LinkedIn Algorithm 2026: What Changed in March, and What the 360Brew Hype Gets Wrong

• Amol Ghemud
• LinkedIn Algorithm 2026: What Changed in March, and What the 360Brew Hype Gets Wrong
• Published: March 2, 2026

Contributors: Amol Ghemud
Published: March 2, 2026

Summary

The LinkedIn algorithm 2026 update is real: on 12 March 2026, LinkedIn confirmed LLM-based retrieval, a Generative Recommender ranking model and action against engagement pods and engagement bait. 360Brew, link penalties, hashtag penalties and the golden hour come from research or practitioner data, not confirmed ranking rules, and this guide shows which is which.

The LinkedIn algorithm 2026 story you keep hearing goes like this: LinkedIn threw out its old feed and replaced it with a 150-billion-parameter AI called 360Brew that reads your posts like a human editor. Parts of that are true. Parts of it don’t appear in anything LinkedIn has published.

So I went back to the primary sources: LinkedIn’s engineering post and Tim Jurka’s feed update (both 12 March 2026), LinkedIn’s research papers on arXiv, and practitioner datasets from AuthoredUp, Buffer and ZenABM. This guide separates what LinkedIn has confirmed from what practitioners observe, then shows what to do.

The uncomfortable part: the feed really was rebuilt. If you’re still optimizing for likes, hashtags and pod comments, you’re playing a game LinkedIn says it’s shutting down.

What Changed in the LinkedIn Algorithm Update in March 2026?

On 12 March 2026, LinkedIn said it’s rolling out a new feed ranking system powered by LLMs and GPUs. A single LLM-based retrieval system replaced several separate content sources, and a sequential Generative Recommender model now ranks posts using more than 1,000 of each member’s past interactions.

LinkedIn published the details in an engineering post, Engineering the next generation of LinkedIn’s Feed, and a member-facing feed update from Tim Jurka. Together they describe 5 changes.

1. One LLM-based retrieval system

The feed used to pull candidates from separate sources, including your network’s activity, trending posts in your geography and collaborative filtering. LinkedIn replaced that mix with a single system built on LLM embeddings. Because the model brings world knowledge, it can connect related topics even when posts use different words, which helps you discover authors you don’t follow.

2. Sequential ranking with a Generative Recommender

Ranking used to score each impression on its own. The new model treats your feed history as a sequence to see where your interests are heading. A February 2026 arXiv paper by LinkedIn engineers describes a transformer-based sequential ranker, Feed SR, that replaced the previous production model, serves the majority of feed traffic, and lifted time spent by 2.10% and likes, comments and reshares by 3.52% in A/B tests.

3. Action on pods and comment automation

Jurka’s update says LinkedIn is working to make engagement pods ineffective and curb comment automation. More on that below.

4. Less engagement bait

LinkedIn said it would show fewer “Comment ‘Yes’ if you agree” posts, fewer posts that pair a video with unrelated text, and fewer recycled thought leadership posts.

5. An Interest Picker for new members

LinkedIn is testing an Interest Picker at sign-up, so new members can choose topics such as Leadership and get a personalized feed from day 1.

Is 360Brew the LinkedIn Algorithm? What’s Actually Confirmed

360Brew is real, but LinkedIn hasn’t confirmed it runs your feed. It’s a 150B-parameter research model from a LinkedIn paper that is now withdrawn on arXiv, and neither of LinkedIn’s 12 March 2026 feed announcements mentions it by name.

What the 360Brew paper says

The 360Brew paper, first submitted in January 2025, calls 360Brew V1.0 a “research pre-production model” with 150B parameters, trained on LinkedIn’s data. It handled over 30 predictive tasks through a text interface and matched or beat production systems on offline metrics. That’s a research result, not a deployment announcement. arXiv now marks every version as withdrawn over a licensing issue.

What LinkedIn’s feed announcements say

The March 2026 engineering post never uses the name 360Brew. The Feed SR paper goes further: an LLM-Ranker that fed post features to an LLM as text “never achieved superior online performance” over the production model, partly because it struggled to encode network relationships in a prompt.

