How the LinkedIn Algorithm Works in 2026

The LinkedIn algorithm is not one rulebook but a set of AI models that picks, for every member, a small selection from millions of posts and puts it in order. LinkedIn Help names three groups of signals: identity (your profile, workplace, skills, location), content (how often a post is viewed or engaged with, what it is about and whether it provides knowledge or advice, how recent it is, whether it comes from someone you follow, its language, how constructive the conversation is) and activity (what you react to, who you interact with, what you spend time on). Since March 2026, LinkedIn has been rolling out a new ranking system built on large language models that understands what a post is about and reads your past interactions as a sequence. LinkedIn does not publish weights. Anyone quoting you fixed point values for likes and comments is not quoting LinkedIn.

My LinkedIn posts are drafted by Conny, my AI employee for content: short posts on AI news, each with an image that shows me as a recurring pixel-art figure. I approve every post myself. How to write and publish a post step by step is in AI for LinkedIn Posts: A Step-by-Step Guide. This article covers what happens next: what LinkedIn does with the post. Everything on the topic is collected on LinkedIn with AI.

How the LinkedIn algorithm works: two stages

On March 12, 2026, LinkedIn's engineering blog described how the new feed is built.

Stage 1, retrieval. Candidates used to come from several separate sources, such as your network's activity, trending posts in your region or industry, and what similar members liked. LinkedIn is replacing that with one system in which a language model describes members and posts as text and compares them. For a post, LinkedIn says that description includes the format, author information (name, headline, company, industry), engagement counts and the post text. LinkedIn's example: an electrical engineer who reads about small modular reactors gets related posts even when no keyword matches.

Stage 2, ranking. A model LinkedIn calls the "Generative Recommender" processes more than a thousand of your past interactions in order. It separates passive actions (click, skip, long dwell) from active ones (like, comment, share). A paper by LinkedIn researchers from February 2026 describes a ranker of the same kind (Feed SR) and says it has been "serving the majority of LinkedIn's Feed traffic for over three months." LinkedIn doesn't say outright whether it is the same model.

Two points from the same blog post matter if you publish. New posts are embedded in near real time, and when a topic gains traction, readers see relevant posts "within minutes, not hours." And demographics are out: LinkedIn says the ranking relies on professional signals and engagement, "never demographic attributes." LinkedIn said the same about gender in November 2025.

Many guides call the new system "360Brew." That was a model LinkedIn researchers described in January 2025 as a "research pre-production model"; every version of the paper is now marked withdrawn on arXiv (for licensing reasons, per arXiv's note), and the March 2026 post never uses the name. There is no evidence 360Brew runs your feed.

The ranking signals LinkedIn names officially

The help article "LinkedIn relevance" has LinkedIn's most detailed list. The right column is an inference, not LinkedIn's words.

For posts recommended outside your network, LinkedIn has a separate list. It aims to suggest professionally relevant content, for example about professional growth and education, being inspired at work, professional products or services, and education on topics like economics or technology. It also checks that such a post "provides sufficient context, isn't overly promotional, and has related elements such as text, images, and video." And it weighs feedback, such as members signalling disinterest or reporting a post.

Post formats: text, image, document, video

LinkedIn's cited sources contain no ranking of formats. What is documented: format is part of how the system describes a post, and for out-of-network recommendations LinkedIn names text, images and video as elements.

A rule of thumb, as a hypothesis: pick the format for the content, not for reach. An argument works as text. An image helps when it shows something the text can't. A document suits a process with several steps, a video suits a demonstration. How to build such a document is in How to Make a LinkedIn Carousel with AI. The limits for each format, from characters to file size, are in the table in AI for LinkedIn Posts: A Step-by-Step Guide.

One thing is documented: LinkedIn's spam policy lists "Emoji / reaction polls that artificially boost engagement," meaning posts like "heart for A, thumbs up for B." The built-in poll feature is not what that refers to.

