Substack adds AI text scanner for posts and comments
The Pangram-powered tool estimates AI use across Substack posts, notes, replies and comments, with rollout starting on web and iOS.
By James Whitfield · Staff Writer
2 min read
Substack is adding an AI detection tool that lets readers check whether text on the platform may have been generated or assisted by artificial intelligence. The feature matters because Substack is trying to give readers more context about who, or what, may be behind the writing they spend time with.
Substack said in a blog post Tuesday that the tool can scan posts, notes, replies and comments. It will provide an estimate of how much of a piece of text may have been produced by AI or written with help from AI.
The scanner is powered by Pangram, an AI detection company, according to Substack. The feature is rolling out on the web and in Substack’s iOS app, while an Android version is planned for release soon.
Readers will be able to use the scanner on content longer than 100 words, Substack said. The option appears in the three-dot menu in the upper-right corner of a post under “Scan for AI text.”
Substack frames the tool as a transparency measure
Chris Best, Substack’s co-founder and chief executive, said the Pangram integration is intended to improve transparency as AI-generated material becomes more common on social platforms. Best described the concern as a gap between what readers believe they are seeing and how the text was made.
Best used the term “Claudefishing” for cases in which readers put attention into material that has no human thought behind it. He said the issue is not the use of AI itself or the quality of AI output, but whether readers understand the role AI played.
Substack is also adding a way for creators to describe their own process. The new “How I make this” statement will let writers tell readers more about how their work is produced, according to the company.
Writers will also be able to run Pangram scans on drafts before publication. Substack said creators can report results they believe are wrong.
Detection comes with limits
Best cautioned that Pangram’s scanner is limited to identifying whether AI was used in making text. He said it cannot judge whether a person put substantial care into the work, and it cannot determine whether AI tools were used only as a source.
That caveat leaves room for cases in which writers use AI as part of a broader reporting, drafting or editing process. Substack’s rollout gives both readers and writers a tool for flagging possible AI involvement, while leaving the final interpretation to people using the platform.
This story draws on original reporting from The Verge.