Substack adds AI detection as 'human-made' becomes premium
The newsletter platform is giving readers a tool to spot machine-generated text, betting that verified human authorship will become a scarce, sellable commodity.
Substack has introduced an AI-detection tool, allowing users to scan posts and comments for machine-generated text. The feature, powered by the detector Pangram, targets what chief executive Chris Best calls "Claudefishing" — the disappointment of discovering an admired piece of writing actually came from a bot. "Claude has a lot to offer," Best wrote. "But when I want Claude’s opinion, I’ll ask Claude."
Readers can now check any text longer than 100 words published from today. The tool estimates the split between human and AI input, displaying results only to the person who initiated the scan. It is currently available on the web and iOS, with an Android rollout planned.
For Substack, this is a crucial defensive business move. The platform operates on direct reader payments built on personal trust. If its feeds fill with undifferentiated, machine-generated text, that core economic engine stalls. Pangram estimates that up to 40% of text on some social platforms is now AI-generated. Substack is effectively wagering that as artificial intelligence floods the wider market, proof of human effort becomes a scarce, premium asset. As Business Insider’s Peter Kafka noted, if AI wins, "made by humans" could be a very good business.
The company is not instituting a blanket ban on AI. Writers can pre-scan their own drafts or attach a "How I make this" statement to explain their workflow. Substack framed the feature around transparency, taking a direct swipe at LinkedIn, which many view as overwhelmed by machine-written posts.
However, the strategy carries technical and social risks. AI detection remains an imperfect science, and Substack concedes the tool cannot measure whether genuine care went into a piece. The Atlantic has previously warned that over-reliance on such detectors risks sparking baseless witch hunts against writers.
Critics also argue the tool may ultimately be outmaneuvered. Developer Perry Metzger pointed out that public detectors can be used to train AI systems to evade detection. Researcher Mor Naaman called detection a losing battle, predicting writers will simply route around the checks. As one commenter noted, the presence of AI does not prove the absence of a human.
