Substack is launching a new feature that allows readers to scan posts, notes, replies, and comments for AI-generated text. The tool, which integrates technology from Pangram, will appear in the top-right menu of articles on the website and iOS app. Users can select the scan option for any content longer than 100 words to receive an estimate of how much text was likely created or assisted by artificial intelligence. An Android version is scheduled for release soon. The announcement comes as writers and publishers grapple with the growing prevalence of machine-assisted content on the platform.
This update addresses the practical need for verification without making definitive claims about authorship. Detection models are not perfect and often struggle with nuance, meaning the results serve as guidance rather than proof. The feature aims to restore some confidence in reading habits by giving users a quick way to assess the origin of the material they consume. It does not prevent AI use but offers a method to identify it after the fact.
