Substack's New AI Detection Tool Tells Readers Who Wrote It

What happened

Axios reported that Substack partnered with AI-detection company Pangram to let readers scan posts, Notes, replies, and comments for an estimate of human-versus-AI authorship. The scan works through a three-dot menu option on eligible content, covers any text over 100 words, and applies to anything published on or after July 21, 2026, live now on web and iOS with Android support still coming. Substack co-founder and CEO Chris Best described the target problem as Claudefishing, the gap between a reader assuming a human wrote something and the reality that no one actually did, and said the explicit goal is avoiding the fate of platforms already flooded with unlabeled AI content.

Why Substack's AI detection is a distribution signal

Substack is betting that authenticity becomes a real subscription driver as AI content volume keeps rising everywhere else online, which makes this a positioning move as much as a product feature. For B2B operators building a newsletter as a distribution channel, it raises the bar on disclosure and process, not just on output quality alone. A detector that flags heavily AI-assisted writing without any context for how it was actually produced is a fairly blunt instrument, and Substack admits as much: it can't see whether a human directed the argument, dictated the draft aloud, or fact-checked an AI-assisted first pass carefully.

1. 01
   Decide your AI-disclosure policy before a reader scans

   If you use AI tools anywhere in your newsletter workflow, at any stage, write a one-line disclosure policy now rather than reacting after a subscriber flags a post publicly. Transparency about process reads far better than getting caught by a detector later. Put the policy where a reader can actually find it, a footer link or an about-page note, not a policy nobody outside the company ever sees.
2. 02
   Keep primary research and quotes in your own voice

   Pangram's detector can't tell whether an AI merely summarized someone else's research or drafted your original argument from scratch. Protect the parts of your writing that are genuinely yours, original data, interviews, specific opinions, since those differentiate you regardless of what any scanner says.
3. 03
   Watch whether other platforms follow Substack's move

   If LinkedIn or Beehiiv add similar detection tools, disclosure expectations will shift fast across every channel at once, not just newsletters. Build your process now assuming this becomes a standard requirement, not a Substack-only quirk. Set a calendar reminder to check competing platforms for similar detection features quarterly, rather than reacting individually to each one as it launches.

By the numbers: the AI detection scan only evaluates text over 100 words and only applies to content published on or after July 21, 2026, so nothing already in a newsletter's back catalog gets retroactively flagged or scored.

What to do this week

If you publish a newsletter on Substack, open the three-dot menu on your last post and run the AI-text scan on yourself before a reader does it for you. Use the result to calibrate how much you disclose about your writing process, especially if you draft with AI assistance and edit heavily afterward.