# How the LinkedIn Algorithm Works in 2026 and What Changed Since 2025

> LinkedIn now ranks posts with LLM retrieval and a model that reads each member's history. What changed in 2025 and 2026, and what to post now.

Source: https://linkedinlikes.com/blog/linkedin-algorithm · Updated: 2026-11-12

## Key takeaways

- LinkedIn's feed now picks candidate posts with one LLM-based retrieval system and ranks them with a model that reads more than a thousand of a member's past interactions.
- In March 2026 LinkedIn said it would show fewer bait posts, fewer posts with unrelated video and fewer recycled generic posts.
- Richard van der Blom's 2025 study of more than 1.8 million posts put the drop in organic reach at nearly 50% year over year.
- His November 2026 update found that 85% of a post's reach now arrives after its first two hours.
- LinkedIn says links are not penalized on purpose, yet a 2026 study still measured about 16% less reach for posts with one link.

The LinkedIn algorithm in 2026 works in two steps. First, one LLM-based retrieval system picks posts whose topic matches what a member has shown interest in. Then a ranking model that reads the member's last thousand-plus interactions as a sequence decides the order. LinkedIn rolled this out over 2025 and early 2026, and paired it with filters against bait posts and automated comments.

Below is what LinkedIn has published about the system, what independent studies of millions of posts measured, and what to change in how you post.

## How does the LinkedIn algorithm work in 2026?

LinkedIn described the new feed on its [Engineering blog in March 2026](https://www.linkedin.com/blog/engineering/feed/engineering-the-next-generation-of-linkedins-feed). It has two layers, and each one rewards something different.

### Step 1: retrieval picks the candidates

Until 2025 the feed pulled candidates from several separate sources: a chronological index of your network's activity, trending posts in your region, collaborative filtering based on similar members, industry trends and several embedding systems. LinkedIn replaced them with one system. A fine-tuned LLM turns every post and every member into an embedding, and the feed fetches the posts closest to you in under 50 milliseconds.

Your side of that match is built from your profile (skills, work history, education) and an ordered list of posts you reacted to, commented on or otherwise acted on. The post side includes its format, the author's name, headline, company and industry, how popular it is so far, and the text itself. New posts get their embeddings in near real time, so a post about breaking news can reach interested readers within minutes.

The change matters most for topic matching. LinkedIn's own example: a member who works in electrical engineering and reads about small modular reactors can now be shown posts on power grids and renewable energy, because the language model knows those fields are related. Keyword matching would have missed the link.

### Step 2: ranking decides the order

The ranking layer is a Generative Recommender, a transformer model that reads more than a thousand of your past interactions in the order they happened. It predicts several actions at once: whether you will click, skip, stay on the post for a long time (LinkedIn calls this a long dwell), like, comment or share. Your history is read as a trajectory, so reading about machine learning on Monday and distributed systems on Tuesday shapes what you see on Wednesday.

LinkedIn also says the ranking runs on professional signals. In an [August 2025 explainer](https://www.linkedin.com/pulse/how-does-linkedin-feed-work-tim-jurka-oxraf/), Tim Jurka, a LinkedIn vice president of engineering who has worked on the feed for over a decade, wrote that the system weighs an author's experience, skills and shared content, and does not use demographic data such as age, race or gender.

| Signal | What it tells the feed | Where LinkedIn describes it |
| --- | --- | --- |
| Profile: industry, experience, skills, location | Which topics you belong to | Tim Jurka, March 2026 |
| Posts you read, liked, commented on, returned to or scrolled past | What you want next | Engineering blog, March 2026 |
| Long dwell | Whether people stayed on a post | Engineering blog, ranking section |
| Popularity, stored as percentile buckets | How a post compares with others | Engineering blog, retrieval section |
| Author's professional identity | Who is qualified to talk about a topic | Tim Jurka, August 2025 |
| Freshness | Whether a post is about something happening now | Engineering blog, serving section |

## What changed in the LinkedIn algorithm in 2025?

2025 was the year LinkedIn moved ranking toward large language models and started explaining the feed in public.

