LinkedIn Details How AI Agents Are Fighting Platform AI Slop

LinkedIn has offered a closer look at how it is tackling AI slop, the flood of generic, repetitive AI-generated posts that has been wearing down the quality of its feed. In a post published on Oct. 8, Vice President of Engineering Tim Jurka outlined how the platform identifies low-value content and how its detection systems are trained.

Reporting Tool Drives Training

The effort builds on a feature introduced in August that lets users flag suspected AI slop directly in their feeds. LinkedIn defines the term as generic or repetitive content that adds no real insight. The option caught on quickly, with more than one million people using it within three weeks of launch.

Jurka explained that these reports help train LinkedIn’s systems through a teacher-student setup. Larger “teacher” models track emerging slop patterns and label examples accurately. That high-quality data is then used to train smaller models so they can recognize new patterns quickly.

Policy-Based AI Agents Add Human Oversight

LinkedIn also uses AI agents, each assigned a specific policy that sets the criteria for assessing a post, such as whether it is promotional, marks an achievement or is timely. When an agent struggles with a complicated case, it can pass the example to human reviewers, learn from their guidance and refine its policy for similar situations.

By combining both frameworks, LinkedIn has expanded its classifiers to cover all posts distributed beyond a user’s immediate network. According to Jurka, they now detect AI slop with 94% precision. The company previously reported a 40% reduction in views of AI slop.

The update points to growing pressure on social platforms to manage AI-generated content as skepticism about what appears in feeds rises and users look for ways to limit it. LinkedIn’s approach of pairing automated detection with human guidance could offer a template for other platforms facing the same problem.