The LinkedIn algorithm decides which posts each member sees by matching what a post is about to what that member cares about, then ranking the matches by how likely the member is to read, react, comment or share. In the feed LinkedIn described in March 2026, a language model does the matching, so your words do the targeting. LinkedIn lists three groups of signals: who you are, what the post is, and how people act on it.
Key Takeaways
- LinkedIn rebuilt its feed in 2026. Its engineering team describes one retrieval system built on embeddings from a large language model, replacing several older sources, and a ranking model that reads more than a thousand of a member’s past interactions.
- LinkedIn’s help center groups ranking signals into Identity, Content and Activity. Time spent viewing a post, how recent it is, and how recent its engagement is all appear on that list.
- Dwell time has counted since at least 2020: LinkedIn starts measuring it when half of a post is visible on screen.
- LinkedIn says ranking uses professional signals and engagement, “never demographic attributes”, and that it is working to make engagement pods ineffective.
- LinkedIn has published nothing on a link penalty, a “golden hour” or the best number of hashtags. Treat confident numbers about those as other people’s measurements, not LinkedIn’s rules.
What LinkedIn has said, and what it has not
The quickest way to judge advice about the LinkedIn algorithm is to sort it by source. The left column below comes from LinkedIn’s own engineering blog, newsroom and help center. The right column is common advice that LinkedIn has not confirmed in anything we could find.
| LinkedIn has documented | LinkedIn has not confirmed |
|---|---|
| A language model matches posts to members by meaning (March 2026) | A fixed “golden hour” that decides a post’s fate |
| Signals grouped as Identity, Content and Activity | A specific reach penalty for posts with external links |
| Time spent viewing a post is a signal; measured from half-visible | That editing a post soon after publishing hurts it |
| How recent the post is, and how recent its engagement is | An ideal number of hashtags, or that hashtags add reach |
| Profile signals such as industry, skills, experience and geography | That posting more than once a day is penalised |
| Never demographic attributes; posts from different creators audited to “compete on equal footing” | That comments longer than a set number of words count more |
| Engagement pods and comment automation are being made ineffective | That a post’s first hour of comments sets its whole reach |
None of the right-hand claims is necessarily false. Some come from third-party analyses of real posts. At best they describe how things behaved on someone’s sample, at one point in time, and LinkedIn has never stated them as rules.
How does the LinkedIn algorithm work in 2026?
It works in two steps. First, retrieval picks candidate posts out of the millions available by comparing the meaning of each post with a member’s interests, using embeddings from a large language model. Second, a ranking model orders those candidates by predicting what the member will do with each one, using their recent history of reading, reacting, commenting and scrolling past.
LinkedIn’s engineering post of 12 March 2026, written by Hristo Danchev, describes what this replaced. The old feed drew candidates from several separate sources: “a chronological index of your network’s activities, trending posts in your geography, collaborative filtering” and other embedding systems. The new feed has “a single, unified retrieval system powered by LLM-generated embeddings.”
The ranking step is what LinkedIn calls a Generative Recommender, which its newsroom says is “rolling out”. It “processes more than a thousand of your historical interactions”, which the post describes as “what you’ve read, liked, commented on, returned back to, or simply scrolled past.” Long dwells, likes, comments and shares are named as the actions it learns from. The same post says new posts reach the system “within minutes, not hours”, and that the feed serves more than 1.3 billion professionals.
The practical consequence is that a post is matched to people by what it says. A post that explains how a support team cut its first-response time is retrieved for members whose history shows interest in customer support operations, whether or not they are connected to you.
What signals does LinkedIn use to rank posts?
LinkedIn’s help center lists three groups. Identity covers who the member is: profile details such as location, workplace and skills. Content covers the post: what it is about, whether it gives knowledge or advice, its language, how recent it is, how constructive the conversation is, and who is tagged. Activity covers behaviour: what the member engages with, who they interact with, and which content they spend the most time viewing.
The help article, “LinkedIn relevance”, spells the lists out. A few lines matter more for a writer than the rest:
- “What the post is about and whether it provides knowledge or advice.” This is the Content signal a language model is built to read. A post with a clear subject gives it something to match.
- “How constructive or professional the conversation is.” The quality of the comment thread counts, not only the number of comments.
