How the LinkedIn Algorithm Works: Dwell Time, Your Network and Relevance
LinkedIn is one of the few platforms that has explained its feed in its own engineering blog. Two of those posts, years apart, tell you most of what matters.
The LinkedIn algorithm decides which posts appear in your feed and in what order, and LinkedIn has described how it works in its own engineering blog. Two signals stand out. Dwell time, how long you spend on a post, counts alongside clicks and reactions. And since a rebuild LinkedIn described in March 2026, language models match posts to members by topic, so relevance matters more than who you're connected to.
In practice, that means a post that people stop and read can rank well without many likes, and a post on a subject you've shown interest in can reach you from outside your network. Engagement bait, pods and automated activity work against a feed built to predict what each member finds worth their time.
Most algorithm advice on social media is guesswork dressed up as a leak. LinkedIn is unusual because it has written about its feed in public, in posts from its engineering team, and those posts are a better guide than any list of hacks.
The first idea is dwell time. In a post titled Understanding dwell time to improve LinkedIn feed ranking, LinkedIn's engineers explained that clicks and reactions miss a lot of what people value. Plenty of members read a post carefully and move on without liking it. So LinkedIn measures time: how long a post is at least half visible on screen as someone scrolls, and how long they spend after clicking into it. A later engineering paper describes the ranking model predicting whether a member's dwell will pass a threshold that depends on the kind of content, so a two-line post and a long one aren't held to the same bar.
The second idea is relevance. On March 12, 2026, LinkedIn published Engineering the next generation of LinkedIn's Feed, describing a feed rebuilt around language models. Posts and member profiles are represented in a shared space, so the system can retrieve posts that fit a member's interests, and a sequence model reads a member's past interactions in order to rank what to show next. The effect LinkedIn describes is that content can reach beyond direct connections when the topical fit is strong.
There's a third factor that no other network in this catalogue has: every reaction on LinkedIn shows the reactor's name and headline, usually their job and company. A like from someone in your industry is visible as exactly that to everyone who opens the list. That's why reactions carry social weight on LinkedIn that they don't elsewhere, and why the reaction count under a post is read closely. It's also the number that LinkedIn likes change.
How does the LinkedIn algorithm work?
Put together from LinkedIn's published engineering material, as of September 2026, the feed works in two broad steps:
- Retrieval: from everything posted, the system gathers candidate posts for you. Since the 2026 rebuild, a single retrieval system uses language-model representations of posts and members, so candidates can come from outside your network when they match your interests.
- Ranking: a model orders the candidates by predicting what you'll value. LinkedIn's engineering posts describe predictions of actions such as reacting, commenting and sharing, and of dwell time.
- Your history: the ranking model reads your past interactions as a sequence, so what you engaged with recently shapes what comes next.
- Quality controls: LinkedIn's policies prohibit fake engagement and automation, and its systems demote content designed to bait reactions.
What is dwell time on LinkedIn?
Dwell time is the time a member spends on a post. LinkedIn measures it in two places: in the feed, starting when at least half of the post is on screen, and after a click, when the member opens the post, article or document. It matters because it captures interest that doesn't produce a reaction. A careful read with no like is still a signal that the post was worth someone's time.
Did LinkedIn change their algorithm?
Yes, substantially, and it said so. The engineering post published on March 12, 2026 describes replacing several separate retrieval sources with one unified system built on language-model embeddings, and ranking with a sequence model LinkedIn calls a Generative Recommender, which reads more than a thousand past interactions in order. LinkedIn reports that retrieval and ranking run in well under a second.
What it means for posting: the old intuition that a bigger network automatically means more visibility is weaker than it was. The system is designed to match posts to people by topic, so a clear, consistent subject helps the model understand who a post is for.
What about monthly LinkedIn algorithm updates?
Searches for a LinkedIn algorithm update by month are common, and so are posts claiming one. LinkedIn does change its systems continuously, but it rarely announces a named update. Treat any specific-date claim without a LinkedIn source with caution. The two engineering posts cited here are published by LinkedIn and can be read in full on its engineering blog.
What content performs best on the LinkedIn algorithm?
LinkedIn doesn't publish a ranking of formats, and the honest answer follows from the signals it does describe. Content that holds attention and fits a clear topic tends to do well, because those are the two things the feed measures.
- Posts people read to the end: a useful point, a specific example, a lesson from real work. They earn dwell time even when readers don't react.
- Posts with a consistent subject: the 2026 feed matches content to members by topic, so writing about the same area helps the system find your readers.
- Posts that invite real comments: a question the reader has an opinion on. LinkedIn post engagement goes through how to write for that.
- Documents and carousels that people page through, which produce long dwell.
What hurts reach on LinkedIn
Engagement bait, such as asking for reactions to vote or comments with a single word to get a file, is the kind of content LinkedIn has said it demotes. Engagement pods, groups that react to each other's posts on schedule, and automation tools that like or comment on your behalf break LinkedIn's rules against fake engagement and automated activity, and the feed's predictions are built to see through patterns that don't match real interest.
Why reactions carry more weight on LinkedIn
Open the list of reactions on any LinkedIn post and you'll see who reacted, with their headline beside each name. On most platforms, a like is anonymous in practice; on LinkedIn, it's a professional endorsement with a job title attached. That shapes how readers judge a post: a few reactions from recognizable people in a field read as credibility.
The reaction count is also the first number a reader sees under a post, before they decide whether to spend time on it. That's the part LinkedIn likes change. Be clear about the limit: they add to the visible count; they don't add dwell time, they don't write comments, and they don't change the topic match that decides who the feed shows the post to. For comments, which LinkedIn also displays with name and headline, LinkedIn comments is the separate service. LinkedIn engagement rate covers how to measure the whole picture, and the rest of the catalogue is on the LinkedIn services page.
Does your network still matter?
Yes, just not as much on its own. Posts from people you're connected to or follow remain natural candidates for your feed, and early reactions from them are often how a post starts to travel. But the 2026 system can also bring in posts from outside your network when the topic fits, and it can pass over posts from connections that don't. A network of people who care about your subject is worth more than a large network that doesn't. LinkedIn followers vs connections explains how the two kinds of audience differ.
Questions people ask about the LinkedIn algorithm
How does the LinkedIn algorithm work?
It gathers candidate posts for each member, then ranks them by predicting what that member will value, including reactions, comments and dwell time. Since 2026 it uses language models to match posts to members by topic.
What is dwell time on LinkedIn?
The time a member spends on a post, measured in the feed when at least half the post is on screen and after clicking into it. LinkedIn uses it as a ranking signal.
Did LinkedIn change its algorithm in 2026?
Yes. On March 12, 2026, LinkedIn's engineering blog described a rebuilt feed using language-model retrieval and a sequence-based ranking model that reads members' past interactions.
What content performs best on LinkedIn?
LinkedIn doesn't rank formats. Its signals favor posts that hold attention and fit a clear topic, and it demotes engagement bait.
Do likes still matter on LinkedIn?
Yes. Reactions are among the actions the feed predicts, and on LinkedIn each one shows the reactor's name and headline, which gives it more social weight than on most platforms.
Are LinkedIn engagement pods allowed?
No. LinkedIn's rules prohibit fake engagement and automated activity, and coordinated reacting in pods falls under that.
Keep reading
- LinkedIn likes — The reaction count is the first number a reader sees under a post, and it's what LinkedIn likes change.
- LinkedIn post engagement — How to write posts that earn comments and dwell time.
- LinkedIn followers vs connections — The two kinds of audience the feed draws on.