Substack adds an AI detector to help spot blogs written by no one
The Verge Emma Roth ● Covered by 3 sources
Substack just added a tool that scans posts to guess if AI wrote them. It's trying to stop readers from unknowingly trusting words nobody actually thought through.
Based on reporting by The Verge, Emma Roth — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Substack is rolling out a feature that lets anyone check whether a post, note, reply, or comment was likely written by AI. The tool, powered by a detection company called Pangram, scans text longer than 100 words and spits out an estimate of how much of it might be machine-generated. It's live now on the web and iOS, with Android coming at some unspecified later date, and readers can trigger a scan from the three-dot menu on any post.
Cofounder and CEO Chris Best framed the move less as an anti-AI crusade and more as a trust repair job. His argument: the issue was never that AI exists or that its output is bad. The issue is when a reader assumes there's a person behind the words and there isn't. Best gave this phenomenon a name — Claudefishing — the awkward moment when you invest attention in something that turned out to have zero human thought behind it.
Substack isn't just adding a detector and calling it done. Writers now get a "How I make this" statement where they can explain their process, whether that's fully human, AI-assisted, or something in between. Creators can also run their own drafts through Pangram before publishing, and there's a mechanism to flag results that seem wrong, which suggests Substack knows the tool won't be perfect.
Best was upfront about the limits, too. Pangram can flag that AI touched a piece of text, but it can't measure how much genuine effort or judgment a human put into it, and it can't tell you if AI was merely used as a research source rather than the actual author. That's a meaningful gap, but Best's bigger point is about incentives: platforms that let fake content pass as real, he argues, end up punishing the writers who are doing the work honestly, AI-assisted or not.
My take — AI-written commentary, not fact-checked reporting
Slapping a detector on top of a platform is an easy headline, but it doesn't fix the actual incentive problem — writers who quietly use AI well have nothing to fear, and the ones gaming the system will just get better at dodging Pangram's guesses. Substack deserves some credit for pairing this with the "How I make this" disclosure option, since transparency from the writer's side is a far sturdier fix than a probability score from a third-party tool. Detection is a patch, not a cure, and everyone should stop pretending otherwise.
Read more about this at: The Verge