Substack adds built-in AI detection powered by Pangram

Reading a viral essay and wondering “did a human actually write this?” Substack just built a button for that.

The first platform with detection built in

On July 21, 2026, the platform partnered with Pangram, one of the strongest AI-detection tools on the market, and became the first major publishing platform with AI detection built right in. Readers can now scan any post longer than 100 words and get an estimate of how much of it was written by AI.

The rollout specifics: it works on posts published from July 21 onward, on web and iOS, with Android coming.

Substack is also adding a disclosure space where writers can state upfront if and how they used AI. Its CEO frames the whole feature as a defense against “claudefishing”: passing off AI-generated text as your own human writing.

The fine print on detection

The caveats matter as much as the feature. Detectors estimate, they do not prove. Pangram can flag AI-styled text, but it cannot tell whether AI was a ghostwriter or just a brainstorming buddy, and no detector is immune to false positives.

That distinction will shape how the button gets used. A high AI score on a newsletter is a data point, not a verdict, and writers who used AI for outlining or editing may score very differently from writers who pasted a chatbot’s output wholesale. The tool measures style, and style is circumstantial evidence.

The signal under the feature

Still, the strategic message is loud. Substack is the platform where writers make real money from direct reader relationships, and it just told its audience: human writing is the product, and we are ready to check.

In a content ecosystem drowning in generated text, verifiable humanity is becoming a selling point. Substack just made it a feature. The next “is it AI?” argument about a viral essay now comes with a scan button attached.