Substack adds AI detection to all notes, posts

Substack announced today that it is launching AI detection across its platform through a partnership with Pangram, described as the leading AI-detection tool. Users can now scan notes, replies, comments, and posts to see an estimate of how much of the text was written by hand or with AI assistance.

The feature is available immediately on web and iOS, with Android support coming soon. It applies to text longer than 100 words that was published from today onward, and analysis will be shown only to users who request it. According to Substack, Pangram can detect whether AI was used to create text but cannot determine whether human care went into the process or whether AI was used as a research source. The company notes that independent research suggests Pangram detects AI-generated text with high accuracy.

Substack positioned this move as a response to what it sees as a growing problem: estimates suggest that as much as 40% of text on some social media platforms is AI-generated, making it harder for readers to distinguish real human voices from machine-generated content. The company cited "Claudefishing"—a term from culture critic Freddie deBoer—to describe the deception of leading readers to invest attention in content with no human thought behind it. Substack argues this undermines the platform's core mission: building an economic engine for culture by helping real people make money from work they believe in. "When readers have to wonder if what they're reading is real, it undermines trust in authorship and threatens the livelihood of writers," the company stated.

Creators have new tools to work with the feature. They can run Pangram scans on their drafts before publication, add a "How I make this" statement to explain their creative process and set reader expectations, and report or remove any scans of their own work they believe are inaccurate. Substack emphasized that it does not oppose AI use—the company itself uses AI for software, research, and product features—but believes readers and writers should have transparency about when and how AI was involved. Depending on user feedback, Substack is considering additional tools: the ability to set preferences around AI content in community Reply Rules, features to help creators express their individual value, reader options to set preferences about AI-assisted content in recommendations, and continued improvements to systems fighting spam, bots, and scams.

Substack's move reflects a broader tension on the internet: AI tools are becoming ubiquitous, but readers increasingly distrust feeds filled with machine-generated content. The company cites Pangram's estimate that as much as 40% of text on some platforms is AI-generated, describing this as a problem not because AI itself is bad, but because readers and writers deserve to know what they're dealing with. Substack frames the issue as "Claudefishing"—when content that appears human-made deceives readers into investing attention in work with no human thought behind it. This framing is central to Substack's business model: the platform depends on trust between readers and creators, and on the ability of individual writers to build sustainable livelihoods. When AI-generated content pollutes the commons, it makes discovering and supporting genuine human voices harder.

The company is explicit that it does not oppose AI use—Substack itself uses AI for software development, research, and features like clipping and translation. Instead, the goal is transparency: readers should be able to choose whether to engage with AI-assisted work, and creators should be able to explain their process. Future tools Substack is considering—such as reader preferences for AI content, creator tools to express their individual value, and community-level controls—suggest the company sees this as an evolving space where norms and user preferences will shape how the feature develops.