- **January 2025:** LinkedIn researchers published [360Brew](https://arxiv.org/abs/2501.16450), a 150-billion-parameter decoder model trained on LinkedIn data that matched or beat production systems on more than 30 ranking and recommendation problems in offline tests. The paper calls it a research pre-production model, so treat claims that "360Brew runs your feed" with care.
- **August 2025:** Tim Jurka wrote that LLMs now read the context of a post, such as advice, a job opening or career news, and that strong posts can reach interested members who do not follow the author.
- **August 2025:** Rishi Jobanputra of LinkedIn's product team answered the link question in a [short video post](https://www.linkedin.com/posts/rishij_back-to-answer-more-feed-questions-and-reminder-activity-7361548020384976897--dhX/). His takeaway: make sure the post gives enough context to stand alone without the link.
- **September 2025:** LinkedIn's Oscar Rodriguez [wrote](https://www.linkedin.com/pulse/linkedin-where-real-conversations-matter-oscar-rodriguez-ejwee/) that the feed was limiting the reach of posts that look artificially inflated, cracking down on third-party tools such as browser extensions, and showing fewer clickbait posts, overly salesy posts and automated comments.

Reach fell over the same period. A [summary of Richard van der Blom's Algorithm Insights Report 2025](https://mercermackay.com/thinking/blog/a-leaders-guide-to-the-linkedin-algorithm-what-the-data-says), which analyzed more than 1.8 million posts, puts the drop in organic reach at nearly 50% year over year. The same report found text-only posts losing ground to visual formats.

## What changed in the LinkedIn algorithm in 2026?

- **March 12, 2026:** LinkedIn announced the new ranking system in a [post by Tim Jurka](https://www.linkedin.com/pulse/updates-linkedin-feed-focusing-authentic-relevant-tim-jurka-umwnc/) alongside the Engineering blog article. The same post promised to "filter out engagement bait": fewer "Comment 'Yes' if you agree" posts, fewer posts that attach a video unrelated to the text, and fewer recycled posts with little substance. It also announced a test of an Interest Picker so new members can choose topics at sign-up.
- **March 2026:** the same update repeated that LinkedIn works to stop comment automation, including third-party software and browser extensions, which it says are not allowed on LinkedIn.
- **Mid-October 2026:** authors can pin a comment under their own post again, [according to van der Blom](https://www.linkedin.com/pulse/80-first-hour-your-linkedin-post-lying-now-richard-van-der-blom-lrnce).
- **November 2026:** LinkedIn began rolling out a [post performance table](https://www.linkedin.com/pulse/understanding-your-new-analytics-linkedin-sam-corrao-clanon-szake) that shows saves, sends on LinkedIn, followers gained from each post and the split between in-network and out-of-network reach.

The new analytics hint at what LinkedIn counts as a good post. A save means someone wanted to come back to it, and a send means someone forwarded it in a message. Both now sit next to reactions in your stats, and the out-of-network share shows how far topic matching carried a post past your followers.

## Does the first hour after posting still matter?

Less than it used to. In the [November 2026 update of his Algorithm Insights Report](https://www.linkedin.com/pulse/80-first-hour-your-linkedin-post-lying-now-richard-van-der-blom-lrnce), built on more than 1.2 million posts, van der Blom found that 85% of a post's reach now arrives after its first two hours, because the system keeps finding new interested readers for days.

The early window still sets the pace. Of posts that got no real reactions or comments in their first 120 minutes, only 8% ended above the author's own average. And each author's average from earlier months explained 51% of the differences in reach between posts, so a steady record counts for more than any single launch.

His suggested routine for publishing day:

1. Block at least 20 minutes from the moment the post goes live.
2. Reply to as many comments as you can in the first two hours. Fast replies brought 64% more reactions and comments in his data, though under 10% more reach.
3. Pin a comment of your own with an extra fact or an open question.
4. Check the numbers once, at the two-hour mark, and leave a post alone if people are responding.
5. Repost it yourself once, about six hours later, with a line of new context. One self-repost added up to a quarter more reach; a second one cost visibility.
6. Judge the post after seven days, by profile views, connection requests and messages as well as reach.

For the rest of the reach checklist, see our guide on [how to get more impressions on LinkedIn](https://linkedinlikes.com/blog/how-to-get-more-impressions-on-linkedin).

## Do links hurt your reach on LinkedIn?

LinkedIn's position is that a link alone does not trigger a penalty, and Rishi Jobanputra's advice is to write posts that stand on their own. Measured reach still drops. Van der Blom's [July 2026 link study](https://www.linkedin.com/pulse/67-link-penalty-just-changed-most-people-missed-richard-van-der-blom-9iese), covering posts from March to June 2026, found:

| Links in the post | Reach compared with posts without links |
| --- | --- |
| One external link | About 16% lower (18.8% lower in his April 2026 measurement) |
| Two links | About 7% lower |
| A pile of links | About 24% lower |

The old habit of moving the link to the first comment is weaker too. A [roundup of four 2026 link studies](https://eksis.io/blog/linkedin-link-in-post-or-first-comment) quotes van der Blom reporting that comments with links are hidden from everyone but the author in up to 80% of cases under the default comment sort. Practical version: lead with the full idea, put one link at the end, and save link posts for times when the click is the goal.

