LinkedIn says more than a million people have clicked its "seems like AI slop" button since it launched on 30 July. The figure came from chief product officer Hari Srinivasan in a Thursday update, as The Verge reported, along with a second number: members are now seeing 40 percent fewer views on posts LinkedIn classifies as AI slop, compared with a few weeks ago.
The shovel was in-house
There's a detail at the centre of this success story that the announcement doesn't mention. LinkedIn, like most platforms, has spent years shipping AI writing tools to its users. The most on-the-nose was a prominent button that would enhance your post with AI, rewriting whatever you had written into whatever the model preferred. Engadget notes that button was removed when the slop report rolled out, on the same day.
The company has also been cracking down on comments generated at scale with little or no human involvement, which suggests the slop problem was never confined to posts. A platform that hands out paintbrushes shouldn't be surprised by the murals.
Forty-one percent is the number that forced this
The button didn't appear from nowhere. In July, the AI detection firm Pangram analysed LinkedIn's longform posts and concluded that 41 percent were fully AI-generated, a finding 404 Media reported at the time. LinkedIn announced the slop button and new AI classifiers a few weeks later. Coordination or coincidence, the 41 percent figure is now the reference point for everything the company says about the problem.
The 40 percent claim carries a qualifier
Read the claim carefully: members are seeing 40 percent less views on what LinkedIn classifies as AI slop. That isn't a claim that slop is down 40 percent. It's a claim that LinkedIn's own distribution changes have cut reach on posts its own classifiers flag, measured against its own baseline from a few weeks ago. The company says it uses many signals, and has built safeguards to stop individual reports being weaponised against particular members. Sensible. All of that can be true, and the number can still be doing more work than it can carry.
Srinivasan's own framing was careful. He described approaching the feature assuming good intent, and added that he's increasingly conscious of how to not sound like AI. A platform executive worrying about sounding like AI is either the most honest sentence in the post or the funniest.
What to watch: the notification's blast radius
The new feature tells authors, inside post analytics, that some members reported their post as seeming like AI. With 41 percent of longform posts flagged by an outside detector, plenty of human-written posts are going to get that message. The interesting question isn't whether the button works. It's what a writer does after getting the notice, and whether the classifier can tell a stiff human from a language model. On that, LinkedIn has published nothing.