If you publish on LinkedIn regularly, you’ve seen the pattern. Within a minute or two of hitting post, a wave of comments arrives that restates your post back at you with a “couldn’t agree more” wrapper. Maybe a “love to connect” tucked in. None of them say anything you didn’t already write.
I’ve taken to replying: “Thanks for the recap. What’s your actual take?” Most never reply, and the ones who do can’t produce one.
That pattern is exactly what LinkedIn just moved to fix.
Last week, Laura Lorenzetti, Executive Editor at LinkedIn News, announced that the platform has built detection systems targeting AI-generated content that lacks any unique perspective, automated bulk comments, and responses that just restate the original post. The initial accuracy figure: 94%. Detected content gets reduced reach beyond the poster’s immediate network. LinkedIn’s verification filters now cover 100 million-plus members and can be applied to profile views, job applications, and feed comments.
That 94% number is the real story. It’s an engineering statement dressed up as a policy announcement. A classifier that accurate at LinkedIn’s scale means the economics of AI-generated thought leadership just collapsed. If your “personal brand” was a pipeline of generated posts and coordinated comments, the algorithm stopped distributing it this week.
For serious B2B marketers, this is good news.
The market already knew
Before LinkedIn’s announcement, the market had already rendered its verdict. It’s a continuation of a shift that started months ago with LinkedIn’s semantic feed changes.
At Oktopost, we see a huge volume of B2B social content move through our platform every day. After enough years watching this, you develop a pretty good read on which posts were written by a person and which were generated. The pattern is consistent. Posts that look generated get engagement early, when the algorithm gives them an initial push. Then conversation flatlines. Comments don’t land. The discussion that turns followers into pipeline never materializes.
Posts where a human wrote the first draft and used AI to sharpen behave the opposite way. The comments are real. People disagree. Someone tags a colleague. The conversation extends past the first 48 hours.
That “AI to sharpen” step has its own bar, though. If the model doesn’t know the writer’s tone of voice, it’ll smooth every sentence into the same generic register, and you’re back to slop with a human signature on top. The model has to learn how that specific person sounds, including the things they’d never say. It also has to hold the company’s brand voice and whatever content governance the marketing team operates under. For regulated industries like financial services or healthcare, that means knowing what compliance has approved and what it’s banned, ideally codified in an enterprise social media governance framework, before anything ships.
B2B buyers are senior, time-poor, and very good at detecting inauthenticity. They were already tuning out AI content before LinkedIn built a classifier to catch it. LinkedIn is now enforcing what readers already decided.
The “personal branding engine” problem
Over the last year, this pattern keeps showing up in customer conversations. A marketing team is proud of their executive thought leadership program. Forty posts per month per executive, all generated from a single quarterly intake document. The system picks topics, writes the posts, schedules them, and spins up comment replies. When I ask how often an executive has read a post before it ships under their name, the answer is always some variant of “they approved the intake doc.”
The intent is usually good. The team wants exec visibility on LinkedIn without pulling execs away from their actual jobs. The execution turns into a content factory running on behalf of people who weren’t in the room. Nothing only that executive could have written.
Programs like that will not survive a 94%-accurate classifier, and they never should have been built. They weren’t building anything real.
The comment pod industry is next
There’s a shadow industry that grew up alongside AI content: coordinated engagement pods and comment networks, often sold by agencies as “LinkedIn reach amplification.” Some are human pods. Many have migrated to automation, running bots or bulk-generated AI comments across large account sets to simulate organic discussion. LinkedIn has been publicly outlining measures against these pods for over a year. The new classifier and verification layers are the next phase.
LinkedIn’s verification filters close that off. If you’re filtering comments by verified members, those bot accounts don’t pass. And even if they did, a 94%-accurate content detector on the comment side catches AI-generated replies before they boost reach.
The agencies selling “one million impressions per quarter” packages built on these networks are facing an extinction event. The whole infrastructure depended on LinkedIn’s algorithm not being able to tell the difference. Now it can.
Human-in-the-loop is not a workflow preference
I want to say this plainly, because I think it gets treated as optional when it isn’t.
A human in the loop is mandatory. Not preferred. Not a nice-to-have for cautious teams. Mandatory.
Authenticity is the asset on LinkedIn. Your executive’s reputation, your brand’s credibility, the trust your buyers extend to the company voice. AI can help you produce more. It cannot produce authenticity. Only the person with actual experience of the topic, the deal, the customer call, the decision that went sideways, can produce that. A human in the loop is the only mechanism that preserves authenticity at the scale modern marketing demands.
LinkedIn’s 94% classifier is, in effect, an external enforcement layer for exactly this requirement. The platforms are now doing the policing the market wasn’t doing on its own. If your workflow doesn’t already have a human reviewing before anything ships, the algorithm just imposed one for you by killing your reach.
The right model puts the human at the start. The opinion or experience underneath the post has to come from a person, not from the prompt. AI then sharpens and personalizes at scale. Once a real human contribution is upstream, downstream automation is a workflow choice. Without that upstream contribution, automation becomes the “personal branding engine” problem above: 40 posts a quarter that an executive never thought or lived through, running under their name until the classifier catches them.
That’s exactly what we built our Employee Advocacy Agent to do. The social media manager writes the source piece. The Agent holds a deep digital model of each advocate’s tone of voice, built from how that person writes on LinkedIn, and rewrites the social media manager’s draft into something that sounds like the advocate, not like a generic share. Advocates can review each version before publish, or let the Agent post automatically in their trained voice. That’s the only model we thought would survive the feed getting smarter about real contribution.
If a human can’t defend the post in a conversation, the post shouldn’t have shipped.
What to do this week
Two things that take under an hour each.
First: look at the last 10 posts published under your executives’ names on LinkedIn. For each one, ask whether someone with that executive’s specific background and opinions could have written it. If the answer is “anyone with ChatGPT could have written it,” that post isn’t building thought leadership, it’s making noise the algorithm will soon stop distributing.
Second: look at what drives real conversation in your feed. In mine, it’s almost always a specific number we measured, a decision that went wrong and what we learned, or a contrarian take we’re willing to defend in the comments. Those things require someone to have the experience. Start there. Use AI to make the draft better. Don’t use it to skip the part where someone has a real thought.
LinkedIn’s classifier and the human-in-the-loop requirement now define what thought leadership looks like on the platform. LinkedIn just enforced what serious B2B marketers already do. Put the human contribution at the start of your workflow now, before reach collapse makes the case for you.
For more on how AI intersects with B2B LinkedIn strategy, see why AI authenticity is the new competitive gap in B2B marketing.
The post LinkedIn just made AI slop unprofitable. Good. appeared first on Oktopost.
