Substack tells TechCrunch the Pangram integration is meant to nudge writers to disclose AI use rather than ban it — even as it could expose newsletters that aren't fully human-written.
Substack has folded Pangram, a tool that spots AI-written text, into its app, giving readers a way to estimate how much of a post came from a person versus a machine. The integration, added recently, returns a split between human and AI authorship.
Inside the app, the scan applies to any post, note, reply or comment that runs longer than 100 characters.
Substack told TechCrunch the tool is not built to ban or penalize writing done with AI help, but to push writers toward disclosing how they work. It will also let authors attach an optional AI author's note, spelling out that they leaned on the technology.
Writers get some cover against false positives. They can run their own drafts through Pangram before publishing, and if they judge a scan of their own work to be wrong, they can flag it and have it removed.
Chief executive Chris Best endorsed the move as a good use of AI: software should do everything, he argued, except the hard part — an idea worth reading that people will care about and pass along, something both very hard to produce and hugely valuable, and the part that in his view should stay human.
The step could cut against Substack's own near-term interest: revealing which newsletters aren't entirely human-written might dent confidence in its network and its reputation for quality. Over the longer run the calculus could run the other way — such detection could keep the platform clearer of AI slop and give readers a better sense of when what they are reading came from a person, placing Substack alongside other platforms moving to label AI-generated content.