Something Weird Is Going On With the LinkedIn Algorithm

A few weeks ago, one of my LinkedIn posts sat at 1,000 impressions and looked dead. Then it jumped to 40,000 in three hours. I've never seen a post do that.

That was actually the second strange thing. There were four.

My posts take much longer to accumulate impressions now. For years the pattern was reliable: most of a post's reach arrived in the first 24 to 48 hours, then it died. This summer, posts kept collecting impressions for five days, sometimes a week. And honestly, I don't love it. I like responding to comments while I still care about the post. When someone comments on something I wrote six days ago, it's old news to me. I'm over it. I've mentally moved on to the next thing.

The topic matching got absurdly specific. I comment and repost about LinkedIn findability, which is a topic I almost never see anyone else cover the way I do. I posted about it one day, and my feed served me five other posts on the exact same subject. Not "LinkedIn tips." Findability. The precision was strange enough that I noticed it immediately.

And my best-performing post of the month was me announcing I hit 7,000 followers. Which, let's be real, is not that big a deal. Nobody in my actual life would congratulate me for it. There are more exciting things happening in the world. But that throwaway post beat weeks of my real work.

I haven't seen anyone talking about this. So I went digging into what LinkedIn has actually published about its feed, plus the independent research, to figure out what's going on.

Here's what I found. Some of it I could confirm with primary sources. Some of it I couldn't, and I'll tell you which is which, because the LinkedIn commentary industry is terrible at that distinction.

The short version: almost every piece of LinkedIn advice you'll read this year was written for a system that no longer exists.

What I could confirm

There was no summer algorithm update. I went in assuming LinkedIn shipped something in July. Wrong. LinkedIn's engineering blog still lists March 12, 2026 as its most recent major feed announcement. There are July creator posts claiming another update. None of them cite a primary source, because there isn't one.

What actually happened is a rebuild that took over a year, shipped in stages. The timeline matters because it explains why the weirdness crept up instead of arriving on a date:

In June 2025, LinkedIn deliberately loosened recency. People started seeing posts two and three weeks old, and LinkedIn's VP of Product confirmed to Business Insider it was intentional. They wanted a better balance between recent and relevant. So the long-tail thing I noticed this summer was a direction LinkedIn chose over a year ago.

In October 2025, LinkedIn researchers published a retrieval paper describing a system that pulls roughly 2,000 candidate posts from a pool of hundreds of millions, matched to a member's topical interests, specifically to surface content from people they don't follow.

By early 2026, a new ranking system had replaced the old one for most feed traffic. It reads over 1,000 of your past interactions as a sequence, modeling how your interests evolve over time instead of scoring each post in isolation.

Then in March 2026, LinkedIn announced the full next-generation feed: AI-generated representations of every post and every member, matched on meaning. Not keywords. Meaning. Their own example is a member who reads about small modular reactors getting matched to relevant content that shares zero vocabulary with anything they've engaged with before.

Posts can now get "refreshed" and re-enter circulation. This is the finding that explains my 1,000-to-40,000 post, and it's the single most underreported thing LinkedIn has disclosed. When an existing post gains engagement, LinkedIn says its representation gets dynamically updated and re-enters the retrieval index within minutes. Popularity, engagement rate, recency and affinity are all baked into that representation. A post isn't a one-time event anymore. It's an object sitting in a giant content bank, and its eligibility gets recalculated as signals come in.

My findability observation is the system working exactly as designed. LinkedIn builds your interest profile from your engagement history, and their engineers found that positive engagement (what you comment on, react to, dwell on, repost) works better as a signal than everything you scroll past. My comments and reposts about findability told the system I care about findability. It went and retrieved findability content from across the entire platform, including from people I've never heard of. The specificity that weirded me out is the product feature. LinkedIn even gives an example of inferring combined interests: someone who reads hospital operations content and engages with cybersecurity gets recommended posts about protecting patient data. It's not matching topics. It's matching you.

Creators finally got a way to see this happening. On June 3, 2026, LinkedIn rolled out post-level analytics showing what percentage of your impressions came from in-network versus out-of-network. One LinkedIn employee noted her out-of-network number was often the bigger one. More on why this matters below.

The independent data half-matches my experience, and the half that doesn't is instructive. Richard van der Blom's 2026 research, based on 1.3 million posts, reports the distribution curve changed shape at both ends. His numbers: about 40% of eventual reach now lands in the first hour, up from 20% in 2024, while the 48-hour mark dropped from 85% to 65%, with the last chunk arriving six to eight days out.

The tail matches what I'm seeing exactly. The front end does not. I'm getting maybe 5 to 10% of my reach in the first hour, nowhere near 40. So either my account is an outlier, or a platform-wide average is hiding enormous variation between accounts. Probably the second, and that's the useful lesson about every third-party algorithm statistic: these are proprietary estimates averaged across accounts that behave nothing alike, with methodology nobody can audit. LinkedIn has never published a decay curve. Treat all of it as directional. The direction we agree on is the fat tail, and the fat tail is the part that changes how you should behave.

