Substack adds an AI detector to help spot blogs written by no one
Substack has introduced an AI-detection tool to help readers gauge whether posts, notes, comments and replies were written by artificial intelligence. Built on technology from detection firm Pangram, the feature aims to boost transparency on the platform as AI-generated content becomes more common on social media, addressing what chief executive Chris Best calls "Claudefishing" — when readers unwittingly invest attention in text with no human thought behind it.
The tool, announced in a blog post on Tuesday, is rolling out on the web and Substack's iOS app, with Android support due "soon". Readers can scan any post longer than 100 words via the "Scan for AI text" option in the three-dot menu, and writers can check their own drafts too, with the option to flag inaccurate results. Substack is also adding a "How I make this" statement so creators can explain their writing process, though Best acknowledges Pangram can only detect AI use, not the quality or human care behind a piece.
- Substack launches Pangram-powered AI detector for posts and comments
- Tool flags "Claudefishing" — AI text posing as human-written
- Rolling out on web and iOS now; Android coming soon
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Supporters welcome the tool as a service to readers and to the integrity of the platform, arguing that people who subscribe to a newsletter are implicitly trusting that a human voice and judgement lie behind the words, and that as AI-generated text becomes harder to spot, giving readers even an imperfect signal helps preserve that trust. They see it as consistent with Substack's stated mission of supporting genuine writers and note that transparency tools of this kind can deter low-effort, mass-produced content from crowding out original work, ultimately benefiting the writers who invest real time and thought in their pieces.
The case against
Sceptics worry that AI-detection tools are notoriously unreliable, prone to false positives that could unfairly brand careful human writers as fraudulent or wrongly flag legitimate uses of AI for editing, research or translation. They argue this risks creating a chilling effect and reputational harm based on a probabilistic guess rather than fact, while also questioning whether a platform built on trusting individual voices should be in the business of algorithmically policing authenticity at all, particularly when many writers use AI as one tool among many in an otherwise wholly original creative process.