- “Content you spend the most time viewing.” Time is a signal in its own right, separate from reactions.
- “Engagement others have with your posts and how recent it is.” Recency applies to engagement too, which is the documented basis for the advice to stay around and reply after you post.
- “Whether the content comes from someone you follow or are connected to.” Your network still counts, as one listed signal among several.
The engineering post adds the profile side: ranking uses “information you’ve chosen to share on your own profile such as industry, experience, skills and geography.” Your stated industry, experience and skills help the feed decide who your posts are for, so a stale profile works against you.
Why does dwell time matter so much on LinkedIn?
Many members rarely react, but they all scroll and stop. LinkedIn’s engineers explained in May 2020 that click and share actions can be rare, are mostly all-or-nothing, and that clicks are noisy, while time spent is measurable for every post anyone sees. So the feed learns from how long people look at a post, and also from which posts they scroll past quickly.
The 2020 engineering post, “Understanding dwell time to improve LinkedIn feed ranking”, defines two kinds. Dwell time “on the feed” “starts measuring when at least half of a feed update is visible as a member scrolls through their feed.” Dwell time “after the click” is the time spent on content after tapping into it. The team also defined “skipped updates”: posts a member views for less than a threshold time without acting on them. That gave the model a negative signal it never had from reactions alone.
Two things follow for anyone writing posts:
- The first lines decide the skip. A reader sees the opening before the “see more” cut. If it does not tell them what they will get, the post is skipped, and the skip is recorded.
- Length is not the goal. A long post that nobody expands earns little dwell time. A short post people read twice can earn more. The 2026 post lists “long dwells” next to likes, comments and shares as actions it learns from, so the aim is attention that holds, not word count.
Does LinkedIn penalise links, edits or posting time?
LinkedIn has not said so in its engineering posts, newsroom or help center. Its 2026 engineering post does not mention links, hashtags or posting time at all. Confident claims about these come from third-party analyses of post samples, not from LinkedIn, and they do not agree with each other. LinkedIn has confirmed none of them.
There is a plausible mechanism for the link effect, and it runs through dwell time rather than a rule. A reader who taps a link leaves LinkedIn, so the post earns less time on the feed and fewer chances of a comment. That is a hypothesis, not a published fact. If you want to know whether links cost your own posts reach, test it: alternate links in the post and links in the first comment over a few weeks and compare impressions in LinkedIn’s post analytics.
Posting time works the same way. The help center lists how recent a post is, and how recent its engagement is, as signals, so a post published when your audience is asleep starts with less of both. Nothing LinkedIn has published sets a fixed window. Your own past results are a better guide than a global “best time” chart.
Hashtags have their own article: our answer to whether hashtags work on LinkedIn covers what changed after LinkedIn removed hashtag following and the three cases where a tag still helps.
Do engagement pods and comment automation still work?
LinkedIn says it is working to stop them. Its newsroom post of 12 March 2026 says the company is “working to make engagement pods ineffective and curb comment automation or third party tools”, and names “unauthorized third-party tools” among the behaviours it acts against. The same post says “We want the Feed to be relevant to your interests, not a popularity contest”, which is the opposite of what a pod is built to exploit.
A pod is a group that agrees to like and comment on each other’s posts as soon as they go up. Pods rest on the belief that an early burst pushes a post to more people. Under the 2026 feed, LinkedIn says, the ranking learns from a member’s own behaviour, and the help center counts “how constructive or professional the conversation is.” Ten “Great post!” comments from the same people every day are not the constructive conversation that signal describes.
Comment automation is the same problem in software form. A tool that posts generic comments under your name is what the newsroom post names as a target. The risk is not only lost reach: comments that read as automated cost you credibility with the people you were trying to reach.
How do you write for the 2026 LinkedIn feed?
Write for one kind of reader and say plainly what the post gives them. The language model matches meaning, so a specific subject in plain words reaches the right members. Dwell time rewards an opening that earns the “see more” tap. The conversation signal rewards posts that invite a real answer. Then stay to reply, because recent engagement is a listed signal.
A useful check before you publish has three questions, one for each signal group:
- Reader (Identity): who is this for? If your profile says you work in finance operations and the post is about finance operations, the feed has two consistent signals. If your posts wander across unrelated subjects, the model has less to go on.
