LinkedIn algorithm 2026: what changed and what actually works

The short version

LinkedIn's feed stopped being a popularity contest between your connections and became an interest-matching system. In March 2026 LinkedIn described, on its engineering blog, a rebuilt feed that uses LLMs and GPUs to understand what a post is about and who it is relevant to, then ranks it against each member's recent activity rather than mostly their history. Social Media Today's summary of the update is the clearest public write-up.

For a founder or executive, that has three practical consequences:

• Who you are matters as much as what you post. Profile details, skills and geography are inputs. A post about pricing strategy from someone whose profile says "B2B SaaS CEO" has a clearer audience than the same post from a vague profile.
• Consistency of subject compounds. The system learns what you are about from your last several weeks of posts. Ten posts on one theme teach it more than ten posts on ten themes.
• Cheap engagement is now a negative signal. LinkedIn said it would filter "Comment YES if you agree" prompts, unrelated video bait and recycled thought leadership.

What LinkedIn actually confirmed in 2026

Most algorithm content online is vendor opinion. The parts LinkedIn itself has said publicly are narrower, and worth separating out:

What "360Brew" is, and what it is not

You will see the 2026 algorithm called "360Brew" everywhere. The name comes from a LinkedIn research paper describing a 150-billion-parameter ranking model, which the paper itself called a research pre-production model. As upGrowth's breakdown notes, that paper was later marked withdrawn on arXiv, and LinkedIn's March 2026 feed posts do not use the name.

Treat "360Brew" as shorthand for the LLM-era feed, not as a documented system with known weights. Anyone quoting precise percentages for how 360Brew scores a post is guessing.

The signals that matter most now

Topic consistency

The interest graph rewards authors it can categorize. Pick two or three content pillars tied to the business you want to grow, and keep roughly four out of five posts inside them. A fintech founder who alternates between payments regulation, hiring and cricket is teaching the model three unrelated audiences. If you need help choosing those pillars, that is the core of our personal branding strategy work.

Profile and content match

Because profile fields are ranking inputs, your headline, About section and experience should describe the same expertise your posts discuss. Run your profile through our free LinkedIn profile grader and check one thing in particular: would a stranger reading only your headline predict what your last five posts were about?

Dwell time and saves

Time spent reading is a stronger signal than a quick like. Saves are rare, which is exactly why they carry weight: AuthoredUp data cited by upGrowth puts saves at under 3% of posts, with one save worth several times the reach of one like. Content that people save is reference material: frameworks, numbers, checklists, decision tables, hard-won specifics.

Comments with substance

Comments count for more than reactions, and longer, specific comments appear to count for more than "Great post". The fastest legitimate way to earn them is to end with a precise question that only people in your field can answer, then reply to every serious comment in the first hour or two.

External links: real effect, exaggerated number

The claim that link posts lose "60% of reach" circulates widely, but nobody has traced it to a LinkedIn source. The data is mixed. AuthoredUp found single-link posts had the lowest median reach, while Buffer found link posts had a slightly higher median engagement rate than text-only posts. Our working rule:

• Do not put a link in a post whose job is reach.
• When the link is the point (a report, a launch, a hiring page), put it in the post, accept some reach loss, and make the post worth reading without clicking.
• Moving the link to the first comment is no longer a reliable workaround.

Engagement pods, bait and AI slop

LinkedIn's March update targeted engagement bait and automation, and in July 2026 it added a member-facing "Seems like AI slop" report option. LinkedIn later said content it classifies as slop was getting about 40% fewer views. We cover that change and how readers spot AI writing in our post on the LinkedIn AI slop button.

The practical message is blunt: a pod of 20 people liking each other's posts within minutes is now a pattern the system can see, and generic AI-written posts are a pattern members can report.

Which formats perform in 2026

Format data comes from tools and vendors, measured differently, so treat it as direction rather than law.

Figures as reported in upGrowth's 2026 summary. Other datasets, such as the ones Dataslayer compiles, show the same ranking (documents first, text last) with smaller absolute numbers.

Why your reach dropped, and why that is not the whole story

Richard van der Blom's Algorithm Insights research, cited in Dataslayer's February 2026 roundup, reported views down about 50% year over year for the creators he tracks. If your impressions fell, you are in the majority.

Two things can be true at once. Average reach per post is down, and the reach that remains is better targeted. A post that gets 4,000 impressions from founders and CFOs in your market is worth more than one that got 20,000 from a general audience in 2023. Measure profile views from your ICP, connection requests from relevant titles and inbound DMs, not impressions alone.

How often to post

Buffer's analysis of more than 2 million posts found 2 to 5 posts a week to be the sweet spot, with higher frequency still adding impressions. For most busy founders, 3 posts a week on two or three pillars, every week, beats 7 posts one week and none the next. Consistency is what lets the model learn your topic.

A practical checklist for 2026

1. Rewrite your headline so it names who you help and the outcome, using the same words your posts use.
2. Choose 2 to 3 content pillars and keep about 80% of posts inside them.
3. Write hooks that make a specific promise in under 140 characters (see our hook examples).
4. Put a number, a name or a real situation in every post.
5. Use documents for frameworks and data; use text for stories and opinions.
6. Leave links out of reach posts.
7. End with a question your peers can answer, not "Agree?"
8. Reply to serious comments within the first hour or two.
9. Comment thoughtfully on 5 to 10 ICP posts a day; it trains the same interest graph.
10. Never join a pod and never post AI text you did not rewrite.
11. Review saves, profile views and DMs monthly, not just impressions.

What we see working on founder accounts

Our co-founder Deepanshu Udhwani built his own LinkedIn from zero to 300K+ average impressions per post, with top posts at 521,800 and 473,643 impressions. The pattern behind it is the one above: a narrow set of topics, specific stories and numbers, and no bait. The case study walks through it.

If you want that system run for you, our LinkedIn ghostwriting service turns one monthly interview into 3 to 5 posts a week, written by humans and approved by you. It is built for people like SaaS founders who know they should post and do not have the hours.