LinkedIn now carries a native report control for AI slop. Readers can flag a post as machine-written in two taps, and enough of them do that the option exists at all.
The awkward part is what triggers it. The formatting that gets posts reported is, almost item for item, the formatting every LinkedIn growth guide has recommended since 2019.
The shape that gets flagged
Open any feed and the pattern is immediate.
One sentence per line. A hook that withholds. Line breaks between every thought, so the post occupies more vertical space than its content justifies. Three bullet points, because it is always three. A question at the end inviting comments.
That shape was optimised for a feed that rewarded dwell time and comment volume. It worked. Then models learned it, because a model trained on the public internet learned what a LinkedIn post looks like, and what a LinkedIn post looks like is that.
So the format stopped signalling effort. A post written that way in 2019 read as someone who understood the platform. The same post in 2026 reads as output.
What the reader is actually reacting to
Three things, and none of them is the line breaks on their own.
Nothing is claimed. The post asserts that leadership matters, or that consistency compounds, or that most people give up too early. Every one of those is true and none of them is information. Ask what a competitor could not have posted verbatim and the answer is nothing.
The rhythm is flat. Sentences arrive at the same length, one per line, which the format disguises for about two seconds. Machine writing clusters between 14 and 18 words per sentence. Edited human writing swings between four and thirty.
The vocabulary is borrowed. Not just "delve", which everybody now watches for, but the whole register. Journeys. Unlocking potential. Fostering culture. The words carry a tone without carrying a meaning.
What a report costs you
LinkedIn does not publish the mechanics, so treat the specifics as inference rather than fact. What is observable is that reported content gets suppressed on most platforms, and that suppression on a professional network is expensive in a particular way.
Your reach on LinkedIn is not a stream of anonymous impressions. It is your actual professional network: former colleagues, current clients, people considering hiring you. A post that reads as machine-written is not neutral in front of that audience. It suggests the author outsourced their own opinions, which is a strange thing to advertise to people evaluating your judgment.
That is the real cost. Not the algorithm. The colleague who reads three of your posts and quietly downgrades their estimate of you.
Fixing it without abandoning the format
The line breaks are not the problem, so do not start there.
Put a number in the first two lines. Not a claim, a number. "We cut onboarding from three days to forty minutes" outperforms "onboarding is broken at most companies" because one of them can be wrong.
Break the rhythm deliberately. Follow a long sentence carrying an argument with a short one that lands it. Three words. Then twenty-six. Your reader will not consciously notice, and that is the point, because they consciously notice the alternative.
Cut the third bullet. If you have three, check whether the third one restates the second. Models invent a third item to complete a rhythm, and so do people copying models. Two specific items beat three where one is padding.
Name something. A client, a tool, a date, a mistake with consequences. Specificity is the one thing a model cannot fake on your behalf, because it does not know what happened to you.
Keep the question at the end if you mean it. The question is not the tell. The question following four paragraphs of nothing is the tell.
The rewrite, side by side
Here is the standard shape:
Most founders get this wrong.
They focus on output.
But real leadership isn't about doing more.
It's about creating space for others to thrive.
Here's what changed everything for me:
• Trust your team • Communicate clearly • Lead by example
What's your take? 👇
Nothing in that is false. Nothing in it is checkable either, and the "not X, it's Y" construction in the middle is one of the most reliable machine markers there is.
Now the same post with the same formatting, carrying actual content:
We cut our weekly meeting load from 12 hours to 2.
I expected output to drop. It went up.
Three features shipped two weeks early that quarter, and customer-reported bugs fell 30%.
The thing I got wrong: I thought meetings were how I stayed informed. They were how I stayed comfortable.
Two changes did most of it:
• Standups moved to a written thread • Anything under four people became async by default
Still not sure the second one scales past 40 people. Anyone run this at a larger size?
Same line breaks. Same bullet habit. Same closing question. The difference is that every line commits to something, the sentence lengths vary, and the closing question is one the author actually wants answered.
Checking a draft before you post
Paste the post into the free workbench before you publish it. It flags the borrowed vocabulary in context, measures your sentence-length variance against the 0.60 human threshold, then returns a score out of 100.
Most people find their vocabulary is fine and their rhythm is not, which is the same result I got running my own articles through it.
The deeper fix is upstream. If a model drafts your posts, give it a ban-list rather than a request to sound human, cap the em dashes, tell it to vary sentence length deliberately, and require every post to carry one claim a reader could dispute. Models follow constraints they can count. They ignore instructions about tone, because tone is not something a decoder can check while it generates.