The new flex on Substack may not be a bigger list, a shinier badge, or another Notes dunk. It may be a reader looking at a paywalled essay and thinking: a person actually wrote this. That is the weird little turn in Substack’s Pangram move. AI detection is being packaged less like a moderation hammer and more like a receipt for human-made work. ## What Substack Is Actually Shipping According to Simon Hernandez-Arthur at Axios, Substack is partnering with Pangram to let users estimate how much platform text, including posts and comments, was made with AI. Substack CEO Chris Best framed the move in Against Claudefishing as a trust problem, writing that it is getting harder to tell what is real on the internet. The Verge’s Emma Roth reported that the tool can scan posts, notes, replies, and comments, which means this is not confined to newsletter essays. It touches the whole Substack social layer, aka the part where creators cultivate subscribers before the paywall gets involved. On Substack later said the tools are available on web, iOS, and Android, and that publishers can run Pangram on drafts before publication, disable scanning on a post by post basis, and add a “How I make this” statement. The same post says these features do not use publisher content, through Substack or Pangram, to train generative AI models. Translation: Substack is trying to make disclosure feel like publishing infrastructure, not a scarlet letter. The platform gets to say it supports AI use while nudging creators to explain the workflow before readers start doing amateur forensics in the comments. ## Why Human-Made Is Becoming a Paid Signal Axios is doing the monetization math out loud here. Hernandez-Arthur reports that Substack is betting subscribers, and the sponsors those subscribers attract, value human-created work and are willing to pay for it. That is the actual platform update hiding inside the product announcement. The scan is not just about labeling synthetic text; it gives creators a new way to explain why a subscription is worth renewing and gives sponsors a cleaner story about the publication they are adjacent to. The counterintuitive bit is that detection could make AI-assisted writing more normal, not less, because the product prompt becomes explain your process. On Substack said the company’s view is that AI use is not necessarily the problem, while a lack of transparency is. That framing matters because creators are already navigating a messy middle: drafts, summaries, edits, research prompts, and fully generated posts all get flattened into one giant discourse blob. Substack is trying to separate the workflow from the deception, which is a useful distinction if the product does not overclaim what detection can know. ## The Creator Risk Is Reputation, Not Just Policy AI Weekly, summarizing 404 Media reporting, captured the writer anxiety: percentage scores can feel less like context and more like public grading. It cited Backstage Pass writer Mack Collier saying he is “not going to apologize for using AI in the creation process,” while Alice Lemee warned that “these detectors are notoriously, wildly inaccurate,” and that a false accusation can harm a writer’s reputation. That is the drama without the costume change: creators want credit for labor, readers want fewer mystery meat posts, and nobody wants a machine-generated shrug deciding who looks honest. The practical answer is to treat Pangram as a signal, not a courtroom. On Substack’s own wording says readers can see how much text is estimated to be written by hand or with AI assistance, and that word estimated is doing a lot of load-bearing. Platforms have a long tradition of launching trust tools, then letting edge cases become creator support tickets. Put this in the running tally of platform promises we should verify in the wild, not just applaud at launch. ## What Creators Should Do Next Substack’s Model Behavior series says it wants to ask writers, creators, and thinkers about prompts, tool stacks, and analog workflows, which is basically the platform telling creators where the new social norm is headed. If you publish paid work, write a short process note before readers ask for one. Say whether AI helps with drafting, editing, research, or not at all, then keep it boring and consistent. The goal is not to perform purity; it is to reduce uncertainty at the exact moment someone is deciding whether to pay. For platform builders, the lesson is bigger than Substack. Trust is becoming a product surface, not just a policy page, and users will judge it by controls, context, and error handling. Watch next for whether sponsors actually ask for human-made inventory, whether readers use scans sparingly, and whether Substack keeps creator controls intact. For now, Substack has turned human-made from a vibe into a product feature, and every writer selling access should decide what they want that feature to say. ## Sources - Substack bets readers want to pay for content written by humans with new AI detection tool - Axios

Sources