How the LinkedIn algorithm works in 2026

In 2026, LinkedIn's feed runs on large language models (LLMs). One system reads posts to find candidates for your feed. A second model ranks them using your last thousand-plus interactions and predicts whether you'll stop to read, skip, like, comment, or share. LinkedIn's engineering team explained this in March 2026, the most detailed look at the feed LinkedIn has ever shared. The same day, LinkedIn said it would cut reach for engagement bait and recycled "thought leadership."

Most algorithm guides mix those facts with guesses. This one keeps them apart: every claim says whether it comes from LinkedIn or from outside practitioners, and practitioner numbers stay out.

What LinkedIn has actually published

The main sources, newest first, plus a few public comments from LinkedIn's product team in 2025:

• March 2026, engineering blog. "Engineering the next generation of LinkedIn's Feed" by Hristo Danchev. How the 2026 feed finds and ranks posts.
• March 2026, Tim Jurka. "Updates to the LinkedIn Feed focusing on authentic, relevant conversations." What LinkedIn will show less of.
• January 2025, research paper. "360Brew," a ranking model from LinkedIn researchers. More on it below.
• May 2020, engineering blog. "Understanding dwell time to improve LinkedIn feed ranking."

How the 2026 feed works

LinkedIn's March 2026 post describes two steps.

Step 1: finding candidates. LinkedIn used to pull posts from several separate sources (your network, trending posts, topics you follow). It replaced them with one LLM-based system that turns both members and posts into a description of what they're about. That lets it match posts by meaning, not just keywords. LinkedIn's own example is "discovering content from authors you don't follow, on topics you care about, expressed in different terminology than you might use."

Step 2: ranking. A model LinkedIn calls a Generative Recommender looks at your feed history as a sequence, "more than a thousand" past interactions, instead of scoring each post on its own. Each post goes into the model as text: its format, the author's name, headline, company, and industry, its engagement counts, and the post text itself.

Two more details matter:

• It learns from positive engagement. LinkedIn wrote that when it trained only on posts that got positive engagement, "the model learned much more effectively."
• It moves fast. Popularity and recency signals refresh "within minutes," so a post that starts getting traction can spread quickly.

What about 360Brew?

You'll see "360Brew" in almost every 2026 algorithm guide, often described as the model that now runs the feed. Here's what's on the record.

In January 2025, LinkedIn researchers published a paper introducing "our research pre-production model, 360Brew V1.0, a 150B parameter, decoder-only model that has been trained and fine-tuned on LinkedIn's data and tasks." The idea: one large model that reads text and handles many ranking tasks. arXiv's administrators later withdrew the paper over a licensing issue with the submission.

LinkedIn's March 2026 feed post doesn't use the name 360Brew. It describes LLM-based retrieval and a Generative Recommender. The two clearly share an approach (read the text, use the member's history), but LinkedIn hasn't said publicly that the 2026 feed is 360Brew. When a guide quotes 360Brew's parameter count as "the algorithm," treat it as a reasonable guess, not a fact.

Signal by signal

Dwell time

What LinkedIn says: dwell time is how long a post stays on screen (at least half visible) and how long people spend after clicking it. LinkedIn added it in 2020 because clicks, likes, and shares are rare for people who mostly read. "Long-dwell" is still one of the outcomes the 2026 model predicts, next to click and skip.

What to do: make the first two lines worth stopping for.

Comments and conversation

What LinkedIn says: the 2026 model predicts likes, comments, and shares as "active" outcomes. Tim Jurka's March 2026 post is titled around "authentic, relevant conversations" and says LinkedIn will cut reach for engagement bait such as "Comment 'Yes' if you agree."

What practitioners say: many consultants claim comments count far more than likes. LinkedIn hasn't published weights, so we don't repeat their multipliers.

What to do: end with a real question when you have one, and reply to comments in the first hours. Never ask for one-word comments.

Saves and sends

What LinkedIn says: in September 2025, LinkedIn added saves and sends (shares by private message) to post analytics, as Social Media Today reported. It hasn't said how they're weighted. The March 2026 post lists click, skip, long-dwell, like, comment, and share as predicted outcomes and doesn't name saves.