Comments: what counts and what is spam

According to LinkedIn, comments are among the active actions the ranking model predicts. In 2020, LinkedIn explained that a comment "provides valuable feedback to the creator" and may encourage them to post more often.

The limits are spelled out too. The spam policy names "excessive, irrelevant, or repetitive comments or messages," chain-letter posts asking for likes, and posts that misrepresent a LinkedIn feature such as the double tap. The Professional Community Policies add: "Don't do things to artificially increase engagement with your content," and "don't agree with others ahead of time to like or re-share each other's content." That covers engagement pods. Bots that comment or like are prohibited by the User Agreement.

What follows, as a hypothesis: reply to comments yourself, adding something. An AI can suggest the reply; you send it.

When to post: what is actually documented

LinkedIn names no official best time. In July 2024, LinkedIn's marketing blog summarized studies from HubSpot, Sprinklr and Buffer and noted "some inconsistencies and conflicting data points." Per that summary, weekdays generally do best, while the hours differ. LinkedIn's conclusion: "the timing is often less important than the substance."

What the new system adds: LinkedIn balances freshness against relevance and refreshes a post's data as it gains engagement. The often-quoted "golden hour," where the first 60 or 90 minutes decide everything, appears in none of the LinkedIn sources cited here. A suggestion, as a hypothesis: post when your readers are at work, then compare in your post analytics after a few weeks. Where to find them is in the FAQ of AI for LinkedIn Posts.

LinkedIn algorithm myths without a source

On the last one: LinkedIn says content that "feels generic, repetitive, or lacks a clear point of view is less likely to be widely distributed," however it was written. How to avoid that with AI is in Writing with AI: A Beginner's Guide.

An example: the short news post

This is how a post in Conny's format could look. An example, made up: a vendor renames an AI subscription. The post has five lines: what changed, who it matters to, what to do now, plus the image with the pixel figure.

Why that fits LinkedIn's signals, as a hypothesis: the topic is unambiguous, so the model can match it to interested readers, including outside my network. It is timely, it offers knowledge rather than a pitch, and it has an image. Whether it travels is up to the readers; nobody can guarantee that. How to turn AI news into a steady supply of topics is in LinkedIn Post Ideas: 25 Formats and AI News. What matters on your profile, since it feeds into how every post is described, is in LinkedIn Profile Optimization.

If you want LinkedIn for client contacts rather than reach, that is a separate topic: Finding B2B Leads: LinkedIn and Alternatives. Where the line is when AI writes for you is in AI as Your LinkedIn Ghostwriter.

Frequently asked questions

How does the LinkedIn algorithm work in 2026?

In two stages: a language model retrieves the posts that match your interests from millions of candidates, then a ranking model orders them based on more than a thousand of your past interactions.

Does LinkedIn penalize external links?

None of the help pages, policies and engineering posts cited here contain such a rule (as of November 2026). Figures that put a number on a penalty come from third parties. Check your own post analytics.

What is the best time to post on LinkedIn?

There is no official one. LinkedIn itself says timing is often less important than substance. The studies it cites favour weekdays and disagree on hours.

Can I see my feed in chronological order?

Yes. LinkedIn Help says you can select "Most recent first" for a single session, or set "Most recent posts" as your default in Settings. Outside the European Economic Area, the per-session option may only be available on desktop.

How do I increase my LinkedIn reach?

LinkedIn's own sources point to substance: knowledge or advice, enough context for people outside your network, nothing overly promotional, real conversation in the comments. Paid reach exists only as labelled sponsored content.

Where to go from here

Check your last five LinkedIn posts against the table above: one topic? Clear to strangers? Knowledge or a pitch? Write the next one so every answer is yes, and compare in your analytics after a few weeks. My weekly newsletter "From the Engine Room" comes with any resource on my free templates page. Conny herself is a ready-made package in my community, the AI Employee Framework. Whether to use LinkedIn's own newsletter feature or your own list is covered in LinkedIn Newsletter vs. Your Own List.