If more posts mean more reach, why did my LinkedIn numbers collapse?

That was the question sitting in front of me two weeks ago. I had been running a consistent 12 posts per week, two per weekday, and the numbers were moving in the wrong direction: impressions down, engagement down, the whole curve bent wrong.

Two problems showed up in the diagnosis. Both of them were things I was doing to myself.

Cadence cannibalization

LinkedIn’s feed ranker will not surface two posts from the same account to the same person in a short window. When I was posting twice a day, both posts were landing in the same golden hour for my network. The ranker picked one, the other got buried. I was not getting twice the reach. I was splitting one unit of reach between two posts and diluting both.

The fix was obvious in retrospect: cut to one post per day. Seven per week instead of twelve. I kept the late morning and midday slots that had the strongest reach data, kept the Tuesday and Wednesday carousels. The second daily post just went away.

Less content, more visibility. It feels wrong until you look at how algorithmic feeds actually work.

Topic monotony

Every signal source I was pulling from was AI and dev news. The result: 100 percent of my posts were Claude and agentic systems content. My audience is college students, founders, builders. They saw the same note repeated every day and tuned it out.

I needed variety in the topic pool without abandoning the core. So I added a reflection stream: 36 evergreen seeds covering lessons, ideas, and thoughts that have nothing to do with the latest model release. The mix went to roughly one in three posts being a reflection piece, with AI and dev staying the two thirds majority.

The implementation detail that matters: the seeds rotate oldest first. Every seed has to run before any seed repeats. This prevents the rotation from becoming its own kind of monotony, where the same “lessons from shipping” angle shows up every third post because it always scores highest. Recency scoring forces genuine variety.

The prompt for reflection posts also changes the implied audience. Instead of writing at AI builders specifically, the framing opens to anyone who has shipped something and learned something. Broader surface, same voice.

What I would do differently

I should have caught the cadence problem earlier. The data was there: second posts consistently underperformed first posts by a wide margin. I kept posting twice a day because the input felt productive. Twelve posts per week sounds like more work than seven, and more work should mean more output. That instinct is wrong when the platform is working against you.

On topic mix: I waited too long to add variety. The reflection seeds I added are not new ideas. They are things worth saying that never got said because the queue was always full of AI news. A topic pool that only tracks live feeds will drift toward whatever is loudest that week. You need evergreen material to anchor the voice across cycles or you end up accidentally running a news aggregator with your face on it.

The full run now passes 189 tests. Plan shows one slot per day. Selector is realizing around 28 percent reflection. A reflection draft cleared the quality gate end to end.

We will see what the next two weeks of data looks like.