Substack just gave readers a way to check whether the post in front of them was written by a person or a machine. The company is rolling out an AI detection tool that scans posts, notes, replies, and comments, then estimates how much of the text may be AI-generated or written with AI help, according to The Verge AI. The feature is powered by an outside detection company called Pangram, and it’s live on the web and iOS now, with Android coming soon.
This is Substack planting a flag on a question every reader is starting to ask: is there a human on the other end of this?
How it works
The mechanics are simple, as The Verge AI reports:
- Open any post longer than 100 words.
- Tap the three-dot menu in the top-right corner.
- Choose “Scan for AI text.”
- Substack returns an estimate of how much of the writing looks AI-generated or AI-assisted.
The tool works on more than just full articles. It covers notes, replies, and comments too, so the check extends across the social side of the platform, not only long-form writing.
Writers get their own controls
Substack isn’t only handing this to readers. Creators can scan their own drafts with Pangram before publishing, and they can report results they think are wrong. There’s also a new “How I make this” statement, which lets writers explain their process up front instead of leaving readers to guess.
That two-sided design matters. It frames the tool as a transparency layer rather than a gotcha machine pointed only at authors.
The reasoning behind it
Substack co-founder and CEO Chris Best laid out the thinking in a blog post. His argument, quoted by The Verge AI, is that AI use itself isn’t the villain. “The core problem is not people using AI, or the quality of its output,” Best writes. “The problem is when there is a mismatch between a reader’s expectation and reality, especially when they unwittingly invest their attention in something with no human thought on the other end. That’s Claudefishing.”
Best also tied the move to writers’ livelihoods. “When readers have to wonder if what they’re reading is real, it undermines trust in authorship and threatens the livelihood of writers, including those who use AI tools thoughtfully to produce work they believe in,” he writes. “Platforms that reward fakeness will create a race to the bottom.”
What stands out here is the positioning. Substack isn’t banning AI. It’s betting that trust, not prohibition, is the thing worth protecting.
The limits worth knowing
Best is upfront that detection has hard edges. Pangram “can only detect whether AI was used to make the text, not whether great human care went into creating it, nor whether AI tools were used as a source,” he notes.
That’s an important caveat. A high AI score doesn’t mean a piece is low-effort or dishonest, and a low score doesn’t guarantee quality. AI detectors across the industry have also struggled with false positives, which is likely why Substack built in a report button for inaccurate results. The estimate is a signal, not a verdict.
Why it matters
Substack is one of the biggest homes for independent writers, and it’s now treating AI disclosure as a platform feature rather than a policy buried in the terms of service. That’s a meaningful shift. As AI-written content floods social feeds and inboxes, the platforms that survive may be the ones that help readers spend their attention with confidence.
Expect more platforms to face the same pressure. If readers start expecting a “was this human?” check the way they expect spam filters, tools like this stop being a novelty and start becoming table stakes. For now, Substack is early, and the Android rollout plus how accurate Pangram proves in the wild will tell us whether the bet holds.
Full details are available at the original report from The Verge AI.