Substack Now Flags AI-Written Newsletters Before You Read Them
The newsletter platform has partnered with AI detection software Pangram to show readers how much of any post was written by a human. It won't punish writers who use AI, but it will make them harder to hide.
Key points
- Substack announced a partnership with Pangram, an AI writing detection company, to scan newsletters, comments, and replies for AI-generated content.
- The tool is available for any piece of writing above 100 characters on the Substack app.
- Substack CEO Chris Best framed AI disclosure as a feature, not a penalty, saying writers should still do "the hard part".
- Writers can run the detector on their own drafts before publishing and can flag scans they believe are wrong.
- Substack will offer an optional "AI Author's note" field so creators can openly disclose how they use AI in their work.
If you subscribe to newsletters on Substack, the platform where writers publish directly to paying readers, you may soon notice a new label on posts you read. Substack has teamed up with Pangram, a piece of software that estimates how much of a text was written by a human versus an AI, to let readers scan any post, comment, or reply inside the Substack app.
The estimate appears as a percentage split: how much reads as human, how much reads as machine. It is not a perfect science. AI detection tools carry a known margin of error, and Substack is letting writers flag scans they think got it wrong.
Substack is not banning AI writing. That matters. The company told TechCrunch that the goal is transparency, not punishment. Writers who use AI tools are encouraged to add an optional disclosure note explaining their process, something like a recipe card for how a piece was made.
CEO Chris Best put it plainly in an online chat with Pangram founder Max Spero. "We'll do everything for you except the hard part," he said, describing his original pitch to writers. The hard part, in his view, is having an idea worth reading. Software, he argued, should handle the rest.
The feature works on any content above 100 characters, which is roughly two short sentences.
What does this mean for readers?
For now, it means you get more information before you decide whether a newsletter is worth your time or money. A scan won't tell you a piece is bad. A writer might use AI to check grammar or structure a draft, and the final voice might still be entirely theirs. What the label signals is that something beyond a human keyboard was involved.
In the short term, some popular newsletters may face awkward questions if their AI use turns out to be heavier than readers assumed. Trust is fragile in independent media. A disclosure that surprises a loyal subscriber can sting.
Longer term, the bet is that clear labelling builds more trust, not less. Readers who know what they are getting tend to stick around. Readers who feel misled do not.
Substack joins a growing list of platforms adding AI content labels. Social media sites already flag AI-generated photos and videos. Music streaming services have started labelling AI tracks, with some penalising them in recommendation systems.
The Pangram integration is live now in the Substack app.
