Facebook News Feed ranking

Meta’s official Transparency Center documentation confirms a user’s relationship closeness to whoever posted, whether it’s a close friend, family member, or a followed Page, feeds directly into ranking, and that this factor was deliberately strengthened following the 2018 pivot toward Meaningful Social Interactions. This affinity signal is one of the clearest through-lines from the earliest EdgeRank formula, where it was literally one of three multiplied terms, all the way to the current ML-driven system.

Confirmed by both Meta’s official documentation and the 2021 leaked internal documents as a core Meaningful Social Interactions signal, this is the model’s prediction of how likely a post is to generate a comment or reshare, treated as a strong positive boost. The signal’s use is confirmed directly by Meta; the specific numeric weight it carries relative to other predictions is known only through the leak, which is why this row is graded Confirmed for use but Speculative for the exact weight.

Internal leaked documents, corroborated by CNN’s independent reporting, revealed that emoji reactions historically carried substantially more ranking weight than a simple like, commonly cited as roughly five times as much. This reflected an explicit design choice by Meta’s own engineers: a reaction requires slightly more deliberate engagement than a like, and the system was built to reward that extra effort with a proportionally larger boost.

The angry reaction has an unusually well-documented history: internal leaked documents show it was initially weighted at 1.5 times a like, then Meta officially confirmed reducing its ranking weight to zero in 2020, following internal debate visible in the same leaked materials about whether rewarding anger was actively harmful. This is one of the few individual signal weights Meta has ever effectively confirmed, even if only implicitly, by publicly announcing the change rather than the original weight itself.

Confirmed by both official Meta documentation and the leaked internal materials, basic engagement in the form of comments and likes remains a genuine ranking input, with comments consistently treated as a stronger signal than a simple like, consistent with the platform’s broader pattern of rewarding higher-effort interactions more heavily. This is the most foundational engagement layer beneath the more sophisticated MSI predictions layered on top of it.

Leaked internal documents revealed a mechanism sometimes called downstream MSI, where engagement occurring further down a chain of reshares, not just on the original post, was historically factored into ranking, amplifying content that spread virally through multiple reshare hops. This mechanism was later curtailed according to the same leaked materials, though the specifics of that curtailment and its current status are not confirmed by any official Meta statement.

Meta’s official documentation confirms that content format, video, photo, link, plain text, or Reels, is factored into ranking adjustments tied to current platform goals, meaning the relative treatment of different formats can shift as Meta’s broader product priorities change over time. This gives Meta a direct lever to push the platform toward whichever content type it’s currently trying to grow, independent of any individual post’s actual engagement quality.

Official Meta documentation confirms recency as a genuine ranking input, with fresher posts generally favoured, a direct descendant of the original EdgeRank formula’s explicit time-decay term. Even as the system has grown vastly more sophisticated with ML-driven predictions, this basic principle, that a post’s freshness matters, has remained a consistent, confirmed part of the ranking logic throughout.

Meta’s official documentation confirms a broader “integrity” layer that reduces the ranking of clickbait, misinformation, and borderline content, content that skirts policy violation without technically crossing the line, functioning as a quality-control counterweight to purely engagement-driven ranking. This is confirmed to exist as a real, deliberate mechanism, though exactly which content qualifies and by how much its distribution is reduced is described only in general, qualitative terms rather than specific published thresholds.

Meta officially announced in 2021 that political and civic content would be down-weighted by default across Feed, a confirmed policy decision reflecting the company’s broader post-2018 effort to reduce potentially divisive content in favour of more personal, friends-and-family material. This policy shift has itself been controversial, drawing criticism from publishers and civic organisations concerned about reduced visibility for legitimate news and activism.

Official Meta documentation confirms that explicit negative feedback, hiding a post, using “see less,” reporting content, or unfollowing an account, suppresses that content and similar future content in a user’s ranking. This gives users a direct, confirmed mechanism to shape their own feed, complementing the positive-engagement signals with an equally real negative counterweight.