When I built a playable model of the LinkedIn feed, I set one rule. Every mechanic had to carry a stamp and a source a player could open: Proven if LinkedIn published it, Measured if someone outside LinkedIn counted and said how, Disputed if the counts disagree, Invented if it traces to nothing. Building the model was the easy part. Stamping it was the work, and the stamps took apart a rule I had carried for years.
I expected posting frequency to be the most reliable lever there is. Every other platform taught me that: post more, post often, feed the machine. I couldn't find a source that says LinkedIn rewards volume. What LinkedIn has published points somewhere else. Its retrieval matches posts on meaning, and your headline, company and industry are part of the text that describes every post you write. A steady drumbeat on one subject you can own lines up with that. Ten topics a week scatter it.
The short list
Here is everything I could stamp Proven. LinkedIn filters engagement bait, and quotes "Comment 'Yes' if you agree" as its own example. It is making pods and automated comments ineffective. It reads your headline into every post. Time spent reading counts. That is the list. It is short and a little boring: don't bait, don't pod, write about what your headline says you do, and write something worth finishing.
I misread the last one at first. I decided that spending more of my own time on LinkedIn, reading for research, would lift my posts. Nothing says that. Dwell time is the reader's time on your post, not yours on the site. I had invented a rule while stamping everyone else's. The research habit is still good for the writing; it just isn't a ranking signal.
The long list
Most of the model is Measured, and every Measured number comes from someone who sells something. AuthoredUp counted 3 million posts and sells a posting tool. Documents reach 1.39x a typical post, images 1.20x with the best engagement, and video reach fell 36% in a year. Richard van der Blom sells a paid report; he puts reach for active creators down about 60% over two years. Interact sells quizzes, and reports that 40.1% of quiz starts become leads. You are playing one.
The numbers are real, and they are pointed. Polls reach 1.78x a typical post and draw 0.37x the engagement, both figures from the same report. Which half ends up in a carousel depends on who is selling.
The reach number surprised me most. LinkedIn says it chose relevance over reach, which is defensible for a feed and hard on anyone who hasn't been discovered yet. The next voice worth hearing now starts with less room than the last one did. I haven't repeated a vendor number as fact, as far as I know. I have repeated stale advice. "Post more" was burned into me on other platforms, and I passed it along on this one.
Disputed, and invented
Only one topic landed in Disputed: links, which is also where the advice is most confident. Van der Blom measures 16% less reach for a post with an external link. Ordinal, which sells a first-comment tool, says personal profiles see almost none. Nobody has published how a link in the first comment behaves. I put mine there anyway, like almost everyone. It may be another "post more," a habit we all carry because someone once said it with conviction.
Three cards in the model are Invented outright: +21% for commenting on three posts before you publish, a 60% link penalty, and ending on "Thoughts?" None trace to data. Neither does the title. The 80% comes from van der Blom estimating, 23 minutes into a podcast, how many early comments on his own posts were AI-written. One researcher, one account, one guess. It circulates as a statistic.
I kept it anyway. The number is invented, but the feeling it names is real. More of what fills a comment section now is copy written by a program to pass as a person, and that is where the internet is heading. The model's last screen stamps its own title Invented, which seemed like the only honest way to use it.
The turn
My own 107 posts sit in Measured with every caveat I put on AuthoredUp: one account, engagement rather than reach, one person doing the coding. I'm not trying to get you to buy something, but I am selling something: a podcast, this model, this essay. My data says promotional posts draw 0.75x my median. You are reading one.
I wanted my data to show I had stumbled into a strategy without trying. It showed trial and error, which only gets you so far. A strategy needs to be run like an experiment, with a control, or you learn nothing from the result. The same data says a specific closing question pulls a median of 9 comments against 2. I use it, though not consistently, which is its own finding.
The model runs on 102 weights, and 65 are mine. The structure comes from LinkedIn's published engineering. The weights are ours, and anyone who tells you they have the weights is selling something.
Four rules came from LinkedIn. Everything else came from someone with something to sell, and that includes me.
When you meet a LinkedIn rule, ask who counted and what they sell. The Proven list is enough to post well. Past it, you are choosing whose pitch to believe, or developing taste. I'm narrowing to product strategy for media companies and the creator economy, and staying on them for twelve weeks, which happens to be the length of one run of the model. You can play it or check every stamp.