What practitioners say: "saves beat likes" is a common claim. It's plausible, since a save shows intent, but LinkedIn hasn't confirmed it.

What to do: write posts worth keeping (checklists, how-tos, document posts), then check saves in your analytics to see what your audience bookmarks.

External links

What LinkedIn says: in September 2025, Rishi Jobanputra said links aren't deliberately penalized, as long as the post gives "enough context that it could stand alone without the link."

What others report: Social Media Today noted in November 2024 that LinkedIn shows a smaller link preview on unpaid posts, and pointed to outside studies finding link posts get less reach. Those studies come from third parties, and their figures vary.

What to do: make the post useful without the click. Test the link in the post against the link in the first comment on your own account.

Hashtags

What LinkedIn says: in July 2025, Jobanputra said, "Since hashtags were introduced over a decade ago, I'd say our feed algorithm has really evolved, and we're just doing a better job at what content is about." The 2026 system reads post text to work out the topic.

What to do: use a few hashtags if they help people search, or none. Write the topic plainly in the post; that's what the model reads.

Video

What LinkedIn says: LinkedIn began testing a vertical video feed in its app in March 2024. In early 2025, it told Digiday that video viewership was up 36% year over year. In March 2026, it said posts with a video that "has nothing to do with the associated text" should no longer gain extra reach.

What to do: use video when the video is the point.

Posting frequency

What LinkedIn says: for Company Pages, LinkedIn's marketing blog says "companies that post at least weekly see a 2x lift in engagement with their content." For members, LinkedIn hasn't published a number. Its editor in chief Dan Roth said in 2024 that if you keep sharing knowledge, "over the length of your posting, it is going to work out for you."

What to do: pick a pace you can keep for months, such as 2 or 3 posts a week, and schedule them so a busy week doesn't break it.

Freshness

What LinkedIn says: in July 2025 LinkedIn rolled back a change after members saw "many more 'older posts' than you would have liked." The 2026 system refreshes recency signals within minutes.

What to do: post when your audience is online, so the early response comes fast. Our guide to the best time to post on LinkedIn has starting windows and a two-week test.

Summary table

A practical checklist

Before you post:

• The first two lines say why this post is worth reading
• The post makes sense without the link
• The topic is named in plain words, not only in hashtags
• There's no "comment YES" or "like if you agree"
• Any video or image belongs to this post, not borrowed for reach
• Your headline says what you do, since the model reads it with your post (the free LinkedIn headline generator helps)

After you post:

• Reply to comments in the first few hours
• Check impressions, comments, and saves a week later
• Write more of what got saved and discussed

Keeping a steady pace is easier when the week is planned in one sitting. Our guide on how to schedule LinkedIn posts covers LinkedIn's built-in scheduler and the official API route. If LinkedIn is one of several networks you post to, the Lemonpost LinkedIn scheduler puts your profile and Company Pages on one calendar, with the first comment scheduled alongside each post.

FAQ

How does the LinkedIn algorithm work in 2026? LinkedIn uses an LLM-based system to find posts that match your interests, including from people you don't follow, and a Generative Recommender to rank them using more than a thousand of your past interactions. It predicts outcomes like long dwell, skips, likes, comments, and shares.

Is the LinkedIn algorithm called 360Brew? 360Brew is a ranking model described in a January 2025 LinkedIn research paper. LinkedIn's March 2026 post on the new feed doesn't use that name, so calling the 2026 feed "360Brew" is a guess, not something LinkedIn has confirmed.

Does LinkedIn penalize posts with links? LinkedIn says no, as long as the post stands on its own without the link. Outside studies have found link posts get less reach, and LinkedIn shows smaller link previews on unpaid posts, so test the first-comment approach on your own account.

Do hashtags still work on LinkedIn? They still help people search, but LinkedIn has said its feed now understands what a post is about from the text. Use a few hashtags or none, and name your topic clearly in the post.

Does LinkedIn penalize scheduled posts? LinkedIn has its own built-in scheduler, and nothing it has published suggests scheduled posts are ranked differently. What matters is how people respond to the post.

How often should I post on LinkedIn? LinkedIn says Company Pages that post at least weekly see twice the engagement. It hasn't published a number for personal profiles. Pick a pace you can keep for months, such as 2 or 3 posts a week.