How the LinkedIn algorithm works, from what LinkedIn has said

What LinkedIn says the feed does

LinkedIn's Help Center says its systems "consider hundreds of signals" to decide what shows up in each member's feed. It groups them in three kinds.

• Identity: details from the reader's profile.
• Content: the post itself, including engagement, how recent it is, and whether it is professional.
• Activity: what the reader has engaged with and what they seem to care about.

LinkedIn also says what it does not use. Its Help page states the algorithm does not use demographic details such as age, race or gender to decide who sees a post. And it says its systems "filter out or taper distribution of low quality and unsafe content."

Readers have some control too. Each member can sort the feed by most relevant or most recent, unfollow or mute people, hide a post, and mark a post as "not interested". So a slice of your readers may see your posts in time order, not ranked.

The March 2026 update

On 12 March 2026 LinkedIn published two pieces about a new feed: a news post for members and a longer engineering post. Together they say three things changed.

1. A new way to find posts for you

LinkedIn replaced several separate ways of finding candidate posts with one system built on large language model embeddings. The goal, in its words, is a system that "better understands what a post is actually about." It also says posts from different creators now "compete on equal footing", and trending posts can reach people within minutes.

2. A model of your reading history

The ranking model now reads each member's past engagement as an ordered sequence, looking at more than 1,000 past interactions. The engineering post names the signals it learns from: long dwells, likes, comments and shares, plus the reader's profile, company and industry.

3. Less bait, fewer pods

The news post says LinkedIn is reducing "repetitive, low substance posts and engagement bait, such as 'comment to agree' prompts or videos that don't match the text." It also says it is taking action against "comment automation, engagement pods, and unauthorized third party tools."

Dwell time: why skimmed posts sink

In 2020 LinkedIn's engineers explained how they use dwell time. They measure time on the feed, counted when at least half of a post is on screen, and time after a click. They built a model that predicts the chance a reader will skip a post, and they lower each post's score in line with that chance.

The March 2026 engineering post still lists long dwells as a signal. So the lesson holds: a post people stop and read does better than one they scroll past. That favors a clear first line, short paragraphs and one idea per post.

What LinkedIn has confirmed, and what it has not

A lot of LinkedIn advice is stated as fact without a source. Here is what we could trace to LinkedIn itself, and what we could not.

"Not published" does not mean false. It means LinkedIn has not said so, so any exact number you see for it is a guess. Test it on your own posts before you build a habit around it.

What to do with this

1. Write one clear idea per post. A model that tries to understand "what a post is actually about" rewards a post that is about one thing.
2. Earn the stop. Put the point in the first line so people read past the fold. Our hook writing guide shows how.
3. Invite real comments, not reflex ones. Ask a specific question you want answered. Skip "comment YES if you agree".
4. Reply to comments yourself. Comments are one of the signals LinkedIn names, and a thread with the author in it gives readers a reason to come back.
5. Stay away from pods and auto commenters. LinkedIn says it acts against both, and its User Agreement bans bots that comment, like or share.
6. Post for your readers' day. Time is a small lever, but it is cheap. See the best time to post on LinkedIn.

How the feed has changed over time

LinkedIn has tuned its feed for years. In 2017 its engineers wrote that they watch "the velocity of likes, shares, and comments" to spot posts that might go viral, partly so they can catch spam and low quality posts early. They described a range of actions, from demoting a post in ranking to limiting who sees it. In 2020 they added the skip model. In 2026 they rebuilt retrieval and ranking around language models.

The direction is steady: fewer tricks, more weight on whether a real reader found the post worth their time. Any tactic that works by gaming the count, rather than earning the read, tends to stop working.