The numbers everyone is quoting about LinkedIn's 2026 change mostly do not hold up.

LinkedIn's ranking system changed in 2026. A unified AI system reported as 360Brew replaced the older engagement-history model, and a March 2026 update is widely described as an authenticity crackdown on engagement bait, automation pods and link spam. LinkedIn has not published the mechanics, so everything specific about how it works is marketing commentary rather than documentation, and worth reading as such.

What has been published is the performance data. It does not say what the commentary says it says.

What the largest dataset actually found

Socialinsider analysed 1.3 million LinkedIn posts from 16,645 business pages. LinkedIn's average engagement rate by impressions sits at 5.20 per cent, an 8 per cent rise year on year. Engagement grew across every format measured: video up 7 per cent, images up 9 per cent, text posts up 12 per cent. Native documents lead outright at 7.00 per cent, up 14 per cent.

Set that against the figures in circulation. The most-repeated set, from Richard van der Blom's Algorithm Insights report, has views down 50 per cent, engagement down 25 per cent and follower growth down 59 per cent. Those numbers are from a 2025 report, so they cannot describe a March 2026 update at all, and the engagement figure points the opposite way to the larger dataset.

Two claims do survive. Video views fell 36 per cent year on year across every page size, which is a real and steep decline. And follower growth genuinely slowed, sharply at the top: pages between 100,000 and one million followers grew 6.4 per cent, down from 21.6 per cent.

Why the wrong numbers spread anyway

The collapse story is more useful to the people telling it. A platform in freefall justifies a service, and it explains away a quiet quarter without anyone having to look at what they posted.

There is a real thing underneath it, and it is smaller and less quotable. Video views are down while video engagement is up, which means fewer people are seeing each video and more of the ones who do are responding. Follower counts are climbing more slowly. Neither of those is reach collapsing. Both are the same audience being harder to add to and more selective once you have it.

What the format data says to do

Native documents are the strongest format on the platform and have been for some time, at 7.00 per cent engagement against video's 6.00 per cent and text's 4.50 per cent. In the most recent quarter they slipped slightly, to 6.60 per cent from 6.90 per cent, which is movement inside a range rather than a trend.

Video is worth making and worth judging on the right metric. Anyone measuring LinkedIn video on views will conclude it stopped working. Anyone measuring it on engagement will conclude it is performing better than last year. Same videos.

Polls are the one to be careful with, in both directions. They are the weakest engagement format at 4.50 per cent, and above 50,000 followers they generate more impressions than anything else on the platform. Which of those matters depends on what the post is for.

One caveat belongs with all of it. The underlying study covers January 2024 to December 2025, with quarterly updates through 2026, so most of what it measures predates the March 2026 update. It is the best evidence available on LinkedIn performance and it is not evidence about the update itself. Nobody currently has that.

What we think this means

The direction everyone agrees on is right. LinkedIn is rewarding posts that took thought and filtering out posts that took none, and that is worth building around.

The numbers being used to prove it are mostly wrong, and they are wrong in a way that should make anyone reading them uncomfortable. Three of the four figures in wide circulation either reverse their own source or come from a report published before the thing they claim to measure. We know because we had two of them in our own draft of this piece and pulled it.

That is the actual lesson, and it is bigger than LinkedIn. A business planning its content on numbers it has not traced back to a dataset is planning on nothing, and the marketing industry is currently generating those numbers faster than anyone can check them. The check is not hard. Find the study, read the methodology, note the date, see whether the figure means what the person quoting it says it means. It took us under an hour on this piece and it changed the whole argument. The same discipline is why the Australian ad spend piece quotes the report and not the commentary about it.

For a company page or a founder's profile, the practical version is short. Make documents. Make video and stop judging it on views. Post as someone specific about something you actually know. And when a post tells you the platform is collapsing, ask it where the number came from.