Quick answer: The LinkedIn algorithm is the set of machine-learning models that decide which posts appear in each member's feed and in what order. LinkedIn says it ranks per viewer, using signals about identity, content and past behavior. Everything else, such as saves outranking likes, is creator observation. Write for the reader's relationship with you, then test what works.
Key facts
- The most detailed LinkedIn Engineering post we found that lists feed signals is dated 29 March 2018. It groups signals into identity, content and behavior.
- A February 2026 paper by LinkedIn authors says a transformer-based ranker called Feed SR replaced a DCNv2-based ranker and had served most Feed traffic for over three months.
- The same paper reports A/B gains of +2.10% time spent and +3.52% in likes, comments or reshares against the earlier model.
- LinkedIn's Help Center says the feed shows Top posts by default, and members can switch to Recent.
- On LinkedIn Page analytics, an impression counts when at least 50% of a post is in view for at least 300 milliseconds.
- We found no official LinkedIn announcement dated December 2025 that describes a ranking change.
LinkedIn algorithm 2026: how it works in plain terms
LinkedIn does not publish a weighting table, so no one outside the company can give you exact formulas. What it does publish is architecture: how posts are gathered, scored and ordered. That is enough to reason from.
In plain terms, the feed is built per viewer. Posts are pulled from your network and from topics you engage with, scored by a ranking model, then ordered into the list a member sees. Readers who ask how the LinkedIn algorithm works in 2026 often expect a list of tricks. The honest answer is that ranking is personal, so the same post can rank high for one viewer and low for another.
Two consequences follow. First, reach comes from relevance to specific people, not from a universal score. Second, your history with those people matters, because the models learn from what each member did before.
What has LinkedIn actually said, and when?
The table below is the dated record. Each row is tagged so you can separate what LinkedIn stated from what creators report.
| Date | Source | What it says | Tag |
|---|---|---|---|
| 29 Mar 2018 | LinkedIn Engineering blog | Signals fall into identity, content and behavior. Clicks alone are a poor proxy for value. The team invests in models that reward members who create quality content over time. | LinkedIn stated |
| Jan 2025 | LinkedIn research paper, 360Brew | Describes a 150-billion-parameter decoder-only model for personalized ranking and recommendation. The paper does not tell us whether it serves the live feed. | LinkedIn stated (research) |
| Oct 2025 | LinkedIn research paper on feed retrieval | Describes retrieval for the feed built with causal language models. | LinkedIn stated (research) |
| Dec 2025 | None found | No official announcement found. Claims in circulation come from creator and vendor posts. | Observed by creators |
| Feb 2026, revised Sep 2026 | LinkedIn research paper, Feed SR | A transformer-based sequential ranker replaces a DCNv2-based ranker and serves most Feed traffic. A/B results: +2.10% time spent, +3.52% likes, comments or reshares. | LinkedIn stated |
Notice what is missing: any statement that comments outrank likes, that links cut reach or that a post must be a certain length. Those claims circulate widely, but they are not in LinkedIn's own record.
LinkedIn algorithm update December 2025: what is confirmed?
We searched for an official LinkedIn algorithm update December 2025 and found none. What we found were creator posts and vendor blogs describing changes they observed. Treat those as hypotheses.
The nearest official material is the October 2025 retrieval paper and the February 2026 ranker paper. Both describe research and engineering work, not a dated product announcement.
How to read LinkedIn algorithm news
Algorithm news usually follows a pattern: a creator sees reach move, a vendor writes it up, and the claim spreads. Before you change your habits, ask four questions.
- Does the source quote LinkedIn directly, with a link?
- Is the claim dated?
- Does it rest on one account's results or on many?
- Would the advice still be sensible if the claim were false?
If the answer to the last question is yes, adopt it. If not, wait.
Which signals matter, and which are myths?
This table sorts common claims by evidence. "Unconfirmed" means LinkedIn has not said so in anything we found.
| Signal or claim | Status | What we can say |
|---|---|---|
| Your relationship with the viewer | LinkedIn stated | Behavior signals include whom a member interacts with most |
| Topic, language and age of the post | LinkedIn stated | Content signals include topic, age and language |
| A member's history of interactions | LinkedIn stated | The 2026 ranker is sequential, meaning it reads past behavior in order |
| Comments weigh more than likes | Unconfirmed | Creator observation only |
| Saves weigh more than likes | Unconfirmed | Creator observation only |
| Dwell time on a post | Unconfirmed | The 2018 post cites time spent reading for articles, not posts |
| Links in the post body cut reach | Unconfirmed | Contested among creators |
| Hashtag count changes reach | Unconfirmed | No official statement found |
| Engagement pods help | Unconfirmed | A pod gives you comments from people who may not care about the topic |
Charts on engagement rate are easy to find, but LinkedIn defines it as clicks, likes, comments, shares and follows divided by impressions. That is a reporting metric, not a published ranking factor.
Do formats like carousels and video change reach?
