Substack adds AI detector to flag machine-written posts

Substack has introduced a new AI detection feature designed to increase transparency on its platform as AI-generated and AI-assisted content becomes more common. The tool, powered by the AI detection company Pangram, scans posts, notes, replies, and comments to provide an estimate of how much text could be AI-generated or written with AI assistance. Readers can access this feature by selecting the "Scan for AI text" option from the three-dot menu in the top-right corner of any post, and the tool works on content longer than 100 words. The feature is rolling out across the web and the iOS app, with an Android launch planned for the near future.

Substack co-founder and CEO Chris Best explained the reasoning behind the integration in a Tuesday blog post. He argued that the core problem is not people using AI or the quality of AI output, but rather the mismatch between a reader's expectation and reality, particularly when readers unknowingly invest their attention in content with "no human thought on the other end." Best termed this phenomenon "Claudefishing." He wrote: "When readers have to wonder if what they're reading is real, it undermines trust in authorship and threatens the livelihood of writers — including those who use AI tools thoughtfully to produce work they believe in." To complement the detection tool, Substack is also launching a new "How I make this" statement that allows creators to voluntarily explain their writing process to readers.

Best acknowledged important limitations of the Pangram integration. The detector can only identify whether AI was used to create text, not whether great human care went into the content or whether AI was used as a research source. Writers have the ability to scan their own drafts with Pangram and can report results they believe are inaccurate. Best's underlying message was a plea for platform integrity: "Platforms that reward fakeness will create a race to the bottom."

Substack's move reflects a growing tension on social platforms: the prevalence of AI-generated content and the reader's difficulty distinguishing it from human work. Chris Best frames this not as a problem with AI use itself, but as a mismatch between reader expectations and reality. Best invokes what he calls "Claudefishing" — investing attention in content with no human thought behind it — as the core issue undermining trust in authorship. The platform is also introducing a parallel transparency measure: a "How I make this" statement that lets creators voluntarily explain their writing process. Together, these tools assume that readers and writers value knowing the provenance of content, and that flagging AI involvement helps preserve both reader trust and writer livelihoods. However, Best also notes a meaningful limitation: detection is binary (AI used or not) and cannot assess the quality of human curation or the thoughtful use of AI as a research aid.