On October 8, LinkedIn's VP of Engineering Tim Jurka explained how the platform now catches AI slop. Its classifiers check every post that travels beyond your own network, and LinkedIn says they're 94% precise.
That's the part of the feed that decides whether strangers ever see your post.
What LinkedIn changed
- On July 30 LinkedIn added a Seems like AI slop option to the menu on every post. More than a million people used it in the first few weeks.
- Content LinkedIn classifies as slop now gets about 40% fewer views than it did a few weeks earlier.
- The detector now covers every post shown outside your immediate network, which is exactly where new reach comes from.
- LinkedIn also removed its own AI rewrite feature and replaced it with a proofreader.
How the detector learns
Jurka describes a teacher and student setup. Large models spot new slop patterns and label examples. Smaller, faster models learn from those labels and run on the live feed.
On top of that, AI agents check posts against specific policies, for example whether a post is promotional, celebrates an achievement or is timely. When a case is unclear, they hand it to human reviewers and update the policy. Member reports feed the same loop.
So the system gets sharper every week. A trick that works today gets labelled tomorrow.
AI isn't the problem. Generic is.
LinkedIn hasn't banned AI, and it says people can keep using it. What it's going after is content that looks polished and says nothing: the same hooks, the same lists, the same lessons anyone could have written.
One report doesn't sink a post. Many reports do, and authors who collect a lot of them now get a notification about it.
What we see on our own accounts
The posts that travel furthest for us are the least generic ones. Real numbers from one market, a real name, a screenshot nobody else has. One of them took a profile with 2 connections to tens of thousands of impressions in a week, as in From 2 connections to 83,285 post impressions.
What founders should do now
- Start every post with something only you know: a number, a client message, a price, a mistake.
- Name the market, the city and the company. Generic posts never name anything.
- Drop template hooks and ten point lists nobody asked for.
- Use AI to edit and check facts, not to have the idea.
- Watch reach outside your network after you post. That's the number the detector controls.
Different market, same door. You still have to knock right on LinkedIn.