TL;DR
- LinkedIn replaced its ranking system with 360Brew, a very large language model reported at around 150 billion parameters, which shifted distribution from a relationship graph toward an interest graph.
- Reach fell across the board. Independent analyses of millions of posts recorded views down roughly 50%, engagement down 25%, and follower growth down 59%.
- Posting more no longer works as a lever. Frequency was a reliable tactic under the previous system and is close to neutral under this one; topical consistency replaced it.
- Saves and considered comments now carry the weight that likes used to carry, because the model is inferring value rather than counting reactions.
- Follower count matters less than it did. An interest graph shows your post to people interested in your topic whether or not they follow you, which is bad news for accumulated audiences and good news for narrow expertise.
Contents
- What actually changed under the hood
- The measured decline and what it means
- Six rules that follow from an interest graph
- Formats that hold up in current conditions
- What stopped working
- Measuring content on pipeline rather than reach
- FAQ: The LinkedIn Algorithm in 2026
What actually changed under the hood
For most of LinkedIn's history, distribution ran on a relationship graph. The system asked who you were connected to, how strongly, and how they had interacted with you before, then showed your post to that network and expanded outward if early engagement was strong. This is why posting frequency worked: more posts meant more chances for your existing network to engage, and engagement was the fuel for expansion.
The platform replaced that ranking approach with 360Brew, a large language model reported at roughly 150 billion parameters, which changes the question being asked. Instead of asking who is connected to this author, the system reads the content and asks who would find this valuable.
That is a shift from a relationship graph to an interest graph, and almost every practical consequence flows from it.
| Dimension | Relationship graph | Interest graph |
|---|---|---|
| Primary question | Who knows this person? | Who cares about this topic? |
| Value of followers | High | Reduced |
| Value of posting frequency | High | Close to neutral |
| Value of topical consistency | Low | High |
| Value of likes | High | Reduced |
| Value of saves and dwell | Moderate | High |
| Reach ceiling for narrow expertise | Limited by network | Limited by audience size |
The important asymmetry: an interest graph is harsher on accumulated general audiences and kinder to narrow, consistent expertise. A large follower count built on varied content is now a weaker asset than it was. A small following built on a single well-defined subject is a stronger one.
The measured decline and what it means
Several independent analyses converged on the same picture, which is unusual and worth taking seriously.
Richard van der Blom's Algorithm Insights work, drawing on a sample in the region of 1.8 million posts, and AuthoredUp's analysis across more than 3 million posts, both recorded substantial declines. Scott M. Graffius's post half-life research across 5.6 million posts spanning 11 platforms provides the decay context.
The headline figures: views down roughly 50%, engagement down roughly 25%, and follower growth down roughly 59%.
Three observations about those numbers.
Views fell faster than engagement. Views down 50% against engagement down 25% means the people who do see your posts are engaging at a higher rate than before. The system is showing your content to fewer people but to better-matched ones. That is precisely what you would expect from an interest graph, and it is the single most useful detail in the dataset.
Follower growth fell hardest. Down 59% is the largest of the three declines, and it makes sense: if distribution no longer depends primarily on follower relationships, the system has less reason to surface follow prompts, and users have less reason to follow rather than simply engage with what appears.
Benchmarks from 2024 are actively misleading now. Any content plan built on impression targets from two years ago will read as failure regardless of quality. Rebaseline before you evaluate anyone's performance, including your own.
| Metric | Change | Interpretation |
|---|---|---|
| Views | Down roughly 50% | Narrower, better matched distribution |
| Engagement | Down roughly 25% | Higher engagement rate per view |
| Follower growth | Down roughly 59% | Following matters less to the system |
Six rules that follow from an interest graph
These are consequences of the architecture rather than tactics, which is why they are likely to hold as the model is updated.
1. Narrow the topic range and hold it
The model infers what you are authoritative about from what you consistently publish. Posting across five unrelated subjects means the system has no confident read on where to place any of them. Pick a subject narrow enough that a stranger could describe it in one sentence after reading three of your posts.
