LinkedIn's AI Slop Button Has Been Clicked a Million Times, and Flagged Posts Are Already Losing 40% of Their Reach

Three weeks ago, LinkedIn added a small option to the three-dot menu on every post: "Seems like AI slop." As of August 24, members have clicked it more than a million times, and LinkedIn's chief product officer Hari Srinivasan says the posts it classifies as slop are pulling roughly 40% fewer views than before the button existed (theregister.com). That is a rare thing in this industry: a feature that shipped with a hypothesis and came back three weeks later with a number attached to it.

Why LinkedIn built this in the first place

The button did not appear out of nowhere. In early July, the AI detection firm Pangram estimated that 41% of long-form LinkedIn posts and 30% of short ones were likely AI-generated, the highest share of any major platform it tested (theregister.com). Anyone who has scrolled the feed in the last year could have guessed that number without a study behind it. The three-emoji hook, the "here's what nobody tells you," the numbered list that resolves into a call-to-action generic enough to fit any industry, all of it became background noise fast, and LinkedIn's own engagement data apparently agreed.

What actually changed, and what did not

The mechanism matters more than the headline. LinkedIn is not just running a classifier against your text anymore, it is feeding a million real flags from actual readers into its detection models, and that is a harder signal to game than a static AI-detector because it measures how a post lands with a person, not just its statistical fingerprint. Srinivasan has also been careful to draw a line: LinkedIn is not punishing AI use, it is punishing generic output. Someone using AI to tighten a sentence or get past a blank page is fine. A post that reads like it was written for nobody in particular is the one losing 40% of its reach.

LinkedIn backed that distinction with a product decision of its own. It is phasing out its "write this for me" AI feature and replacing it with one that only proofreads what a person has already written. That is a company that built a generation tool, watched what it did to its own feed, and decided the fix was less generation and more editing. Worth sitting with, given how many agencies reach for exactly the kind of tool LinkedIn just backed away from.

The part that turns into a client conversation

Here is where this stops being a platform-trivia story and becomes an operational one. LinkedIn is rolling the flagging data into post analytics next, so a client who logs into their own dashboard will start seeing, right next to impressions and engagement rate, whether a post got marked as "seems like AI slop." That used to be a private judgment call between you and whatever tool you used to draft the post. It is about to be a visible metric sitting in front of the person paying you.

If your process on a client account is generate, skim, schedule, that gap is no longer invisible. It shows up as a number a client can screenshot and ask about. The agencies coming out ahead here are not the ones swearing off AI drafting, they are the ones already treating a generated draft as raw material instead of a finished post: swap in the real number the client mentioned on a call, the specific example only their business has, the phrase that actually sounds like the founder instead of a prompt. That editing pass used to be a nice-to-have. LinkedIn just turned it into the difference between a post that reaches people and one a million strangers have already trained the algorithm to bury.

What to actually watch

The next few weeks are worth paying attention to for two reasons. First, LinkedIn has more detection work coming behind the scenes, so the 40% number is a floor, not a ceiling, if the pattern holds. Second, once the flag shows up inside client-facing analytics, expect it to become a routine line item in account reviews the same way engagement rate or follower growth already is. Anyone managing multiple client accounts on LinkedIn should build an actual editing checkpoint into the content calendar now, before a client finds the flag before you do.