Substack Launches 100-Word AI Detector as Platform Targets 'Claudefishing'

3 articles · Updated · The Verge · Jul 21

Substack on Tuesday began rolling out an AI detector that lets readers scan posts, notes, replies and comments to estimate whether text was AI-generated or AI-assisted.
Content longer than 100 words can be checked through a “Scan for AI text” option, with the Pangram-powered feature arriving on the web and iOS first and Android coming soon.
Chris Best said the tool is meant to curb “Claudefishing” — content that appears human-written but lacks human authorship — because that mismatch can erode reader trust and writers’ livelihoods.
Substack is pairing the detector with a new “How I make this” disclosure for creators, while also letting writers scan drafts and report inaccurate results.
Best said Pangram can detect AI use in text but not the level of human care behind it, underscoring Substack’s broader push for transparency rather than a ban on AI-assisted writing.