LinkedIn Is Done Pretending AI Posts Are Harmless

LinkedIn has added a new option to the menu on every post. Tap the three dots at the top right of a post in your feed and you will find a choice labeled "Seems like AI slop." Select it and LinkedIn thanks you for the feedback, notes that it helps improve the feed, and hides the post. Users started spotting it in the mobile app this week, and desktop is expected to follow.

The button sits on top of work the company announced in May, when it said it would reduce distribution for content that appears to be generated by AI and lacks a clear perspective. That system relies on machine learning trained to recognize writing that reads as generic or repetitive even when it looks polished. LinkedIn put detection accuracy at 94 percent in early testing. It has not released a false positive rate, which leaves open the question of how often a human writer gets swept up by it.

The scale of the problem explains the urgency. 404 Media reported earlier this month that the AI detection service Pangram estimated 41 percent of long-form LinkedIn posts and 30 percent of short-form posts were likely machine written. Anyone who has scrolled the feed recently will find that number plausible.

What the penalty actually is

Flagged posts do not get deleted. Accounts do not get suspended. What happens instead is quieter and, for most people, worse: the post stops traveling. It stays visible to your existing connections, but it becomes far less likely to surface in front of anyone outside that circle. From the poster's side, nothing looks wrong. The post is still there. The reach simply is not.

Hari Srinivasan, LinkedIn's VP of Product, framed the new button as a training input rather than a moderation tool. Slop resists a fixed definition and the definition keeps shifting, so member reports give the company a way to tune its models against what people actually find worthless.

He also drew a line that deserves attention. AI and slop are not the same category, and plenty of people use these tools to sharpen a thought they already had. LinkedIn appears to be aiming at the absence of a perspective rather than the presence of a tool. To that end, the company is testing a private nudge that will appear in a poster's own analytics when their writing may have come across as inauthentic, and it is replacing the old "enhance your post" feature with a proofreader built to leave the writer's voice intact. The AI writing assistant that lived in the compose box appears to have been pulled.

There is an obvious tension in all of this. LinkedIn is owned by Microsoft, which has invested more heavily in OpenAI than any other company, and LinkedIn has spent two years shipping AI writing features to its own subscribers. Now it is building systems to suppress the output of exactly that kind of tooling. The company would probably say those positions are compatible, and on the specifics it has a case. The distinction it is drawing is between AI as a drafting aid and AI as a substitute for having anything to say.

What it means if you post to build a brand

The quiet assumption behind AI-written posts has always been that nobody can tell. Readers can tell. The tells are well documented by now: sentences broken into their own paragraphs for false emphasis, tidy numbered takeaways, a certain relentless symmetry in the phrasing, an opening that restates the premise before saying anything. The 404 Media reporter who tested the new button found a candidate after roughly six seconds of scrolling.

So the reputational cost was already being paid. What changed this week is that a distribution cost now sits on top of it. Reduced reach and a reader who has mentally filed you under "skim, do not read closely" is a poor return on saved time.

The deeper issue is what gets delegated. Handing AI your calendar, your meeting notes, your first-pass research, or the job of turning a deck into a one-pager costs you nothing. Handing it your published opinion costs you the only thing that makes a personal brand worth having. A generated post does not produce your thinking in a more efficient form. It produces the statistical average of everything written on the topic, attached to your name and your photo. Sounding like everyone else is the specific outcome you were trying to avoid.

The useful version of this technology sits earlier in the process. Use it to pressure test an argument, to find the weak spot in a draft you wrote, to catch the paragraph that is doing no work. Then write the thing yourself.

The incentive finally moved

For most of the past few years, the cheapest content to produce and the most widely distributed content were the same content. That is now less true than it was a week ago, and the gap will probably keep widening as LinkedIn feeds member reports back into its models.

Detection will get things wrong. The false positive question is real, and people whose natural writing runs formal or list-heavy have a legitimate worry. But the direction is right. A platform that makes synthetic volume less profitable makes original thinking more visible by default, and that is a change worth having.

If your posting strategy depended on the old math, consider this notice. If it did not, the competition just got a lot quieter.