LinkedIn shares how it detects AI slop and cuts views by 40%

LinkedIn added a report button for AI slop in August. More than 1 million people used it, and the platform now catches 94% of AI-slop posts beyond immediate networks.

What happened

Social Media Today reports that LinkedIn has shared new insight into its evolving battle against AI slop, which the platform defines as generic and/or repetitive posts that added no real insight. In August, LinkedIn added a new option that let users report suspected AI-generated content in-stream. Less than three weeks after launch, more than 1 million people had used the feature to weed out AI junk and clean up their feed experiences.

On Oct. 8, LinkedIn VP of Engineering Tim Jurka posted an explanation of how LinkedIn identifies AI slop. He said the platform uses user feedback from the AI slop button to train its detection systems. Jurka described a teacher-student setup: larger “teacher” models identify new AI-slop patterns and generate training data so smaller models can pick up those patterns rapidly.

“Specifically, we use what’s known as a teacher-student setup. The larger ‘teacher’ models are designed to keep up with new AI-slop patterns and accurately identify them, which generates the high-quality training data we need to ‘teach’ or train our smaller models to pick up on those new patterns rapidly.”

Jurka also said LinkedIn deploys a range of AI agents assigned to specific policies. Each agent evaluates content against criteria such as whether a post is promotional, celebrates an achievement, or is timely. When an agent cannot reason about a complex example, it surfaces the example to human reviewers, learns from their guidance, and evolves its policy for similar situations.

Key facts

• LinkedIn added a user option in August to report suspected AI-generated content in-stream.
• Less than three weeks after launch, more than 1 million people used the AI slop reporting feature.
• Jurka said classifiers now cover all posts distributed beyond a user’s immediate network with 94% precision at detecting AI slop.
• LinkedIn previously reported that it reduced views of AI slop by 40%.
• AI slop is defined as generic and/or repetitive posts that added no real insight.

Our analysis

The key change is that LinkedIn is no longer relying only on platform-side guesses about what counts as low-quality AI content. It has converted user reports into training data through a teacher-student model. That suggests the definition of AI slop will keep shifting as new repetitive patterns emerge, making it harder for creators to rely on templated AI output that once slipped through.

This matters for reach. Because LinkedIn says it now classifies all posts distributed beyond a user’s immediate network, a post that looks promotional, generic or achievement-only may be detected and hidden before it gains broader visibility. The 40% reduction in AI slop views shows the platform is actively suppressing this content, not just labeling it.

The source also frames a deeper trust problem: the rise in AI-generated posts has increased overall skepticism about every post in social feeds. For brand marketers and community operators, that means even authentic content may face more skeptical audiences, and the bar for proof of human insight is rising.

What it means for operators

• Audit LinkedIn drafts against the platform’s own criteria: promotional, celebrates an achievement, or timely. If a post reads as generic or repetitive with no real insight, rewrite it before publishing.
• Do not post AI-generated images or video without a specific human story, process detail, or operational lesson. Social Media Today notes most such posts lack real human connection or value and are more likely to be suppressed.
• Treat reach beyond immediate connections as a new threshold. Since LinkedIn now classifies all distributed posts at 94% precision, focus on original, first-person experience rather than volume of templated posts.
• Use the feedback loop. If posts are reported or hidden, note the pattern and adjust your content guidelines; the system learns from human reviewers and user reports, so repeated slop-like signals can train the model against your brand.