Substack Rolls Out Pangram Scanner for AI Detection
Substack ● Covered by 3 sources
Substack just added an AI-detection scanner from Pangram so you can check if a post, note, or comment was written by a human or a bot. It's a first move against "Claudefishing" — AI content pretending to be human — which Substack says is flooding social feeds.
Based on reporting by Substack — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Substack has a new tool for a very 2024-into-2025 problem: you can no longer assume the person on the other end of a post is, in fact, a person. Starting today, the platform is rolling out an integration with Pangram, an AI-detection company, that lets users scan notes, comments, replies, and posts longer than 100 words to get an estimate of how much of the text was likely written by a human versus generated or heavily assisted by AI.
The timing isn't random. Substack cites Pangram's own estimate that up to 40% of text on some social platforms is now AI-generated, and frames the issue less as an AI-quality problem and more as a trust problem. Plenty of AI-assisted writing is good, and plenty of human writing is slop, so the company isn't trying to police quality. What it's targeting is the gap between what a reader assumes they're getting and what they're actually getting — a phenomenon Substack is calling, only half-jokingly, "Claudefishing."
The mechanics are fairly narrow by design. Pangram's scan only tells you whether AI was likely involved in producing a piece of text; it says nothing about how much human judgment, editing, or original thought went into it, and nothing about whether AI was merely used as a research source rather than a ghostwriter. Substack acknowledges the tool isn't flawless but points to independent research suggesting it's highly accurate at flagging AI-generated text. Analysis only appears when someone actively requests it — there's no ambient score stamped on every post.
Alongside the scanner, Substack is giving writers a way to get ahead of the question themselves: a new "How I make this" statement where creators can spell out their process, plus the ability to run Pangram checks on their own drafts before hitting publish, and to dispute scans they think misfire on their own work. The features are live now on web and iOS, with Android support coming later.
Substack is framing this as a first step rather than a finished policy. It's floating further ideas — letting communities set AI-content preferences in their Reply Rules, giving readers control over what gets recommended to them, and building out anti-spam systems as AI-generated bot content scales up. The company is explicit that it isn't anti-AI; its own staff use AI tools daily for coding and product features. The stated goal is narrower: make sure readers know what they're getting, and let them decide, post by post, whether they're in the mood for a human's take or a chatbot's.
My take — AI-written commentary, not fact-checked reporting
Detection tools are a patch, not a fix, and Pangram will eventually get gamed the same way every AI-detector before it has — but I'd rather platforms try transparency now than wait until the whole feed is indistinguishable slop, which is exactly the LinkedIn-ification Substack says it's trying to dodge. The real tell here is that Substack isn't banning AI writing at all, it's just refusing to let it pretend to be something it's not, which feels like the only sane middle ground between AI-maximalism and AI-panic.
Read more about this at: Substack