## What does the LinkedIn algorithm reward in 2026?

Put together, LinkedIn's statements and the 2026 data lead to a short list of habits:

1. **Stay on two or three topics.** Retrieval matches posts to readers by meaning, and an account that keeps posting about the same field gives the model a clear picture of who should see it. Tim Jurka suggests reviewing your own posts by topic, format and timing to see what finds an audience.
2. **Keep your profile current.** Skills, headline and work history feed your member embedding and the author signals on every post you publish.
3. **Give people a reason to stay.** Long dwell is one of the actions the ranking model predicts. A clear first line, short paragraphs and one concrete example keep readers on the post, and our free [LinkedIn hook generator](https://linkedinlikes.com/linkedin-hook-generator) drafts first lines you can test.
4. **Earn comments and saves.** Comments, saves and reposts moved posts fastest in van der Blom's data, and LinkedIn now shows saves and sends in your analytics. End with a question people can answer from their own work, and include something worth keeping, such as a checklist or real numbers.
5. **Drop the bait.** "Comment YES" lines, unrelated video and recycled advice are the three patterns LinkedIn named in March 2026.
6. **Pick formats that hold attention.** In [Socialinsider's 2026 benchmarks](https://www.socialinsider.io/social-media-benchmarks/linkedin) for company pages, native documents averaged a 7.00% interaction rate and multi-image posts 6.45%, ahead of video (6.00%), single images (5.30%), text (4.50%) and link posts (3.25%).
7. **Post when your readers are online, then stay.** Our [best time to post on LinkedIn tool](https://linkedinlikes.com/best-time-to-post-on-linkedin) shows typical windows; the replies in the first two hours matter more than the exact minute.

| Old habit | How the 2026 feed treats it | Source |
| --- | --- | --- |
| "Comment YES and I will send the PDF" | Shown less | LinkedIn, March 2026 |
| Video that has nothing to do with the text | No extra reach | LinkedIn, March 2026 |
| Automated comments and browser extensions | Reach limited, tools not allowed | LinkedIn, September 2025 and March 2026 |
| Judging a post after one hour | 85% of reach comes later | van der Blom, November 2026 |
| Link in the first comment | Comment often hidden under the default sort | van der Blom, via Eksis |

## Where does LinkedInLikes fit in?

The early window rewards people who actually read and reply. On LinkedInLikes, members open each other's posts and react, comment, repost or save by hand from their own LinkedIn accounts. Comments are written in each member's own words, and post owners can't dictate the text or the tone. We never ask for your LinkedIn password, cookies or a browser extension; you only share a public link.

You can ask for [comments](https://linkedinlikes.com/linkedin/comment) or [saves](https://linkedinlikes.com/linkedin/save) on a post you want to give a fair start, spending points you collect by doing the same for other members. It will not turn a weak post into a strong one, so write the post first.

Next step: open your analytics, sort your last 20 posts by saves, and write your next post on the topic at the top of that list.

## Frequently asked questions

### What is LinkedIn 360Brew?

360Brew is a 150-billion-parameter ranking model that LinkedIn researchers described in a January 2025 paper. They called it a research pre-production model. The production feed LinkedIn described in March 2026 uses LLM-based retrieval and a Generative Recommender for ranking.

### Why did my LinkedIn reach drop in 2025?

Most accounts saw it. Richard van der Blom's 2025 report measured organic reach down by nearly 50% year over year, while LinkedIn shifted the feed toward topic matching and showed more posts from outside each member's network.

### Does LinkedIn show my posts to people who don't follow me?

Yes. LinkedIn says strong posts can reach interested members who don't follow the author, and the November 2026 analytics show what share of your reach came from outside your network.

### Does LinkedIn penalize AI-written posts?

LinkedIn has not said it penalizes text written with AI. In March 2026 it said it would show fewer recycled, generic posts with little substance, so a post that reads like everyone else's is likely to travel less, whoever wrote it.

### Do comments count more than likes on LinkedIn?

The ranking model predicts several actions, including long dwell, likes, comments and shares. In van der Blom's 2026 data, comments, saves and reposts moved posts further than reactions alone.

### How long does a LinkedIn post keep getting views?

Days. Van der Blom's November 2026 update found that 85% of a post's reach arrives after the first two hours, so judge a post after about a week.

### Should I put links in the first comment?

It helps less than it used to. Comments with links are often hidden under the default sort, so many creators now put one link at the end of the post and make sure the post makes sense without it.