Substack Introduces AI Detection Tool to Verify Authenticity of Posts

Substack announced on July 21, 2026, that it has introduced a tool on its platform to determine whether articles or comments are generated by AI. The Verge’s Emma Roth reported on the development.

The tool leverages technology from AI detection company Pangram. It scans posts, notes, replies, and comments to provide an estimate of how much of the text was potentially created or assisted by AI. Currently available on the web version and iOS app, the feature is expected to roll out on Android “soon.” Users can analyze content consisting of more than 100 words by selecting the “Scan for AI text” option from the three-dot menu in the top right corner of a post.

Substack co-founder and CEO Chris Best explained the initiative, saying, “The core issue isn’t that people use AI or the quality of its output. The problem arises when there’s a mismatch between readers’ expectations and reality, particularly when readers unknowingly engage with content that hasn’t involved human thought.” Best referred to this phenomenon as “Claudefishing.”

Alongside this update, Substack has introduced a new feature called “How I make this,” which allows creators to explain their writing process to readers. Writers can also scan their drafts using Pangram and report inaccurate detection results.

Best acknowledged the limitations of Pangram’s technology, stating, “Pangram can only detect whether AI was used in text creation. It cannot determine the extent of human effort involved or whether AI tools were used as a source of information.” He further emphasized, “When readers have to question the authenticity of what they’re reading, it undermines trust in authors and threatens writers’ livelihoods. Platforms that reward inauthentic content create a race to the bottom.”

Limitations and Purpose of AI Detection

The detection provided by Pangram is not definitive. As Best himself admitted, AI detection tools can estimate whether AI was used but cannot evaluate the degree of human involvement in creating the content. There is a significant difference between using AI for brainstorming or drafting assistance and fully delegating the writing process to AI. However, the detection tool cannot distinguish between these scenarios.

To address this challenge, Substack has gone beyond offering a mere detection tool by implementing a self-reporting mechanism for creators. The “How I make this” feature serves as a way for writers to be transparent about their writing processes with readers. By allowing authors to describe their own methods, Substack aims to build trust without solely relying on AI detection results.

The timing of Substack’s introduction of AI detection technology aligns with the growing prevalence of AI-generated content on its platform. In particular, “Claudefishing”—a phenomenon where AI-generated text deceives readers into believing it was written by a human—has raised concerns about trust. Substack, as a platform rooted in newsletter subscriptions, recognizes the importance of maintaining a continual trust relationship between its subscribers and writers. The company views the rise of opaque AI-generated content as a potential risk to this trust.

Industry Reactions and Challenges

The field of AI detection technology has long been a subject of debate. Even OpenAI abandoned its own AI detection tool due to accuracy issues. While the precision of Pangram’s detection remains unclear, Best’s admission that the tool is “imperfect” suggests that overreliance on it should be avoided.

Many writers use AI assistants to support their writing process, such as organizing ideas, suggesting structures, or editing drafts. If Substack’s tool labels such legitimate uses as “AI-generated,” false positives could become a significant issue. While the platform has implemented a mechanism to report inaccurate results, this alone may not resolve the underlying concerns.

Conversely, distributing low-quality, fully AI-generated content as paid subscriptions could undermine the overall value of the platform. As Substack operates on a subscription-based model, it relies on the premise that readers value “human-written content” enough to pay for it. If this assumption is undermined, the business model itself could face challenges.

Editorial Opinion

In the short term, this tool aims to alleviate readers’ mistrust towards AI-generated content. However, if false positives become a widespread issue, it might lead to friction with writers who use AI responsibly. Given the current uncertainty regarding the tool’s accuracy, some initial confusion is to be expected. While Substack has included a feature for reporting false positives, its effectiveness will depend on how well it is managed.

In the long run, the industry seems poised to adopt AI detection as a standard feature across platforms. Substack’s move could serve as a leading example for other platforms. However, considering the limitations of AI detection technology, combining it with transparency initiatives like the “How I make this” feature may offer a more practical solution. Instead of outright banning AI use, making it visible and empowering readers to make informed choices appears to be a forward-thinking approach. The concept of “Claudefishing” itself introduces a new angle to consider in platform design, drawing attention to the importance of balancing technology and trust.

References

• “Substack adds an AI detector to help spot blogs written by no one”, by Emma Roth — The Verge, 2026-07-21T19:22:28.000Z (ARR)
• Source URL: https://www.theverge.com/ai-artificial-intelligence/968855/substack-pangram-ai-detecting-tool