Substack Deploys Pangram AI Detection
Substack has integrated AI detection technology from Pangram into its platform, a move designed to reveal how much of a post was likely generated by artificial intelligence. Alongside the automated assessments, the company has introduced a new interface allowing creators to explicitly disclose their use of AI tools.
Transparency Labels for Readers
The system provides readers with an estimate of AI-generated content within a post. According to Substack’s official announcement, the feature aims to increase transparency regarding content origin.
Beyond automation, Substack created a dedicated space for writers to describe their process. The company stated the goal is to ensure “people should know what they’re getting” when they read a newsletter.
The Engineering Limits of Detection
Substack acknowledges that its new tools are not foolproof.
Research from outlets like The Atlantic has noted that AI detectors often struggle to maintain high accuracy rates across different writing styles and prompts.
Taking Aim at Platform “Slop”
In its announcement, Substack offered a pointed critique of how other platforms manage AI content. The company contrasted its transparency-focused approach with services like LinkedIn, suggesting that other networks have become saturated with low-quality, AI-generated “slop.” By coining the term “Claudefishing” to describe the practice of using AI to mimic human connection, Substack is positioning its new labels as a defense of the reader-writer relationship.
Future Expansion and Android Rollout
Despite the firm language regarding competitors, Substack maintains that the Pangram integration is an evolving project. The company hinted at the future addition of features regarding AI preferences and content filtering. For now, the detection tools remain limited to web and iOS, with an Android rollout expected to follow.
Keep reading