Do not index
"Does it matter if people think my posts sound like AI?" That is the question founders and agency owners keep bringing me, usually with more worry in the phrasing. The answer is yes, and it now matters more than any ranking factor you are currently optimizing for. Over a million people clicked LinkedIn's "seems like AI slop" button within a few weeks of launch, according to TechRadar Pro, and views of flagged content dropped roughly 40 percent. Detection stopped being an algorithm problem the moment it became a crowd behavior.
The stranger part is what happened next. People who clicked the button report that their feeds look exactly the same. LinkedIn's own explanation covers why. "Importantly, no single piece of feedback determines how content is distributed. We look at many signals together, and we've built safeguards to help prevent individual feedback from being used to unfairly target other members," Srinivasan told TechRadar Pro, adding that the platform catches hundreds of thousands of automated comment attempts every day.
Read that carefully, because it is the whole story. The platform is deliberately diluting the signal, which means the feed will not clean itself on your behalf. The million people who tapped that button did not change what they see. They changed what they do while reading. A button that lets you publicly accuse a post of being machine output trains an entire audience to run a suspicion check on every paragraph. That habit does not switch off just because the flag failed to move distribution.
The standard moved from ranking to believing
The old question was whether a post would rank. The new question is whether the person reading it believes a human wrote it. Those are different jobs and they reward opposite instincts. Ranking rewards consistency, tested hooks, and formatting that survives a three line preview. Believability rewards specificity, admission, and detail that could not have come from anywhere except the room you were actually in.
This applies to founders running personal brand content, ghostwriters charging $5k to $30k per month, and agency owners between $200k and $2M in revenue who publish under a client's name. Those three groups carry the same exposure. You are producing volume under a real person's face, and your reader now has a one tap way to call it fake in public, in front of that person's buyers. A single flag does nothing. A pattern of readers who quietly stop believing you does everything.
Skip this if you publish twice a quarter and treat LinkedIn as a digital resume. The suspicion tax lands hardest on people posting at volume, because volume is what pattern recognition feeds on. And if you are still grading the program on impressions alone, this will not change your model. Start instead with how you decide whether LinkedIn is actually working, because a believability problem never shows up in a dashboard until it has already cost you the deal.
The Human Signature Test
Here is what I would actually do before anything goes out. Run what I call the Human Signature Test. Take any three consecutive sentences in the draft and ask whether a competent stranger working from the same brief could have written them. If the answer is yes, that passage reads as machine output whether or not a machine touched it. If the answer is no, something in there is load bearing: a number only you know, a decision you regret, a client objection phrased the way the client actually phrased it, a date, a room, a cost.
Most posts pass at the hook and fail in the middle. The opening is sharp because everyone edits openings. The closing line is sharp because everyone edits closings. Then there are four paragraphs of connective tissue that any tool on earth could generate, and that is the stretch a reader is scanning when they decide what you are. Fix the middle. One concrete number or one named consequence per two hundred words is enough to carry a post, and it is far more efficient than rewriting the hook a fifth time.
None of this is an argument against using AI. Snapchat drew the same line a week later by making only videos from real people eligible for Spotlight recommendations while leaving its own AI editing tools untouched. Two platforms, two enforcement bets, one shared distinction: assisted is fine, generated is not. That distinction is not really about tooling. It is about whether a human made the judgment calls. AI can handle mechanics all day. What it cannot do is decide which of your seven client stories is the one worth telling this week, and that decision is the part readers are actually detecting.
The trajectory here is worth sitting with. Reach used to be won by people who understood distribution mechanics better than their competitors. That advantage is closing, because the mechanics are now available to everyone at zero marginal cost and the audience has been handed a tool to punish the output. What is left is proprietary experience, and the willingness to put the specific, slightly uncomfortable version of it on the page. Over the next year the accounts that keep compounding will be the ones whose posts could only have been written by one person. Everyone else will be optimizing for a feed that has quietly stopped believing them.
