Substack partners with Pangram to flag AI-written posts and comments

What happened

Substack is rolling out a new AI-detection tool that lets readers estimate how much of a given piece of content may have been written or assisted by AI. The feature, powered by AI-detection startup Pangram, can scan posts, notes, replies, and comments across the platform.

According to a blog post published on Tuesday, readers can trigger the scan by selecting the "Scan for AI text" option from the three-dot menu in the top-right corner of a post. The tool works on content longer than 100 words and returns an estimate of how much text appears AI-generated or AI-assisted.

The integration is live now on the web and in Substack's iOS app, with an Android rollout described as coming "soon." The Verge reports that the tool can also pick up on a pattern the platform calls "Claudefishing" — a term suggesting AI-generated content masquerading as human writing across posts, comments, notes, and replies.

Substack co-founder and CEO Chris Best framed the move as a transparency initiative rather than an anti-AI crackdown. "The core problem is not people using AI, or the quality of its output," Best wrote. "The problem is when there is a mismatch between a reader's expectation and reality."

Why it matters

Substack has built its brand on direct relationships between writers and paying subscribers, where trust and authenticity are the core currency. As AI-generated content floods social platforms and search results, readers increasingly want to know whether a post reflects a human's genuine voice or a model's output. This tool gives Substack's audience a lightweight, in-platform way to check.

The partnership with Pangram also signals a broader industry trend: platforms are beginning to treat AI provenance as a reader-side feature rather than a writer-side restriction. Instead of banning AI-assisted content outright, Substack is betting that disclosure-by-detection — letting readers make their own judgments — is the more sustainable approach.

However, AI-detection tools remain imperfect. False positives can unfairly flag human writers, particularly non-native English speakers, while sophisticated AI outputs can still slip past detectors. Substack's framing as an "estimate" rather than a verdict is notable, but the tool's accuracy at scale will determine whether it builds trust or creates new friction.

What to watch

• Accuracy in the wild: How often does Pangram's detector produce false positives or miss AI-assisted content on real Substack posts? Reader feedback in the coming weeks will be telling.
• Writer response: Will Substack writers view the tool as a helpful transparency layer or an accusatory surveillance mechanism? Some may push back if their human-written posts get flagged.
• Android rollout timeline: Substack only says Android support is coming "soon" — a concrete date has not been provided.
• Platform precedent: If Substack's approach gains traction, expect other content platforms (Medium, Ghost, WordPress) to explore similar reader-facing detection features.
• Regulatory context: As AI-labeling requirements take shape in the EU and elsewhere, voluntary platform-level tools like this could become a baseline expectation.