LinkedIn’s fight against AI slop, which it defines as generic or repetitive AI-generated posts that add no real insight, is becoming a full detection system. On Oct. 8, VP of Engineering Tim Jurka shared how LinkedIn defines and detects AI slop and what impact the work has had so far.
In August, LinkedIn added an in-stream option to report suspected AI-generated content. Less than three weeks after launch, more than 1 million people had used it. LinkedIn has previously said it reduced views of AI slop by 40%.
Detection starts with those reports. LinkedIn uses a teacher-student setup: larger “teacher” models keep up with new AI slop patterns and identify them accurately, generating high-quality training data that teaches smaller “student” models to pick up new patterns quickly.
On top of that, LinkedIn runs AI agents that are each guided by a specific policy, such as whether a post is promotional, celebrates an achievement or is timely. When an agent hits a complicated example, it surfaces it to human reviewers, learns from their guidance and updates its policy.
Using both approaches, Jurka said LinkedIn has expanded its classifiers to cover all posts distributed beyond a member’s immediate network, with 94% precision at detecting AI slop. Any post trying to reach people outside your connections passes through this filter.
Social Media Today expects demand for an AI off switch to grow across social apps as AI content erodes trust in feeds.
On LinkedIn, posts aimed at new audiences are screened. If you draft with AI, add the numbers, examples and opinions only your company can provide. Generic takes are likely to lose distribution.
The policy examples include promotional content and achievement posts. Frame awards and deal news around what you learned or the process behind the number. The same push toward originality shows up in YouTube Shorts Algorithm Update Favors Original Content.
Social posts are increasingly cited in AI answers, so content that survives the filter can compound into search visibility. See AI Search Visibility Runs Through Social Media.
Generic or repetitive AI-generated posts that add no real insight.
User reports feed large teacher models that generate training data for smaller student models, alongside policy-specific AI agents and human reviewers. Classifiers now cover all out-of-network posts with 94% precision.
The target is generic, repetitive content rather than AI use itself. Posts with original data, examples and opinions are less likely to be classified as slop.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
TikTok’s Road to Black Friday series shows how top TikTok Shop sellers prep for BFCM: Princess Polly’s 6,000 creator videos a month, Dossier’s sampling engine and SharkNinja’s LIVE strategy.
Meta shared how small businesses use its Muse AI agents: analyzing ad performance, building campaigns through conversation and automating workflows with Shopify, Canva and HighLevel. Prompts included.
Reddit outlined its U.S. midterm election plans, including the Up the Vote hub, AI detection that cut spam exposure 20% and removes nearly 2 million fake upvotes daily, and AMA rules for political ads.