LinkedIn's algorithm is now suppressing generic AI content, and a 46% enforcement surge is the proof

LinkedIn's September DSA disclosure reports a 46% jump in detected inauthentic activity in H1 2026, as the platform blocks hundreds of thousands of automated comment attempts daily and demotes AI 'slop' beyond a poster's immediate network.

LinkedIn’s September 1 EU Digital Services Act disclosure reports a 46% rise in detected inauthentic activity in the first half of 2026 versus the back half of 2025, and the numbers underneath that figure describe a platform quietly rewriting the economics of AI-assisted posting.

Chief product officer Hari Srinivasan, in a late-July post cited by TechCrunch and Social Day, said LinkedIn is now blocking “hundreds of thousands” of automated comment attempts daily and has stopped billions of automation attempts in recent months. In H2 2024 the company removed 80.6 million fake accounts at registration, more than 97% caught by automated defences before any member reported them. The framing is deliberate. “People come to LinkedIn to connect with real people and share their real perspectives, ideas, and expertise,” Srinivasan wrote.

The DSA filing also shows LinkedIn added 1.4 million EU users in H1 2026, reaching 56.5 million active in the region. That’s roughly 30% of the EU member base and implies a global active audience near 433 million, far below the 1.3 billion registered-member figure LinkedIn typically cites. Growth is slowing while enforcement accelerates.

The classifier build-out is where this gets operational. LinkedIn is retiring its own “enhance your post” AI writing feature and replacing it with a proofreading tool. It shipped a “Seems like AI slop” report button that more than 1 million people used in its first two weeks. Content the platform internally tags as slop is seeing 40% fewer views. Creator product lead Sam Corrao Clannon described the target as posts that are “potentially sophisticated or polished in its presentation, but lacks substance… empty text that’s posted to take up space and garner attention without effort on the other side.”

The scale of the target is real: a Pangram study cited by Social Day flagged 40% of long-form and 30% of short-form LinkedIn posts as fully AI-generated. Social Day’s read of Srinivasan’s guidance is that flagged posts are “much less likely to spread beyond a user’s immediate network,” and that high-volume posting and templated comments are behavioral signals the classifiers weigh.

For B2B small businesses whose pipeline runs through LinkedIn, the practical read is that the old volume-first playbook is now a distribution penalty. The rewarded behavior looks like a human making a deliberate choice: fewer posts, grounded in specific expertise, sent by a person. That pattern echoes the customer-approved send patterns emerging in enterprise agent rollouts and the multiplayer conversation-context approaches Anthropic is testing, both of which treat human judgment as the last mile rather than a bottleneck.

LemonLime studies a customer’s company, industry, and competitors before preparing LinkedIn content and prospect outreach, and the customer approves every post and send before it goes out. That’s the behavioral signature LinkedIn’s classifiers are built to reward.

LinkedIn's algorithm is now suppressing generic AI content, and a 46% enforcement surge is the proof

Sources