Reddit fights AI-powered spam with AI, blocking millions of posts daily

Reddit is turning to the same technology that has made spam proliferation easier—large language models—to combat the flood of artificial content overwhelming its platform. The company has built new spam-detection systems powered by LLMs that are designed to catch patterns of coordinated inauthentic behavior that older filtering systems routinely missed.

Scale of the problem#

The challenge is substantial. Reddit blocks approximately 23 million spam views every day and identifies roughly 25,000 new spam posts and comments daily. The surge in spam has accelerated as generative AI tools have become more widely available, making it trivial for bad actors to generate convincing synthetic content at scale.

How the new approach works#

Rather than relying on traditional pattern-matching, Reddit's LLM-based tools examine the subtle, interconnected signals of fake engagement and artificial activity. According to the company, this approach proved more effective than legacy systems. Between January and March, Reddit reported a 20% drop in the amount of spam users encountered compared with the prior quarter.

"We leverage LLMs to catch the highly subtle, coordinated patterns of fake behavior and artificial hype that older systems once missed," the company stated.

The broader moderation landscape#

Other major platforms have adopted varied approaches to managing AI-generated material. YouTube, Meta, and Instagram permit users to publish AI-generated content provided they label it transparently. TikTok has gone further, allowing users to control their exposure to algorithmic recommendations for such content.

Faster AI detection capabilities could yield broader benefits—platforms may be able to identify harmful content like hate speech more rapidly. However, content moderation researchers have consistently cautioned that algorithmic systems alone are insufficient; effective moderation requires human oversight working in tandem with AI tools.