One myth to kill while I'm here: it's not 360Brew. A lot of 2026 algorithm guides confidently name 360Brew as LinkedIn's new engine. LinkedIn's VP of Engineering said in April it was a limited experiment that got shut down. Any guide citing it is recycling a claim LinkedIn already denied.

What I couldn't confirm

This section matters more than the last one, because it's the stuff every algorithm guide asserts as fact anyway.

Why now. I can document every piece of the machinery and roughly when it shipped. I cannot document anything that happened in July or August specifically. My best guess, and it's a guess, is that "rolling out" in March doesn't mean fully rolled out in March, the system's model of each of us gets richer every week, and at some point the accumulated effect crossed a threshold where regular posters started feeling it. That's inference, not fact.

A "second wave test." Plenty of creator guides describe a formal protocol where LinkedIn retests posts against new audiences after 24 hours. I found zero documentation of any such rule. What I did find is machinery that produces the same result without needing a rule: post gets modest engagement, engagement refreshes its representation, refreshed post becomes competitive for retrieval against a whole interest cluster, cluster is way bigger than your follower list, boom. Every step of that loop is documented. The loop as an explanation for my specific post is my inference.

A milestone boost. I really wanted this one to be true because it would explain my 7,000-follower post. No evidence. Nothing in any primary source suggests milestone posts get special treatment. The boring explanation is probably the right one: a congratulations post is something thousands of weak ties can respond to instantly, without needing to know anything about what I actually do. A post about member engagement strategy is useful to a much smaller group. Easy engagement arrives fast, fast engagement feeds the popularity signals, and away it goes. Also, it's one post. One post proves nothing about a category.

Whether posting about a topic programs your own feed. The documented signals are your profile and your engagement behavior. My commenting and reposting about findability? Documented signal. The act of authoring a post about it? Not documented anywhere I could find. So the feed probably learned from my engagement around the topic, not from the fact that I published on it.

The actual math. LinkedIn has never published a decay curve, distribution weights, or the thresholds that decide when a post crosses from modest reach to massive out-of-network reach. Anyone quoting exact percentages is quoting third-party estimates, and third-party estimates get copied from creator to creator until they start sounding like official facts. They're not.

What this means for you

If you use LinkedIn to market yourself or your business, here's what I'd actually change.

Stop judging posts at 48 hours. This is the big one. The 24-to-48-hour report card is dead. Give every post seven days before you call it. And don't delete or repost a "failure" on day two, because under this architecture, day two means nothing. My 40,000-impression post was a dud at that point.

Check your in-network versus out-of-network split on every post. This new metric is the most useful thing LinkedIn shipped all year, and most people are ignoring it. It tells you whether a post won because your existing audience showed up, or because it escaped your audience entirely and got recommended to strangers. Those are two completely different kinds of success, and they call for different content decisions. If you want to test whether the retrieval system is driving your late spikes, this is how: late-blooming posts should skew heavily out-of-network. Mine did.

Write posts that are clearly about one thing. The system builds a representation of what your post means and matches it to people who care about that subject. A post that's crisply about one identifiable topic is easier to match than a post gesturing at four things. This is not keyword advice. The whole point of semantic matching is that it works without shared vocabulary. Just write clearly about a real subject.

Specific now beats broad. This is the strategic shift underneath everything. If a post can be retrieved for anyone on the platform whose behavior matches its subject, then a post that strongly matters to 5,000 strangers can outperform a post that mildly interests your 7,000 followers. My findability content is niche. Under the old system, niche capped your reach. Under this one, niche is targetable. That inverts a decade of "broaden your appeal" advice.

Your follower count matters less than it did. Followers are still your seed audience and still valuable. But they're no longer the boundary of your reach, and building follower count as your primary strategy is optimizing for the old system. Van der Blom's data reports a sharp decoupling between followers and reach. That's his estimate, not LinkedIn's, but the architecture supports the direction.

Your engagement is training your feed. Every comment, repost and long read is telling the system what to bring you more of. That's why my feed filled with findability posts. Useful to know for what you consume, and worth knowing your audience's feeds work the same way: the people engaging with your topic are getting fed more of that topic, including yours.

And the annoying part: plan for slow conversations. If you're like me and you want to respond to every comment, that window doesn't close anymore. Comments will trickle in on day five, day six, on posts you're completely over. I don't have a fix for this. I just know the days of a post being "done" are gone, and I've stopped expecting closure.

Here's where I landed. I went looking for an algorithm update and found something bigger and slower: LinkedIn changed what a post is, and never sent a memo. A post used to be an event. Your followers saw it, it lived for two days, it died. Now it's an object in a searchable bank that can get pulled back out whenever it matches someone's evolving interests, whether they follow you or not.

Do I like it? Mixed. The distribution upside for specific, useful content is real, and I write specific, useful content, so I'll take it. But the rhythm I built my posting habits around is gone, and nobody announced it. The machinery shipped over eighteen months of engineering blog posts that almost no one who posts daily will ever read.

Which is exactly why it felt so weird. The system changed underneath us, one stage at a time, and the only announcement most of us got was our own analytics acting strange.

Everything above is either linked to a primary source, attributed to named independent research, or labeled as my inference. Where I couldn't confirm something, I said so.