This is uncomfortable because it feels limiting. It is the single highest-return change available under the current system.
2. Optimise for saves and dwell, not likes
A like costs nothing and signals little. A save signals that someone expects to need this again, and dwell time signals that they actually read it. A model reading content for value weights the expensive signals higher.
Practically: write posts worth keeping. Frameworks, benchmark sets, checklists, and specific breakdowns get saved. Observations and opinions get liked and forgotten.
3. Treat comments as content, not as engagement bait
A considered comment carries far more weight than a reaction, both as a distribution signal and because the model reads it. This cuts both ways: generic comments on other people's posts do nothing for you and mark you as low-value activity, while a substantive comment can reach further than a mediocre post.
Reply to comments on your own posts within the first hour. Early engagement remains the strongest distribution signal you can directly influence.
4. Stop using frequency as a strategy
Daily posting was a reliable lever under the relationship graph and is close to neutral now. Worse, it usually forces topic drift and quality decline, both of which are actively penalised under the current system.
Three strong posts weekly within a narrow subject outperform daily posting across a wide one. This is a real change in the correct behaviour, not a stylistic preference.
5. Write for the buyer, not for your peers
An interest graph will happily deliver your content to other practitioners in your field, who will engage generously and never buy anything. Peer applause is the most common way content programmes fail commercially, and the interest graph makes it easier to fall into because the system optimises for engagement rather than for your commercial interest.
Check your engaged audience periodically. If it is mostly people who do your job rather than people who buy from you, the topic is calibrated to the wrong reader.
6. Post from personal profiles
Personal profiles reach roughly eight to ten times the audience of company pages. This was true before the change and remains true after it. Company page posting is appropriate for announcements and for nothing else.
Formats that hold up in current conditions
| Format | Current performance | Why |
|---|---|---|
| Carousels and documents | Strong | High dwell time, high save rate |
| Text with a specific framework | Strong | Saved and referenced later |
| Text opinion posts | Moderate | Liked, rarely saved |
| Native video | Moderate | Good dwell, weaker save behaviour |
| External link posts | Weak | Distribution penalty persists |
| Polls | Weak | Engagement without value signal |
Carousels and documents are the strongest general-purpose format because they produce both dwell time and saves, which are the two signals the current system weights most heavily.
Frameworks outperform opinions. A post containing a usable structure gets saved. A post containing a view gets liked. Under a system that reads content for value, the first is worth substantially more.
External links still suppress reach. This predates the ranking change and has not improved. Put the link in the first comment or, better, write the substance natively and let people ask.
Polls are effectively dead as a reach tactic. They generated engagement without value under the old system, which is exactly the pattern a content-reading model is designed to discount.
What stopped working
Worth naming explicitly, because these tactics are still widely recommended by people working from stale playbooks.
Engagement pods. Coordinated early engagement from an unrelated group signals nothing to a model reading content, and the mismatch between engagement source and topic is detectable.
Hook formulas. The one-line opener followed by white space was a relationship-graph tactic aimed at maximising early clicks on "see more". It still marginally helps dwell time and no longer substitutes for substance.
Posting at optimal times. This mattered when early network engagement drove expansion. Under an interest graph, distribution is spread over a longer window and matched to interest rather than to who happens to be online.
Broad topical range to reach more people. Precisely backwards now. Range dilutes the model's read on your authority and reduces distribution for everything you post.
Chasing follower count as the primary metric. Follower growth fell 59% and matters less to distribution than it did. Followers remain useful as an owned audience, but growing them is no longer the mechanism by which reach increases.
Measuring content on pipeline rather than reach
Given that reach declined roughly 50% across the platform, measuring content on impressions guarantees a story of decline regardless of how well the programme is working. Three better measures.
Pipeline influenced. Track whether people at target accounts engaged with content before entering a sales conversation. This requires the content audience and the account list to be connected, which is a systems question rather than a content one.
