12 posts per week. That was my LinkedIn cadence, and it was quietly killing my reach.
Not because the content was bad. Because two posts per weekday meant both landed in the same golden hour window and competed for the same network. The feed ranker almost never shows both posts to the same person. So the second post was diluting the first. Every week I was running two campaigns against each other.
I found this by looking at what I actually controlled at the algorithm level. LinkedIn's feed ranker has documented behavior around post spacing and early engagement velocity. When you put two posts in the same window, neither gets full distribution in the first hour. You split your own early signal. And early signal is what determines whether the algorithm amplifies you at all.
The fix was straightforward once I saw it: drop to 1 post per day. Keep the higher reach late morning slot on weekdays. Keep the two carousel posts on Tuesday and Wednesday. That is 7 posts per week instead of 12. Fewer posts, more room for each one to breathe.
The second problem was subtler
Every topic signal I was pulling from was AI and dev news. Which meant 100 percent of my posts were Claude, agentic systems, or some adjacent corner of that world. My audience is college students, builders, and founders. They care about AI, yes. But seeing the same note repeated is how you get unfollowed.
I needed variety without faking expertise I do not have. The answer was a reflection topic type: lessons, ideas, and thoughts that come from actually building things, not from chasing the AI news cycle.
I seeded it with 36 topic prompts covering angles outside AI in voice. Things like shipping decisions, what breaks in solo projects, why planning fails, when to cut scope. Real stuff from building. The mix is now roughly 1 in 3 posts from this reflection pool, with AI content staying at about 2 in 3. Not an equal split. AI is still the core. But the feed no longer looks like one note on repeat.
The implementation detail that matters: the pool rotates oldest first. A seed does not reappear until the rest have run. Otherwise you end up with de facto repeats just from different angles, which defeats the point.
Verification before shipping
After the changes, the plan command showed 1 post per day. The curate command appended all 36 reflection seeds to the queue. The selector was realizing about 28 percent reflection in practice, which matched the configured mix. A reflection draft ran through the quality gate and passed. 189 tests green.
What I would do differently
I should have caught the cadence problem earlier. The data was there. Two posts per day, declining per post impressions, obvious explanation. I kept optimizing the content and ignoring the structural issue. That is the classic mistake: iterate on the thing you are most comfortable touching instead of looking at what the data is actually telling you.
On the reflection stream, I would have started with a smaller seed pool and validated one full rotation before expanding. 36 seeds is probably more than I needed to start. Better to prove the rotation mechanism works at 10 seeds and grow from there.
The broader lesson: most content distribution problems are not content problems. They are structural. Fix the structure first.
Top comments (0)