substack now lets readers scan your newsletter for ai
Substack turned on AI detection this week, built with Pangram. Readers can scan any post, note, or comment over 100 words published from July 21 forward and get an estimate of how much was written by a machine. Writers can disable it on their own work, pre-scan their drafts before publishing, and dispute a bad read.
The more interesting half is the part that isn't detection. Alongside it, Substack added a field called "how I make this" — a place to describe your actual process in your own words. The company was clear this isn't about banning AI-assisted writing. It's about making the process legible. Chris Best framed it as good use of AI: the platform does everything for you except the hard part, which is having something worth reading in the first place. Substack also took a swing at LinkedIn on the way out, coining "claudefishing" for manufacturing human connection out of slop — a bold move given that no detector, including this one, is reliably accurate. Substack said as much itself.
I write a newsletter with AI in the room. So does most of the valley by now — the chamber email, the nonprofit appeal, the realtor's market update. What shipped this week isn't a verdict on that. It's the arrival of a norm: not did you use AI, but can you say how. That's a much easier question to answer well, and the answer is starting to be worth something.
The detector will be wrong sometimes. The disclosure field never is.