LinkedIn did not ban AI. It started letting the feed vote on it.
By 24 August 2026, more than one million people had clicked “Seems like AI slop,” a control the network shipped at the end of July. Hari Srinivasan, LinkedIn’s chief product officer, said accounts that copy-paste AI-written posts are now seeing around 40 percent fewer views than before the button existed.
That number is the story. Not a model release. Not another caption generator. A distribution penalty on the exact workflow most teams still use: prompt, copy, paste, publish.
Srinivasan was careful on two points. First, no single report decides how far a post travels. Reach still comes from a mix of signals. Second, LinkedIn still wants people to use AI. The company draws a line between AI and slop. The first can make a real person sound more like themselves. The second is lazy compilation — the same cadence, the same “here’s the thing,” the same empty authority, posted at scale.
The product changes match that line. LinkedIn pulled the “enhance your post” helper that made one-click slop easy and replaced it with a proofreader. Behind the feed, automated systems were already catching hundreds of thousands of AI-slop comments a day, and billions of automation attempts — mass posting, mass commenting — in the last couple of months. Users will also start seeing notes in their own analytics when viewers think a post is too AI. The crowd is now part of the ranking loop.
For anyone running an agent on social, this is the week the rules got public. An agent that dumps the same block onto LinkedIn, X and Instagram is no longer just a brand-risk. On LinkedIn it is a reach-risk. The feed has a dump button, and people are using it.
The workable pattern is the one we already tell teams for permissions: the model writes, a human keeps the voice, the scheduler sends. Draft-only access. A daily cap so reviewers are not rubber-stamping at midnight. Brand keys split from personal keys. If your agent can hit publish on the company page, the slop button is not your main problem.
ASONET’s useful job here is not “more AI captions.” It is the pause between draft and live: rewrite in the person’s tone, swap the generic hook, put a real detail in the first line, then queue. LinkedIn is training the market to reward that pause.
Source for the counts: reporting on 24 August 2026 quoting Srinivasan, including The Register’s write-up of the million-click mark and the 40 percent view drop.