Outbound performance lift among engaged prospects. People who have seen your content accept connection requests and reply to email at materially higher rates. Segment your outbound results by prior content engagement, and the difference is usually the clearest evidence available that the content programme is working.
Saves per post. The strongest available proxy for whether content is genuinely useful, and the metric most aligned with what the current system rewards. It is also harder to game than impressions or likes, which makes it a better internal target.
| Metric | Use it for | Reliability |
|---|---|---|
| Pipeline influenced | Programme value | High |
| Outbound lift among engaged | Programme value | High |
| Saves per post | Content quality | Good |
| Considered comments | Content quality | Good |
| Impressions | Little | Low |
| Follower growth | Little | Low |
The underlying point: content on LinkedIn is now most valuable as a distribution asset for the rest of your motion rather than as a standalone acquisition channel. A prospect who recognises your name from three useful posts is a different prospect when your outbound arrives, and that effect survived the ranking change entirely intact.
LinkedIn GTM Playbook: Download Free
Get the practical framework for applying this article to your GTM system.
Download the LinkedIn GTM Playbook →
FAQ: The LinkedIn Algorithm in 2026
What changed in the LinkedIn algorithm in 2026?
LinkedIn replaced its previous ranking approach with 360Brew, a large language model reported at around 150 billion parameters, which shifted distribution from a relationship graph to an interest graph. Instead of asking who is connected to the author and how strongly, the system reads the content and asks who would find it valuable. Almost every practical consequence follows from that change: topical consistency now matters more than posting frequency, saves and dwell time matter more than likes, and follower count matters less than it did.
How much did LinkedIn reach actually decline?
Independent analyses converged on views down roughly 50%, engagement down roughly 25%, and follower growth down roughly 59%. The samples are large, including Richard van der Blom's Algorithm Insights work across around 1.8 million posts and AuthoredUp's analysis across more than 3 million. Note that views fell faster than engagement, which means the smaller audience now seeing your posts engages at a higher rate, exactly as an interest graph would predict. Benchmarks from 2024 are actively misleading and should be rebaselined.
Does posting frequency still matter on LinkedIn?
Much less than it did, and treating it as a strategy is now counterproductive. Frequency was a reliable lever under the relationship graph because more posts meant more chances for your network to engage. Under a content-reading model, three strong posts weekly within a narrow subject range outperform daily posting across a wide one, because daily posting usually forces topic drift and quality decline, both of which the current system penalises. Consistency of topic replaced consistency of cadence.
Should I optimise for likes or saves?
Saves, along with dwell time and considered comments. A like costs nothing and signals little, whereas a save signals that someone expects to need the content again and dwell time signals that they actually read it, so a model inferring value weights the expensive signals higher. Practically this means writing posts worth keeping: frameworks, benchmark sets, checklists, and specific breakdowns get saved, while observations and opinions get liked and forgotten.
Do followers still matter on LinkedIn?
Less than they did, which is why follower growth fell hardest of the three measured declines at roughly 59%. An interest graph shows your post to people interested in the topic whether or not they follow you, so a large following built on varied content is a weaker asset than it was, while a smaller following built on one well-defined subject is a stronger one. Followers remain useful as an owned audience, but growing them is no longer the mechanism by which reach increases.
How should I measure a LinkedIn content programme now?
Not on impressions, since platform-wide reach fell roughly 50% and any impression-based target will show decline regardless of quality. Measure pipeline influenced by tracking whether people at target accounts engaged with content before entering a sales conversation, and measure the outbound lift among engaged prospects, since people who have seen your posts accept connection requests and reply at materially higher rates. Saves per post is the strongest available proxy for content quality and is harder to game than impressions.
Yananai A. Chiwuta
Author
Yananai Chiwuta is a Go-To-Market Architect and founder of Forma Nôrden. He builds Go-To-Market growth systems for B2B companies selling into enterprise and mid-market accounts.
Celine Sky
Reviewer
Celine ensures technical accuracy and strategic alignment across all Forma Nôrden playbooks and resources.
