LinkedIn's 360Brew Cut 98% of Creators' Reach and Made Saves Worth 5 Likes
LinkedIn's feed is now ranked by 360Brew, a 150-billion-parameter in-house LLM that reads posts by semantic meaning instead of engagement velocity. Since the rollout, 98% of creators saw reach decline across 621,000 tracked posts, per AuthoredUp. One save now drives 5 times more reach than a like, which means the playbook that worked in 2024 is actively hurting distribution for most B2B teams.
What 360Brew actually does to your post
Before, LinkedIn's feed rewarded early engagement velocity. You got likes in the first 60 minutes, you got reach. The mechanic was broadly understood, and the arbitrage was to write a clean hook, ask a question, and get your team pod to react inside the first hour.
That system is mostly gone. According to LinkedIn's own engineering post published in March 2026 by Senior Staff TPM Hristo Danchev, the company rebuilt the feed around a two-stage LLM architecture. Stage one retrieves roughly 2,000 candidate posts for each user by computing embeddings and matching them on cosine similarity. Stage two is a transformer that reads up to 1,000 of your recent interactions as a chronological sequence and ranks those 2,000 candidates in order of semantic fit to the person scrolling.
That's a real shift. Your post is no longer competing on who liked it fastest. It's competing on how closely it matches the topics a specific user's past interactions have taught the system they care about.
What this looks like for a working marketer: a 47-like, 20-save post from a niche category expert stayed in feeds for more than three weeks, per Shama Hyder's analysis in MarTech. A motivational quote with 2,000 reactions disappeared inside 24 hours. Same old vanity metrics. Completely different outcomes.
The save economy is the real signal now
AuthoredUp's analysis of more than 3 million posts found that one save drives roughly 5x the reach of a like and about 2x the signal of a comment. A saved post also lifts the odds that a viewer follows you by 130%.
The logic tracks. A like is cheap. A save is someone filing your post away because they plan to act on it later. The LLM treats that as a much louder signal that the content answered a real professional need, not that it briefly entertained someone.
So the practical test for anything you publish changes. It's no longer "will this get engagement" but "would a working professional bookmark this to come back to."
From what I've seen, this is roughly the dividing line between posts that hold distribution and posts that die on the vine. If a busy person couldn't execute on something after reading it, there's nothing to save. And if there's nothing to save, the feed moves on pretty quickly.
Your profile is now part of your distribution
Here is the part most creators are still missing. 360Brew bundles your profile metadata into the semantic representation of every post you publish. Your headline, your listed skills, your experience entries. All of it gets fused into the embedding the system uses to decide who this post is relevant to.
This seems to be why generic "thought leader" headlines are quietly torching reach. The system doesn't know who to show the content to because the signal in your profile is mush. A CMO with a headline like "growth, leadership, ideas" is giving the embedding almost nothing to match on. A CMO with "B2B demand gen for early-stage SaaS, ex-HubSpot, pipeline over MQLs" gives 360Brew something to actually cluster against.
If your headline is generic, 360Brew has no idea who your content is for. Your reach is paying for that ambiguity every single post.
That is worth pausing on. You are not just writing a post anymore. You are writing a post that the feed interprets through a learned representation of you. Change the representation, change the distribution.
Why broad-appeal content is getting strangled
There is a second-order effect here that B2B teams should think hard about. The old feed rewarded scale. You wanted the widest possible hook because engagement velocity determined reach, and you needed anyone with a pulse to react.
Now the system is actively penalizing that. Per Oktopost's analysis, distribution begins inside tight relationship and topic clusters and only expands outward when early signals suggest relevance to specific audience pockets. Broad motivational content doesn't cluster well because it doesn't map to any one professional topic, so the system can't find an audience confident enough to push it to.
This is why posts that got 200+ reactions under the old algorithm now get 22. The content didn't get worse. The distribution mechanism stopped being the one it was optimized for.
It's a bit like the GEO shift happening in search. We covered how ChatGPT cites only 1.93% of the Reddit URLs it retrieves, and the takeaway was similar. The AI has the whole web available and deliberately picks a narrow subset that matches user intent. LinkedIn's feed is running the same logic on posts. Retrieval is cheap. The scarce thing is relevance to a specific reader.
A 45-minute audit that will matter more than your next ten posts
Honestly, I think most teams will respond to this by writing more posts, which is exactly the wrong instinct. The useful move is smaller and more tedious.
Here is the audit I would run this week, roughly 45 minutes for one person:
- Rewrite your headline to one concrete topic domain. Not three. Not "and leadership." One. If someone reads it and can't immediately name the specific thing you help with, it is too vague. Benchmark: a domain expert should be able to tell what audience you serve in under 5 seconds.
- Pull your last 10 posts and tag each one by whether it is saveable. Would a working professional bookmark this to reference later? If fewer than 4 of 10 are saveable, the ratio needs to flip. Checklists, frameworks, annotated screenshots, and numbered playbooks save. Opinions about industry vibes do not.
- Stop posting outside your lane. If your headline says demand gen, resist the temptation to post about founder mindset or how AI is changing everything. The system penalizes topic drift more than it used to.
- Reply to comments on your own posts in the first hour. Buffer research cited in the MarTech piece found that 83% of accounts that reply to comments outperformed those that don't. The signal the system reads there is expertise and consistency, not just activity.
Where I think this is headed
I will be direct about what I expect over the next 18 months. The median LinkedIn creator will publish roughly 60% less and focus narrower, because posting broad content into a semantic feed is just lighting effort on fire. The accounts that grow will be the ones whose profile, posts, and engagement all teach 360Brew the same story about what they do.
The uncomfortable truth is that LinkedIn just turned the entire platform into a niche authority game. If you were already doing B2B social with restraint, this is a gift. Dove, Netflix, and Nike figured out the same lesson on Reddit: less posting, sharper focus, let the platform reward the concentration. LinkedIn just formalized that into math.
If you have been writing to be widely liked, the feed has already started routing around you. It is probably worth asking why you are still posting that way.
Notice Me Senpai Editorial
