Table of Contents
Do not index
How is LinkedIn actually catching AI posts now?
That question started showing up in my DMs about three weeks ago. Founders who built a content engine on ChatGPT in 2024 and 2025 are watching their reach drop and most of them assume the algorithm changed again. It did not. The algorithm started reading what they were publishing.
According to The Next Web's reporting on LinkedIn's new detection rollout, "The system could correctly flag generic AI-generated content 94 percent of the time in early tests." Flagged posts do not get a notification. They do not get removed. They get held back to direct connections only, which on most accounts is a quiet death of about 80 percent of the previous reach. The post still exists. It just stops traveling.
Here is the part most people are missing. This is a probability problem, not a one-off mistake. If you post five times a week using AI as your primary drafter, that is roughly 260 posts a year. At a 94 percent flag rate on generic AI patterns, you do not have to wonder whether you will get caught. You have to wonder how many posts are already buried that you never realized got buried.
This article is for agency owners between $200k and $2M in revenue who built a LinkedIn delivery model on AI tools, and for founders running personal-brand content with a ghostwriter who quietly leans on the same prompts. If you are a founder paying $5k to $30k per month for content and you have not asked your writer to show you a draft from start to finish in the past 60 days, you are exposed.
This is not for content creators who already write everything by hand. Skip this if you treat AI as a research assistant and never let it draft. If you are still trying to find the prompt that sounds human, this article will not change your model because the model itself is the problem.
The 6 Percent Rule
Call it what I call the 6 Percent Rule. If 94 out of 100 generic AI posts get flagged, the entire game is whether you operate inside the remaining 6 percent. The 6 percent is not a prompt. It is not a tool. It is a category of writing that has a real first-person operator inside the draft, even when AI was used to start it.
What lives in the 6 percent looks like this. A specific number from a specific client engagement you actually ran. A judgment call you made that turned out wrong, and what you adjusted after. A piece of language your customer used that you wrote down. A pattern you noticed across 15 sales calls that you can name. None of that survives a rewrite this post in a more human voice prompt because the source material is not in the prompt to begin with. It lives in your head and in your notes.
What gets stuck in the 94 percent looks the opposite way. Posts that open with a rhetorical question and then deliver three lessons. Posts structured around five things I learned. Posts where the framework has no name, the numbers are round (always 10, always 5, always 80 percent), and the conclusion could apply to any industry on the platform. LinkedIn's pattern detection eats those for breakfast.
The math gets worse the more you scale. A founder posting once a week with AI assist is rolling the dice 52 times a year. An agency cranking out content for 12 clients five times a week is rolling the dice 3,120 times. At 94 percent flag rate, you cannot get lucky enough to outrun that. The volume itself becomes the signal that triggers the system.
Why agencies bolting AI onto the old workflow are about to lose retainers
The agencies that are going to take damage in the next two quarters are the ones that productized AI as a margin play. They priced retainers assuming one writer could handle four to six clients because AI was doing the heavy lifting. That math worked when reach was forgiving. It does not work when 94 percent of the output gets quietly suppressed.
The first sign a client is going to churn is not a complaint. It is a calendar invite labeled quick chat about results. By then, the founder has already pulled up their analytics, seen impressions cut in half, and started asking around about other agencies. The way you keep that meeting from being a goodbye is to know, before the founder does, which of your drafts have a real human spine and which are filler. This is the same hygiene I would apply to any content quality control system that prevents client churn before the retainer ends. If you cannot audit your own output against the 6 percent in under an hour, your retainer is already on borrowed time.
For founders watching this from the other side, the move is not to fire your writer. The move is to ask for three specific things in your next call. Show me the source material this draft came from. Show me the specific claim, story, or number that only I could have given you. Show me what you removed from the AI draft that would have made it flaggable. If your writer cannot answer those three questions in plain language, you are paying for content that LinkedIn is about to throttle, and you will see it in your reach before you see it in your pipeline.
The strategic implication is simple. The next 12 months will sort content operations into two categories. Operations built on real source material that uses AI as a tool will keep distribution. Operations built on AI with a thin layer of humanization sprayed on top will quietly disappear from the feed without ever getting a warning. Your reach will not crash on a single day. It will erode three percent at a time until the math has already happened to you. The founders and agencies who survive this shift will be the ones who treated the 94 percent number as a forecast, not a story.
