Substack readers can now scan any post, note, reply, or comment longer than 100 characters to estimate how much of the text was generated by AI, the company said. The feature, built with detection startup Pangram, arrives alongside a voluntary AI author's note that lets writers disclose their use of AI tools before publishing. For writers, the move turns transparency from a tacit expectation into an on-platform signal that readers can check.
How the scanner and author note work
Any Substack user can run a scan on eligible text. The tool uses Pangram's model, which the company says detects AI-written prose with high accuracy. Writers can also scan their own drafts before publishing and, if the results look off, add a note explaining the tool's use or the text's provenance.
Results are not always perfect. Quirks in a writer's style-especially when English is not a first language-can produce false positives. Substack encourages writers to address such cases in the AI author's note rather than leaving readers to guess.
Why Substack is doing this now
Substack CEO Chris Best said in a conversation with Pangram founder Max Spero that the feature reflects the platform's original pitch: Substack handles everything else so the writer can focus on "a good idea, well said." Best called it "an appropriate use of AI," framing the tool as a way to support writers, not replace them.
The move also comes as skepticism about AI-generated content grows across platforms. By pairing detection with a disclosure tool, Substack signals that using AI to assist with writing is acceptable-provided the writer is upfront about it.
A tool that could cut both ways
Once readers begin scanning posts, some publications may show high AI scores, which could erode trust and drive subscribers away. That risk extends to the platform itself: a newsletter that relies heavily on undisclosed AI loses credibility, and the association can reflect on Substack's brand.
The scanner's occasional inaccuracies add another layer of risk. A false positive on a human-written essay can damage a writer's reputation if readers take the result at face value. That's why Substack treats the author's note as a safety valve, giving writers a chance to respond before doubts harden.
An industry-wide shift toward AI transparency
Substack's approach mirrors what several large platforms are trying, though detection tools remain imperfect and unevenly deployed:
- Video-sharing sites now auto-tag AI-generated videos.
- Photo and video apps offer tools that distinguish human-made from AI-generated media.
- Music streaming services flag AI-made tracks, limit their promotion, and in some cases cut off monetization when disclosure is absent.
None of these systems work flawlessly, but they represent a clear commitment to making the origin of AI content visible to audiences.
What this means for writers and readers
For writers, the practical change is immediate: any reader can now check a post's AI fingerprint. That makes honesty about AI use a pragmatic choice, not just an ethical one. Resources like AI for Writers can help professionals understand where these tools fit into their workflow and how to disclose their role without undermining their voice.
Readers gain a new piece of context, but the scan does not replace editorial judgment. A high AI score does not automatically mean a post is low quality; a zero score does not guarantee it is worth reading. The tool simply pulls back a curtain that was previously closed.
Why this matters for writers
Writers on Substack should assume that AI scans will become part of a reader's routine vetting process. Including an AI author's note and responding to questionable scan results quickly can protect a reputation that years of human effort built. The platform is betting that candid disclosure will be rewarded with trust-and that silence will start to look like a liability.
Your membership also unlocks:
