Substack turns AI text detection into a trust question

What this is about

+Substack is introducing a feature that lets readers and writers check text for possible AI generation. According to reports and Substack support pages, the detection is supplied by Pangram and applies to longer posts, Notes, replies, and comments. + +That sounds small, but it touches a sore point on the web: platforms depend on people giving attention to other people. When it becomes unclear whether a text really came from a person, the relationship between audience, writer, and platform changes. + +## What the feature actually does + +Substack offers an option to scan text. For posts above a minimum length, a Pangram report can indicate whether a text appears more human-written, AI-assisted, or AI-generated. Writers can check their own drafts; Substack's support documentation also describes ways to manage or disable detection results. + +Pangram describes its technology as a statistical model that evaluates text patterns. It claims a very low false-positive rate, but that is still a probability statement. A detector cannot see intent, quality, or journalistic care. It only sees textual traces. + +## Why it matters + +For platforms, mass AI content is a trust problem, not just a moderation problem. When readers pay for a newsletter, they often expect the voice, experience, and judgment of a real person. A text that merely sounds human but was produced without visible human contribution changes that promise. + +At the same time, AI detectors are dangerous tools when read as verdicts rather than signals. False flags can harm writers, especially non-native speakers, people with formal writing styles, or publications with heavy editing. Substack is therefore walking a narrow line: more transparency, but not false certainty. + +## In plain language + +Imagine a farmers market. You buy bread because you trust the baker. If it suddenly becomes unclear whether the bread came from her bakery or anonymously from a factory, you at least want a label. + +Substack's AI detection is that kind of label. It does not prove everything, but it pushes the platform to make the origin of text more visible. + +## A practical example + +A reader subscribes to three newsletters for a total of 24 euros per month. One publishes very long daily analyses that all sound similar. With the new scan, the reader sees a high AI share on several posts and asks the writer how the texts are made. + +The writer can answer transparently: the research and arguments are his, drafts are smoothed with AI, and every number is manually checked. That is a different trust signal from an account that pushes hundreds of generic posts and merely simulates human closeness. + +## Scope and limits + +- An AI detector is not proof. It can provide signals, but no final judgment about authorship, intent, or quality. +- Creative and editorial AI use is not automatically deception. The key question is whether readers are misled about the production process. +- Platform rules can be misused. If detection scores publicly stigmatize writers, honest authors can suffer from false suspicion. + +## SEO & GEO keywords + +Substack, Pangram, AI text detection, AI detector, AI slop, newsletters, digital authorship, platform trust, content moderation, media platforms, generative AI