In our view, 360Brew stuck because it gave a real shift a memorable name. The lesson survives the correction: LLMs now help LinkedIn understand what your post is about. Just don’t build a strategy on specifics LinkedIn hasn’t published, such as “4 alignment pillars” or a “Topic Authority Score”.

LinkedIn Algorithm 2026 Explained: Which Signals Does LinkedIn Use?

LinkedIn names 2 groups of signals: what you share on your profile (industry, experience, skills and geography) and how you engage with content over time. Its ranking model predicts passive actions such as clicks, skips and long dwells, and active ones such as likes, comments and shares.

Dwell time is confirmed, and has been since 2020

LinkedIn engineers wrote about using dwell time to improve feed ranking in May 2020. Posts viewed for less than a set threshold count as “skipped”, which lowers their score. In 2026, long dwells are still among the actions the ranker learns from, with the threshold depending on post type.

The old game was getting as many people as possible to tap like, fast. The new game is making the right person stop scrolling and actually read.

Likes, comments and saves

LinkedIn hasn’t published how much each action weighs. AuthoredUp’s algorithm guide says comments count twice as much as likes, that fewer than 3% of posts get saved, and that 1 save gives a post 5x more reach than 1 like. Those are practitioner findings. LinkedIn’s March post and the Feed SR paper don’t name saves among the actions they list, so treat saves as a strong observed signal, not a confirmed weight.

Your network’s relevance

LinkedIn says affinity features, which capture how a viewer relates to other members, remain very effective in ranking, and that the feed also surfaces posts from people you’ve never connected with. So connect with people in your niche rather than chasing connection counts.

Confirmed vs Observed: LinkedIn Algorithm 2026 Claims Fact-Checked

LinkedIn has confirmed LLM-based ranking, dwell time as a signal, and action against pods and engagement bait. The link penalty, hashtag penalty, golden hour, best posting frequency and reach-decline figures come from practitioner datasets, and those don’t always agree.

Is LinkedIn reach really down?

AuthoredUp says average reach is down 34% in 2025. LinkedIn hasn’t published reach trends, and Jurka’s update says its anti-automation work isn’t about limiting reach. The question isn’t “is LinkedIn dying”. It’s whether your content is worth distributing.

Does LinkedIn penalize external links?

LinkedIn has never publicly confirmed a link penalty, and we couldn’t trace the widely quoted “60% penalty” to any LinkedIn source. AuthoredUp found posts with 1 link get the lowest reach, while posts with 4+ links see 3 to 5x higher median reach than single-link posts. Buffer’s March 2026 format study found LinkedIn link posts earn a 3.81% median engagement rate, above text-only posts at 3.18%. Test link placement on your own account.

Do hashtags still work?

AuthoredUp found hashtags had “no impact” on reach and that 3 to 5 can slightly reduce visibility. LinkedIn hasn’t published a hashtag rule, but it has confirmed its LLMs understand what a post is about from the text. If you’re still adding 10 hashtags to every post, you’re wasting space. Use natural keywords instead.

How often should you post?

Buffer’s analysis of over 2 million posts found 2 to 5 posts a week is the sweet spot for sustainable growth, 11+ a week added nearly 17,000 impressions per post versus posting once, and LinkedIn doesn’t punish frequent posting. AuthoredUp recommends 3 to 5 posts a week and warns against posting more than once in 24 hours. Our take: start at 2 to 5 and scale only if quality holds.

Is the golden hour real?

AuthoredUp, citing Richard van der Blom’s Algorithm Insights 2025 research, says the first 30 to 60 minutes after posting are crucial. LinkedIn hasn’t confirmed a fixed window, though its engineering post says a post’s data refreshes within minutes as it gains engagement. AuthoredUp also notes posts stay visible for days, so a slow first hour isn’t fatal.

Engagement Pods Are Dead. LinkedIn Said So.

LinkedIn has publicly said it’s working to make engagement pods ineffective and curb comment automation. Its Help Center says third-party software and browser extensions that automate activity aren’t allowed.

