Substack has launched a tool that can detect whether a piece of writing was generated by AI. The tool is powered by a company called Pangram and works on posts, notes, replies, and comments across the platform. The feature was announced on July 21, 2026 (The Verge).

The tool is available on web and iOS, with Android support coming "soon." Readers can check any Substack content longer than 100 words by selecting "Scan for AI text" from the three-dot menu in the top-right corner of a post. Writers can also scan their own drafts and report results they believe are wrong.

Substack's co-founder and CEO Chris Best said the tool is meant to increase transparency as AI becomes more common on social media. Best coined the term "Claudefishing" to describe what happens when a reader expects a human-written post but is actually reading something AI generated. The concept is outlined in Substack's own post on the subject (Substack).

Alongside the detection tool, Substack is launching a "How I make this" statement feature. This lets creators explain their writing process to readers. The two features work together: one lets readers check text for AI involvement, the other lets writers proactively disclose their methods.

Best was open about the limitations. He said Pangram can only detect whether AI was used to produce the text, not whether a human put careful effort into it or whether AI was used merely as a research aid. A writer who drafts everything in their own words but uses an AI tool for fact-checking or brainstorming might still get flagged. Meanwhile, a piece that was mostly AI-generated but then rewritten by a human might not get flagged. The tool tells you where the text came from, not how much care went into it.

Pangram, based at pangram.com, brings real technical credentials to the partnership. Its latest detection model, version 3.3, was described in a July 9, 2026 blog post (Pangram). Pangram claims version 3.3 has a 0.01% false positive rate, meaning about one in every ten thousand human-written texts would be incorrectly labeled as AI-generated. That matters because false positives — wrongly accusing a real person of using AI — have been the biggest problem with detection tools since they first appeared.

Pangram's technology extends beyond Substack. The company offers a Chrome extension called "Feed Scanner" that can detect AI-generated content on X, LinkedIn, Substack, Medium, Reddit, and Google Docs (Pangram). Pangram also had a research paper called "EditLens" accepted for publication at ICLR 2026 (Pangram), giving it credibility in the machine learning research community.

The broader context here is how this fits into Substack's business. Substack depends on readers paying to subscribe to writers they trust. If AI-generated content floods those subscription feeds, that trust is undermined. Best has framed this as a matter of the reader-writer relationship, not traditional content moderation. The platform is not deleting or hiding flagged content. It is handing readers a tool and letting them decide.

There is a real gap between a 0.01% false positive rate measured in a lab and how the tool performs on a live platform full of casual posts, investigative journalism, non-native English speakers, and unconventional writing styles. The in-product option to "report inaccurate results" is an acknowledgment that the system will make mistakes.

The "How I make this" feature may turn out to matter more than the scanner itself. Voluntary disclosure shifts the question from a technical contest between AI generators and detectors to a social habit. If enough writers adopt it, readers learn to judge trust on their own rather than relying on a score. If few writers use it, the scanner becomes the main signal, and the whole burden falls on Pangram's accuracy.

For now, Substack has paired a detection tool with a disclosure feature, betting that the combination gives readers enough information to navigate a world where AI-generated content is becoming common.