The LinkedIn algorithm has no published format weights. What you can reason from is behavior: a format that gives a viewer more reasons to interact leaves the system with more evidence about that viewer's interest.
- Text posts: quick to read and easy to answer, so they suit opinions and short lessons.
- Documents and carousels: each swipe is an interaction. Our LinkedIn carousel guide covers sizes and structure.
- Video: works when the first seconds carry the point, since viewers decide fast.
- Polls: cheap to answer, though a poll with a vague question teaches the system little.
Treat this as reasoning, not data. Test each format against your own analytics for a month before you commit.
What should you change in your posts?
These are our recommendations, built on the signals LinkedIn has stated. They don't depend on any rumored weighting.
- Open with the point. The preview before the "see more" link is what viewers judge. Put the claim or the result first.
- Ask for a specific reply. "Which of these three do you use?" gets more useful answers than "Thoughts?".
- Answer comments early and by name. It builds the relationship history the models learn from.
- Keep links honest. If a link is the point of the post, include it. If you suspect a penalty, run a two-week test of links in the body against links in a follow-up comment.
- Stay on topic. A feed that repeats three subjects teaches the system what to show you for.
- Pick times from your own data. See our note on the best time to post on LinkedIn for the method.
Post-audit checklist
Score each recent post out of 10, one point per yes.
- The first line states the claim or result.
- One idea per post.
- A specific question or prompt to reply to.
- Comments answered within a day.
- A format chosen for the idea, not by habit.
- No link, or a link that earns its place.
- No more than a few hashtags, if any.
- The topic fits your last five posts.
- A named person or company appears only if relevant.
- Impressions checked against the 300 ms and 50% in-view definition before you compare posts.
Posts that score 8 or more are your templates. Posts under 5 need a rewrite, not a repost.
How Publiqa helps with LinkedIn posts
Publiqa turns one idea, article or competitor link into content for 12 channels, LinkedIn among them. It adapts the same thesis to LinkedIn's 3,000-character limit, and it builds carousels of 5 to 10 slides in 4:5 format.
A practical workflow for this topic: scan a competitor's public page with Competitor Radar to see gaps and counter-angles, then open Campaign Studio with the angle you like. Creating the core plus 12 platform versions costs about 10 credits. Review the LinkedIn version against the checklist above, and plan the week in the Smart Planner, which suggests posting times you should test rather than trust.
Publiqa sends content to your channels through outgoing webhooks and headless CMS integrations, which are live. Direct publishing to social networks is rolling out, so check the channels and integrations documentation for the current LinkedIn status.
Measure and iterate
Treat any reported change in the LinkedIn algorithm as a hypothesis until your own numbers agree. Use LinkedIn's own analytics definitions, and compare like with like. An impression counts when a post is visible for 300 milliseconds with at least half of it on screen. Repeat views by one member count again, so look at unique viewers too.
Run a four-week loop. Change one variable per week, such as the opening line, and log impressions, comments and profile visits in a sheet. Keep what repeats across several posts, and drop what appeared once.
FAQ
How does the LinkedIn algorithm work?
It is a set of machine-learning models that retrieve candidate posts, score them for each viewer and order them. LinkedIn's engineering posts describe identity, content and behavior signals. A 2026 paper describes a sequential transformer ranker. Exact weights are not public.
Did LinkedIn change its algorithm in December 2025?
We found no official announcement from LinkedIn dated December 2025. Creator and vendor posts describe observed changes, but they are not confirmation. The closest official items are research papers from October 2025 and February 2026.
Does the LinkedIn algorithm weigh comments above likes?
Many creators report that they do, but LinkedIn has not confirmed it in anything we found. Comments are still worth earning, because they start conversations and deepen the relationship with the people who matter to you.
Does the LinkedIn algorithm penalize links in a post?
This is unconfirmed. Creators disagree, and LinkedIn has not published a rule. Test it on your own account: post two similar updates, one with the link in the body and one without, and compare impressions after a week.
How often should you post on LinkedIn?
No official figure exists. A common starting point is two to three posts a week, which is our recommendation and not a platform rule. Hold that rhythm for a month, then adjust using your analytics.
Your four-week test plan
Week one, audit your last ten posts with the checklist. Week two, rewrite openings only. Week three, test one format change. Week four, review the log and keep the winner.
If you want drafts to test, start free with 200 credits. No credit card required. Turn one idea into content for 12 channels and compare the LinkedIn version with what you would have written alone.
Sources
- LinkedIn Engineering: A Look Behind the AI that Powers LinkedIn's Feed (29 March 2018)
- arXiv: An Industrial-Scale Sequential Recommender for LinkedIn Feed Ranking (2602.12354)
- arXiv: Large Scale Retrieval for the LinkedIn Feed using Causal Language Models (2510.14223)
- arXiv: 360Brew, a Decoder-only Foundation Model for Personalized Ranking and Recommendation (2501.16450)
- LinkedIn Help: Sort your feed by Top or Recent posts
- LinkedIn Help: Content analytics for your LinkedIn Page