I know some of you are still in pods. LinkedIn’s Help Center page on automated activity says it doesn’t allow software or extensions that scrape, modify or automate activity, and accounts restricted for it must remove that software.

The replacement is simple but not easy: genuine, strategic commenting. In our experience, a daily block of thoughtful, specific comments on posts in your niche builds visibility without risking your account. We haven’t found a verified benchmark for how many comments it takes, so track your own profile views.

Which LinkedIn Post Formats Perform Best in 2026?

Document posts (PDF carousels) lead on engagement in Buffer’s March 2026 study, with a 21.77% median engagement rate on LinkedIn. AuthoredUp’s reach data for personal profiles puts polls first with a 1.64x reach multiplier and documents next at 1.45x.

Document posts and polls

Buffer found LinkedIn carousels earned 196% more engagement than video and 585% more than text-only posts. Carousels with a clear payoff on every slide hold readers, which fits LinkedIn’s use of dwell time. Polls regained the top reach spot for personal profiles, but AuthoredUp warns overusing them can look spammy.

Video

LinkedIn’s Help Center confirms a full-screen vertical video experience with a video tab, and says any video you post may appear there based on relevance and performance. Buffer puts LinkedIn video second for engagement at 7.35%, and AuthoredUp gives video a 1.10x reach multiplier on personal profiles. My take: video is worth testing, but the easy-reach story is overstated.

Text and images

Buffer ranks images at 6.52% and text-only posts last at 3.18% median engagement. AuthoredUp’s reach multipliers for personal profiles point the same way: images at 1.18x, text at 0.88x.

The LinkedIn Features Nobody’s Using (and Why They Should)

Newsletters and Thought Leader Ads give you distribution that doesn’t rely only on feed ranking.

LinkedIn Newsletters

When you create a newsletter, LinkedIn lets you invite all your connections or followers to subscribe, and subscribers get push, in-app and email notifications after each edition, according to LinkedIn Help. If you’re a founder or consultant without one, you’re leaving your most dependable channel untouched.

Thought Leader Ads

LinkedIn’s Thought Leader Ads help page confirms companies can sponsor public posts from employees, creators and 1st, 2nd or 3rd-degree+ connections, with permission. ZenABM’s 2026 benchmark of 2,828 ads found a 2.68% median CTR for TLAs versus 0.42% for single image ads, and 77% cheaper landing page clicks, though it notes the CTR comparison isn’t like for like. Before you budget, check how much LinkedIn ads cost.

How to Adapt Your Strategy to the LinkedIn Algorithm in 2026

Write for relevance and attention, not reactions. Post consistently on topics your profile backs up, earn real comments, and test formats, links and cadence on your own account.

Make your profile and posts tell the same story

LinkedIn’s post prompts include the author’s headline, company and industry alongside the text. In our view, a “SaaS expert” who posts motivational quotes sends mixed signals, and posting consistently in a single lane makes you easier to match with the right readers.

Write for dwell time, not bait

Lead with a specific claim and give people a reason to finish. Drop “comment YES” prompts and recycled platitudes. For hooks and structure, see our guide to optimizing content for LinkedIn’s 360Brew-era feed. If you draft LinkedIn posts with AI, keep a human edit pass.

Test before you believe any rule

Track engagement rate per post with our LinkedIn engagement rate calculator, and run link and cadence tests for 4 to 6 weeks before you decide. Care about AI search too? See how LinkedIn content earns AI citations.

LinkedIn isn’t broken. The old playbook is.

LinkedIn Algorithm 2026: FAQs

Your Next Move: Audit Your LinkedIn Playbook

The LinkedIn algorithm 2026 rewards relevance you can prove, not tricks you can copy. Pick 3 posts from last month and ask: would the right reader stop, read and respond?

What’s the biggest change you’ve noticed on LinkedIn in the last 6 months? If you want help turning it into a growth marketing plan, book a strategy call with upGrowth.

About the Author

Amol Ghemud

Optimizer in Chief

Amol has helped catalyse business growth with his strategic & data-driven methodologies. With a decade of experience in the field of marketing, he has donned multiple hats, from channel optimization, data analytics and creative brand positioning to growth engineering and sales.