Substack добавил AI-детекцию на базе Pangram для борьбы с «Claudefishing»

Substack added AI detection based on Pangram to fight "Claudefishing"

In July 2026, Substack launched a built-in AI text detection feature based on the Pangram service — readers got a button in the interface that flags posts presumably written by a neural network. The platform announced this on its own blog under the headline "Against Claudefishing."

What is Claudefishing

Claudefishing is a term coined by Substack for the moment when a reader realizes that a post they admired for its "live human voice" was actually generated by a bot. The word is formed from the name Claude — the chatbot by Anthropic — and "catfishing," online deception under a fake identity.

Substack put the term right into the headline of the manifesto accompanying the feature's launch. The platform positions itself against places where a machine increasingly imitates the author's voice, turning "humanness" of text into an explicit competitive advantage.

• Term: Claudefishing — from "Claude" + "catfishing"
• Detection engine: third-party service Pangram
• Announcement format: a post on the Substack blog "Against Claudefishing"
• Mechanics: an indicator button next to the post, the decision is left to the reader

How the detection works

The check runs on the Pangram engine — a third-party service specializing in recognizing AI-generated text. Substack didn't build a detector of its own but integrated someone else's: an indicator appears next to the publication that lets the reader judge whether the text was written by a human or a model.

At the same time, Substack leaves the decision to the audience: the platform doesn't automatically block or delete suspicious posts, but puts a verification tool directly into readers' hands. This is a fundamentally soft approach — a label, not a ban.

Why this matters for authors

Substack has officially acknowledged, for the first time, the problem of AI content within its own ecosystem — a service built precisely on trust in a personal authorial voice. For a newsletter platform where subscribers pay for a specific author, "a bot in a human mask" strikes at the core asset — the reader's belief that there's a living person on the other side of the screen.

"Claudefishing is the moment when you realize: the post you admired for its human voice was actually written by a bot," the

Substack blog states.

Relying on a third-party detector is a conscious risk. AI detectors are regularly criticized for false positives, especially on texts by authors for whom English is not their native language. That's precisely why Substack leaves the final judgment to the reader rather than issuing a verdict on the platform's behalf.

What this means

The fight for the "humanness" of content is moving from public debate into the product itself: platforms are starting to build AI detection directly into the interface. Substack is betting that subscribers want to read humans, not models — and gave them a button to check it. Whether Pangram proves accurate in practice will be shown by the feature's first months of operation.

Frequently asked questions

What is Claudefishing?

Claudefishing is Substack's term for a situation where a text that impressed a reader with its "lively" style turns out to have been written by an AI bot. The name comes from Anthropic's chatbot Claude and the word "catfishing."

What does Substack's AI detection run on?

The feature is built on the third-party service Pangram, which specializes in recognizing AI text. Substack integrated it into its interface as an indicator button, but left the final decision about whether to trust a post to the reader.