What is LinkedIn’s feed actually ranking for?

The ranking model serving the majority of LinkedIn’s feed traffic in 2026 predicts two named objectives: long dwell — time spent on a post above a post-type-dependent threshold — and contribution, meaning a like, comment or reshare. Both are named in LinkedIn’s own published paper. BrandLoop’s drafting flow is built around those two objectives rather than formatting tricks.

Advice about the LinkedIn algorithm has a citation problem: follow the chain and it usually ends at another agency post rather than a measurement. LinkedIn itself publishes its ranking systems as engineering papers. Everything below comes from those papers or from official company statements.

The feed is a pipeline, not a model

LinkedIn’s engineering team describes a multi-stage architecture: first-pass rankers order candidates within each content source, a second-pass ranker merges them into one personalised list, and a reranker produces the final feed. The 2026 paper confirms the same shape — candidates are retrieved from multiple sources before ranking begins.

The practical consequence: there is no single “algorithm”. Your reach depends first on which candidate pool you entered, then on how you were ordered inside it. Two different stages can drop you for two different reasons.

What the 2026 model predicts

LinkedIn’s February 2026 paper describes a sequential ranker that had been serving the majority of feed traffic for over three months. It names two tasks.

The second is contribution: a like, comment or reshare. So the model optimises for two things — whether people read the post, and whether they add to it. The paper reports online A/B results of +2.10% time spent and +3.52% likes, comments or reshares.

The previous generation, LiRank (2024), points the same way: multiple action probabilities predicted jointly, with long dwell among them.

Dwell time is confirmed — its threshold is not

Dwell time is not folklore. LinkedIn announced it in a 2020 engineering post that defines two measures: on-feed dwell, which starts when at least half of an update is visible, and after-click dwell.

The gap worth naming: LinkedIn has never published the threshold. The 2024 paper defines it as a percentile; the 2026 paper says it depends on post type. Any advice quoting a specific number of seconds is quoting nothing.

LinkedIn says it is not built for virality

That is not an interpretation; it is a sentence in LinkedIn’s own engineering post: “LinkedIn is not designed for virality.” The same post describes a second defensive layer that watches engagement velocity — unusually fast spread gets inspected.

A 2022 company blog post is equally direct about reaction-baiting: “We won’t be promoting this type of content.”

How long does a post live?

LinkedIn has never answered this with a number. Claims that posts live 24 or 48 hours have no source. The only on-record statement came from a LinkedIn engineering director in a 2024 interview: content lives and dies quickly in the newsfeed, but a strong post can be resurfaced much later through the suggested-posts surface. Those are different surfaces; the distinction matters.

Claims with no source behind them

• The “golden hour”: LinkedIn has never described a first-60-minutes window. Every source found cites another marketing source.
• A specific dwell threshold in seconds: only a percentile definition and “depends on post type” have ever been published.
• Hashtag counts, emoji counts, editing a post: none confirmed by LinkedIn.
• 360Brew being “the new algorithm”: LinkedIn’s own paper calls it a research pre-production model, and the paper was withdrawn from arXiv in 2025.

What this means for writing

If the model predicts two things, writing has two jobs: be read, and be answerable. That simplifies most advice. The opening lines exist to say what the post is about, not to manufacture curiosity — people who cannot tell scroll past. The body has to be concrete enough to finish. The ending needs something a reader with actual experience can respond to.

BrandLoop’s drafting flow is built around those two objectives rather than formatting tricks. The cadence question — how often to post — is covered separately in our piece on posting frequency.

If you want to read the sources yourself

Every claim above traces to a document listed below. A short guide for reading them: the 2026 paper describes the ranker in production and names its objectives; the 2024 LiRank paper covers the previous generation and uses the same objective family; the 2020 engineering post defines how dwell time is measured; the company blog posts announce product-side decisions. Outside those four, LinkedIn has no public primary source on feed ranking. Everything else is commentary.

One caveat worth carrying: LinkedIn publishes retrospectively. A paper released in February 2026 describes a system that had already been serving traffic for three months. So even the most current document may describe the previous state of the system. That is why this piece avoids categorical statements.