Substack Adds Pangram AI Detection So Readers Can See How Much of a Newsletter Was Written by a Human
Readers can now scan any Substack post, note, or comment for an AI estimate. Writers can opt out - or add a disclosure explaining how they used AI.
Substack launched an AI detection feature on Wednesday built on Pangram, a third-party AI writing detector, giving readers a way to estimate how much of any newsletter, note, reply, or comment was written by a person versus generated by AI. Substack Pangram scans run on any text above 100 characters published on or after July 21, 2026. Web and iOS support is live now; Android follows later.
How the Detection Works
Readers trigger a Pangram scan directly from a post or comment. Pangram returns a percentage breakdown - how much of the text reads as human-written versus AI-generated. Substack frames this as an estimate, not a verdict. Writers can run the same scan on their own drafts before publishing, which lets them catch a false positive and contest it before readers do.
Creators who want to dispute a scan result on their own published work can flag it as incorrect. Substack will review and, if warranted, remove the scan. That appeals process is a small but real concession to the reality that AI detection tools produce false positives - a fact Substack acknowledged by building the removal option in from day one.
Writers Can Opt Out or Disclose
Substack is not forcing detection on anyone's work permanently. Creators can disable Pangram scans on their content entirely. Alternatively, writers who do use AI can add an "AI author's note" - a short disclosure explaining how they used AI in their process. Substack CEO Chris Best framed this as preferring transparency over prohibition.
We'll do everything for you except the hard part. You have to have something - an idea that's worth reading, that's worth caring about, that's worth sharing. That one thing is very hard and very valuable.
Chris Best, CEO, Substack
Best made those remarks in a conversation with Pangram founder Max Spero, positioning AI as a production aid rather than a replacement for the ideas themselves. Whether Substack's writers - many of whom built paid subscriber bases on their personal voice - agree with that framing is another question.
A Short-Term Risk Substack Chose to Take
Adding AI detection to a platform that hosts independent newsletters is a bet with real downside. Pangram scans could expose popular publications that rely heavily on AI, potentially eroding reader trust in creators who built audiences around the perception of personal authorship. Substack appears to have weighed that short-term risk against the alternative - a platform increasingly full of AI-generated content with no way for readers to tell.
Platforms have been moving in this direction for a while. YouTube labels AI-generated videos, TikTok auto-labels AI content from other platforms, and Spotify updated its AI policy last year to address AI-generated music. AI accounted for 31% of all job cuts in June 2026 - a number that includes writers and content creators - which makes the question of who actually wrote a newsletter less abstract than it was two years ago. For Substack readers paying $8 to $15 a month to support a writer, that question is now answerable.