Substack adds AI detector to fight 'Claudefishing' and fake authorship

Substack is shipping an AI detection tool directly into its platform, letting readers scan posts, notes, and comments to see if they were likely written by a machine. The feature, powered by startup Pangram, is rolling out on the web and iOS now, with Android coming later. It kicks in for any text over 100 words, accessible from a three-dot menu on any post.

CEO Chris Best isn’t framing this as an anti-AI crusade. He’s targeting a specific kind of deception he calls “Claudefishing” — when a reader invests their attention in something they believe has human thought behind it, but doesn’t. “The core problem is not people using AI, or the quality of its output,” Best wrote in the announcement. “The problem is when there is a mismatch between a reader’s expectation and reality.” That’s a more nuanced position than you’ll find on most platforms, and it’s worth paying attention to. He’s not saying don’t use AI. He’s saying don’t lie about it.

Alongside the scanner, Substack is introducing a “How I make this” section where writers can disclose their process — whether they use AI for research, drafting, or not at all. Writers can also scan their own drafts before publishing and flag any detection results they think are wrong. It’s a transparency play, not a policing one. The goal, Best says, is to help readers decide where to spend their time. “When readers have to wonder if what they’re reading is real, it undermines trust in authorship and threatens the livelihood of writers — including those who use AI tools thoughtfully to produce work they believe in.”

I’m skeptical about how well this will work in practice. AI detectors are notoriously unreliable, prone to false positives that can unfairly tag non-native English writers or anyone with a formulaic style. Pangram might be better than most, but no detector is bulletproof. The real test will be what happens when a popular writer gets flagged incorrectly and raises hell. Substack is betting that giving writers a way to push back — and readers a way to check — is better than doing nothing while the platform fills with synthetic slop.