- Topic (Content): could someone describe the post in one sentence after reading the first two lines? “How we cut month-end close from six days to three” passes. “Some thoughts on leadership” does not.
- Stop (Activity): is there a reason to stop scrolling and a reason to reply? A concrete number, a decision someone else faces, a question you actually want answered.
LinkedIn posts can hold up to 3,000 characters, so space is rarely the constraint. For formats that pass all three questions, our list of 30 LinkedIn post ideas has worked examples, and the B2B social media strategy guide covers choosing the audience in the first place.
Members also shape their own feed, which is worth knowing when your reach drops. The help center says members can hide a post with the X icon and say why, unfollow or mute people and Pages, and switch between “Most relevant first” and “Most recent first” for the current session, or set a default in Settings. A member who hides one of your posts can tell LinkedIn why, which the help center says is used to improve future recommendations.
How can PostNext help with LinkedIn posting?
A scheduler cannot change how LinkedIn ranks a post, but it can keep a steady rhythm and show you when your own audience responds. PostNext publishes text posts and posts with one or more images or a GIF to a personal LinkedIn profile through LinkedIn’s official API. On paid plans it suggests three posting slots from how your own LinkedIn posts performed over the last 90 days, and nothing is scheduled until you confirm.
Know the limits of the PostNext LinkedIn scheduler before you plan around it. PostNext does not publish LinkedIn video, document carousels, polls, articles or newsletters, and it does not post to Company Pages yet. For LinkedIn it reports likes and comments per post, not impressions or clicks, so read impressions in LinkedIn’s own analytics. The AI writer drafts posts with a hashtag set you can trim or delete, and every draft stays editable. To check how line breaks and the opening will look before the post goes out, paste it into the free LinkedIn post formatter.
For a wider view of how the other networks rank posts, see how social media algorithms are changing in 2026. For a step-by-step setup, our guide to scheduling LinkedIn posts walks through it.
Frequently Asked Questions
How does the LinkedIn algorithm work?
It matches posts to members by meaning, using a large language model, then ranks the matches by predicting what each member will read, react to, comment on or share. LinkedIn groups its signals into Identity, Content and Activity.
Did the LinkedIn algorithm change in 2026?
Yes. On 12 March 2026 LinkedIn published details of a rebuilt feed: one retrieval system built on language-model embeddings replaced several older sources, and a new ranking model, which LinkedIn says is rolling out, reads more than a thousand of each member’s past interactions.
Does LinkedIn show posts to people outside my network?
Yes. LinkedIn says the 2026 feed surfaces posts whether they come from a member’s network or from professionals they have never connected with, when the topic matches their interests.
Does dwell time affect LinkedIn reach?
Yes. LinkedIn lists the content a member spends the most time viewing as a ranking signal, and its engineers have measured dwell time on the feed since at least 2020, starting when half a post is visible.
Do external links hurt LinkedIn posts?
LinkedIn has not said they do. The claims come from third-party analyses, which do not agree with each other. Test it on your own posts by comparing links in the post with links in the first comment.
Does posting time matter on LinkedIn?
It matters indirectly. LinkedIn lists how recent a post is, and how recent its engagement is, as signals, so posting when your audience is active helps. LinkedIn has not published a fixed best time or a “golden hour”.
Related reading
- Do Hashtags Work on LinkedIn in 2026?
- 30 LinkedIn Post Ideas You Can Use This Week
- How Social Media Algorithms Are Changing in 2026
- How to Schedule LinkedIn Posts
- B2B Social Media Strategy
Sources
- LinkedIn Engineering, Engineering the next generation of LinkedIn’s Feed, Hristo Danchev, 12 March 2026 (read October 2026)
- LinkedIn Newsroom, How LinkedIn Is Improving the Feed to Show More Relevant, Authentic Professional Content, 12 March 2026 (read October 2026)
- LinkedIn Help, LinkedIn relevance: optimizing the member experience (October 2026)
- LinkedIn Engineering, Understanding dwell time to improve LinkedIn feed ranking, Siddharth Dangi, Johnson Jia, Manas Somaiya and Ying Xuan, 12 May 2020 (read October 2026)
- LinkedIn Help, post character limit and attachments (October 2026)
- PostNext, postnext.io/linkedin-scheduler (October 2026)
