LinkedIn reach fell 47%. Here's what actually changed

The short answer

LinkedIn stopped optimising for how many people see a post and started optimising for how relevant it is to each person. A broad post aimed at everyone used to collect reach through inertia. A system that reads the text ties a broad post to nobody, because it speaks precisely to no one. Precise posts were not punished — they were spared.

The number, and where it comes from

Both studies are third-party measurements, not LinkedIn's own numbers. Treat the exact figures as directional and the direction as well established.

What did not cause it: 360Brew

Most articles explaining the drop blame 360Brew, a 150-billion-parameter model LinkedIn's research team described in a January 2025 paper. That paper was a research pre-production model, tested with a small group of members and then shut down — and it has since been withdrawn from arXiv over a licensing issue.

LinkedIn's VP of Engineering answered the question directly on 14 April 2026: is 360Brew used as part of your ranking system? "Short answer: No!" The denial travelled far less than the rumour. What ranks the feed is a different system, documented in LinkedIn's March 2026 feed update and the accompanying engineering post. We walk through it in how the LinkedIn feed actually ranks.

What did cause it

LinkedIn rebuilt both stages of the feed in March 2026. Retrieval now matches the text of a post to a text description of the member, instead of pooling separate sources. Ranking reads a member's recent history — more than 1,000 past interactions — and predicts whether that member will read a post at length, skip it, like it, comment, or share it.

The consequence is structural. Each post is matched to the members it is relevant to, not to the largest available audience. LinkedIn reported the change as a gain — +2.10% time spent and +3.52% on likes, comments and shares across 1.2 billion members — so this is not a bug being worked around. It is the intended behaviour.

Which formats fell hardest

Attaching a video used to buy distribution on its own. That gap has closed, which is worth knowing before commissioning content in a format the post does not actually need.

Why broad posts were hit hardest

A post addressed to "managers" has no member it is precisely relevant to. A post addressed to plant managers at companies under two hundred employees does. The second is easier to tie to a relevance, and it is also the only kind that produces a conversation instead of a congratulation. This is the single most practical change in the whole update: write for an audience you could name in one sentence.

What actually helps now

1. Make your profile a text the system can read. A headline naming a trade and a clientele gives retrieval something to match; a headline written for effect gives it nothing.
2. Write for one identifiable audience, not everyone. Precision is now the difference between being served and not being served.
3. Put something verifiable in the post — your own figure, a client case, an observation from practice. It is also the only thing a competitor cannot copy.
4. Answer comments within the hour. The ranker predicts comments and long reads, and a live exchange feeds both.
5. Post from people, not the page. Personal profiles still out-engage company pages by a wide margin (2.60% vs 1.74% in Metricool's 2026 study).

What this means if you only read

The same change that makes publishing harder makes reading louder. If the feed serves you what it predicts you will engage with, then the posts you actually asked for are competing with the ones designed to hold you. That is the whole problem a focused feed exists to solve: collect the people you chose, and stop depending on a ranker's guess. The related question — why a follow does not guarantee delivery — is answered in why you don't see posts from people you follow.