7 September 2026 · 5 min read

The most disciplined people on LinkedIn are the ones losing the most ground, and the reason is not effort.

Richard van der Blom, who runs the Algorithm Insights study, went through 1.3 million posts from 50,000 creators. Speaking about that data set on 2 June 2026, he put the drop in reach for active creators at 60 percent over two years. Engagement fell by only 20 to 25 percent. The two numbers moving apart is the whole story.

Then the detail that should stop anyone who posts on a schedule. People who reduced how often they posted saw reach go up 10 to 15 percent.

The people who did the most work lost the most, and some of the people who did less gained.

Treat that as one analyst’s study, not a law. But it is the largest recent sample published, and it points somewhere specific.

What is doing the reading now

LinkedIn does not narrate its ranking changes, but its engineers publish. In a paper dated 16 October 2025, a LinkedIn team described the retrieval layer that sits under the feed. They fine tuned a large causal language model as a dual encoder to build embeddings for both members and content, and the phrase in the abstract is the one worth keeping: using only textual input.

That system narrows hundreds of millions of candidates to 2,000 in a few milliseconds, before anything you would recognize as ranking begins. The gains, they report, were largest for newer members, the people with almost no engagement history to score.

A second paper from January 2025 described a 150 billion parameter model handling more than 30 predictive tasks through a text interface, replacing systems built feature by feature over years. It was later withdrawn from the preprint server over a licensing agreement, so cite it carefully. The direction survives the withdrawal.

Both describe the same move. The first pass over your post is a reading, not a count.

Why volume stopped compounding

For a decade the model was mechanical. Post often, collect early engagement, get shown to more people, repeat. Frequency worked because the system counted.

Two things broke that.

The first is that the early signal is no longer trustworthy. Van der Blom said roughly 80 percent of the comments arriving in the first five minutes on his own posts are written by AI. A signal that can be manufactured at zero cost stops being evidence, and any system leaning on it gets corrected.

The second is that the feed no longer needs your history to place you. If retrieval works from text, a post is matched to readers by what it says. Your back catalog counts as subject, not as volume.

Nine percent of the feed is now suggested posts, in the same June 2026 analysis. That is space given on relevance, not on who follows whom.

So posting five times a week across five subjects no longer buys five times the placement. It buys a blurred embedding and four posts competing with the one that mattered. I have written before that your job is not to become good at LinkedIn. This is the mechanical version of that argument.

One subject, four angles

Here is the correction. It runs on a four week cycle and it usually means writing less.

1. Name the one subject you want to be retrievable for

Not your industry. Not your title. The specific problem you get called about. Write it as a phrase a client would use, because that phrasing is what the text encoder has to match. “Board communication for founders after a down round” is a subject. “Leadership” is not.

2. Cut cadence before you cut anything else

Van der Blom’s own recommendation moved from five or six posts a week to two to four. Take the lower end for a month. The point is not rest. The time you free has to go into the next three steps, or the cut does nothing.

3. Write four angles on that subject, not four subjects

  • Week one, the mistake you keep seeing.
  • Week two, the case that changed your mind.
  • Week three, what you refuse to do and why.
  • Week four, the number nobody tracks.

To a human that reads as depth. To a text encoder it reads as one position stated four times.

4. Keep the vocabulary stable

If you call it a positioning problem in one post and a messaging problem in the next, you have described two things. Choose the words a buyer would use and repeat them without apology. This is the least glamorous step and it does the most work.

5. Judge it by who arrives, not by what moved

Reach and engagement now measure different things, and one of them is faked at scale. Track the inbound instead. Who wrote to you, what did they think you did, were they the kind of person you want. Thirty right readers beat 30,000 wrong ones, and only one of those numbers is on the dashboard.

What this costs

The first month is uncomfortable. Narrowing to one subject feels like losing range, and the numbers dip before they settle.

Most people abandon it there, add a second subject, then a third, and end up with a wide catalog and a thin position.

A blurred position is not a small version of a clear one. It is a different outcome.

Almost every engagement starts by taking things out.

The work also compounds in a way volume never did. A subject stated clearly four times outlasts forty scattered posts, and it survives being compressed into a summary by a machine or a colleague, which is where most decisions about you are now made.

Questions people ask about this

Does posting less actually increase reach?

That is what one large data set suggests, not a rule the platform has published. The finding was a 10 to 15 percent lift for people who reduced frequency, in van der Blom’s June 2026 analysis. The mechanism is probably not the cut itself but what comes with it, fewer subjects and better pieces.

Should I delete old posts that are off subject?

No. Let them sit and make the next ten consistent. Recency does the correcting for you.

My comments are full of automated replies. Does that hurt me?

It does not appear to help. If early engagement is discounted because it can be manufactured, comments farmed from pods are effort spent on a devalued signal. Reply properly to the few who are real.

What about video, documents and newsletters?

Format is a smaller lever than subject. Choose the one that lets you make the argument fully, and keep the subject fixed across it.

How long before this shows?

Four weeks to see the shape, about three months before the inbound changes.

Who is this for?

Senior people whose reputation runs ahead of their public record, and who would rather be findable for one thing than visible for everything. This is what LinkedElla does, quietly and under agreement.