LinkedIn says over a million members have used its 'Seems Like AI Slop' button since launch — but the 40% drop in views came from somewhere else

In the same LinkedIn post, Srinivasan said members are now seeing roughly 40% fewer views on the content LinkedIn classifies as slop. Those two numbers have been read together since he published them. They shouldn’t be.

Asked how much of the decline came from member flags, LinkedIn’s Amanda Purvis told Moneywise the 40% is specific to the content classifiers the company rolled out — the detection software it announced the same day as the button.

If you post on LinkedIn to attract clients, businesses, or even recruiters, 40% is a lot of reach to lose. And whether your post lands in that bucket is now partly up to the people reading it.

What the button does, and what changed

The “seems like AI slop” option appears directly under the three-dot menu on any post or comment on LinkedIn. Srinivasan clarified in the post that no single piece of feedback determines how a post gets distributed, since the company looks at many signals together. He also mentioned that LinkedIn has built-in safeguards so members can’t use the feedback to unfairly target other people.

Those other signals matter when reading the 40% figure. LinkedIn rolled out new classifiers to identify slop and low-quality posts on the same day it launched the button, so the drop reflects both. Srinivasan didn’t break out how much came from reader flags.

Srinivasan also added a screenshot to the LinkedIn post showing the message an author may now see: “Some members told us this post seems like AI.” The rest of the note says LinkedIn is passing it along as feedback to consider for future posts. Purvis said in a response to Moneywise that the message is rolling out to members over the coming weeks.

Meanwhile, the button was one of several changes LinkedIn announced on July 30. Srinivasan said that LinkedIn was catching hundreds of thousands of automated comment attempts every day and had blocked billions of other automation attempts over the previous couple of months. He added that the “Enhance post” tool, which used to draft posts for members, was being replaced by one that proofreads without changing a member’s voice.

How much of LinkedIn was AI-written before the button

The company hasn’t yet clarified that, but two detection brands — Pangram Labs and Originality.ai — had already measured it, and both came back with high figures. Both scans were finished before LinkedIn introduced the button, so neither says anything about what the feed looks like now.

Pangram Labs runs a browser extension that scans posts as people scroll past them. Its July 9 report drew on 1,002,627 posts collected that way since the extension launched April 24. It found more than 40% of LinkedIn posts over 250 words flagged as fully AI-generated.

LinkedIn accounted for only about a third of everything Pangram scanned, but it produced 62% of all the AI content the company flagged, more than any other platform in the study.

Originality.ai put the figure far higher, at 81.2% of the 5,000 public LinkedIn posts it examined from July. The two aren’t measuring the same thing. Pangram counts only posts over 250 words that come back as fully machine-written, while Originality.ai looks at posts of 100 words or more and calls one likely AI when its model is at least 50% confident. That gap explains most of the distance between the numbers.

However, Pangram says its model wrongly flags human writing only 0.01% of the time, but its posts came from people who had gone out and installed a slop-spotting tool, so they probably already suspected their feeds contained AI-generated content.

What LinkedIn still hasn’t explained

Moneywise asked whether the volume of AI-generated posts on the platform is falling, rather than just the views those posts receive. Purvis said LinkedIn doesn’t have data to share at this point.

But LinkedIn hasn’t yet clarified what separates a slop post from an ordinary one in its systems, or how much feedback it takes before that message reaches an author. Srinivasan said at launch that slop is hard to define, and that the definition changes.