience is small, the outcome is sensitive to chance, which is why near-identical posts can have wildly different results.
Negative signals count too
Systems track hiding, reporting, skipping quickly and marking content as uninteresting, and these reduce a score as surely as positive reactions raise it.
That is a partial check on material that provokes reaction without satisfying anyone, though it works only if enough viewers use those controls.
Platforms also apply policy layers that suppress categories of content independently of engagement, which operates separately from the ranking itself.
The system reflects the viewer back
Because predictions are built from an individual's history, two people following identical accounts receive substantially different feeds.
This makes disagreements about what is happening online genuinely hard to resolve, since the participants are describing different sets of material.
The feed is a model of a person, and it becomes more confident the longer it is used without interruption. Deliberately watching outside a habit is the only practical way to widen it.