On 21 August 2026, LinkedIn published the first figures on its “Seems like AI slop” button: more than one million reports since its launch on 30 July, and 40% fewer views for content the platform classifies as “slop”, generic posts produced with little effort (announcement by its Chief Product Officer Hari Srinivasan). For brands and executives active on the platform, the question has moved on from whether AI can be spotted to which writing habits trigger the penalty, because the available data show that the structure of the text is what gets flagged, whatever tool produced the first draft.
The facts
- Hari Srinivasan, LinkedIn's Chief Product Officer, announced on 21 August 2026 more than one million uses of the “Seems like AI slop” button launched on 30 July, and 40% fewer views, compared with a few weeks earlier, for content its classifiers place in that category. Flagged authors now receive a notification in their post analytics, and LinkedIn says it applies safeguards so that targeted reporting cannot unfairly penalise a member (Engadget).
- According to a study published on 9 July 2026 by the AI detector Pangram, based on around one million posts sampled across five platforms, 41% of long posts (250 words and more) served to LinkedIn users read as entirely AI-generated, the highest rate of any platform studied; LinkedIn accounts for 62% of all the AI content detected while representing only about a third of the posts analysed (404 Media).
- The editing tool MagicPost, on a sample of 18,784 posts, measured that in May 2026, compared with April, every account bracket above 1,000 followers lost median impressions, by up to 30% for accounts with 25,000 to 50,000 followers, while accounts under 1,000 followers gained 8%.
What actually tips a post into being flagged?
The analyses agree on one point: the trigger is structure rather than vocabulary. Testing the feature in late July, 404 Media spotted a flaggable post within six seconds of scrolling, without reading a single sentence closely, from its one-line paragraphs, its emoji bullets and its “it's not X, it's Y” construction, a turn of phrase whose visibility LinkedIn had already reduced in its algorithm before the button existed. MagicPost puts a figure on each marker, in reach relative to each author's usual average: generic advice of the “stop X, start Y” kind costs 6.7% of reach, the contrast construction “it's not X, it's Y” costs 4.9%, a “curiosity gap” opening 4.3%. The most discriminating factor, according to the analysis published by VYRAL, remains the complete absence of concrete detail: no real figure, no named person or company, no dated moment.
Some habits are rewarded rather than penalised: in the same MagicPost data, a sincere closing question gains 3%, an admission or an openly stated vulnerability 7 to 10%. The phenomenon also remains a minority one across authors: of 45,965 posts analysed by the same tool, only 3.1% read clearly as AI, and 73% as human writing. One category even runs against the general trend: according to Originality.ai, AI-generated “leadership and inspiration” posts outperform their human equivalent by 75%, while healthcare and the public sector reward human writing, which limits the real reach of the flagging on the most formulaic content of all. The penalty therefore targets uniformity rather than the tool, the risk of smoothed-out content we had already documented across the marketing function.
AIxH's view
Measurements like this confirm what we already check before every client post in our Content & Social media team: a post that anyone could generate identically, in thirty seconds, on any subject, does not deserve to be published under someone's name. The scrutiny bears on the structure of the text and on the specificity of what it says, whatever the role of AI upstream: this is exactly the logic of “human editorial direction, AI-accelerated production” that our social media agency in Luxembourg applies to the accounts it runs, with a brief fed by figures, names and dated facts specific to each client. If you are wondering where your LinkedIn presence sits on this grid, a quick audit of your latest posts is enough to find out.