- Substack has launched an AI detection feature using Pangram to estimate how much of a post is human-written versus AI-generated.
- In the short term, this could hurt trust by revealing AI-written newsletters, but long-term it may help maintain quality and user trust.
- The feature is part of a broader trend of platforms labeling AI-generated content, including social media and music streaming services.
- Substack CEO Chris Best endorsed the tool, emphasizing that the hard part—ideas worth reading—should remain human, while software handles the rest.
- The detection works on posts, notes, replies, and comments over 100 characters, and writers can add an optional AI author's note.
- The tool is not punitive; it encourages transparency and allows publishers to scan drafts and report errors.