Substack rolls out AI detection tool powered by Pangram to flag machine-generated posts

Substack has integrated an AI detection system from Pangram that allows readers to identify potentially machine-generated content on the platform. According to The Verge, the tool scans posts, notes, replies, and comments to estimate what percentage of text may be AI-authored or AI-assisted, launching across web and iOS with Android support planned.

How the detection feature works

Readers access the detector by selecting “Scan for AI text” from the three-dot menu on posts exceeding 100 words. The feature is rolling out to all Substack users as part of a broader initiative to surface authorship transparency. Writers can also pre-scan their own drafts using Pangram before publishing, with an option to report inaccurate assessments.

Substack’s framing: transparency over bans

Substack CEO Chris Best emphasizes that the platform’s concern is not AI usage itself but undisclosed use. According to The Verge, Best describes the core issue as “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.” He frames this phenomenon as “Claudefishing”—a parallel to clickbait that exploits reader attention.

Alongside the detection rollout, Substack is introducing a “How I make this” statement that lets creators explain their writing process and AI tool usage to readers. This positions transparency as a collaborative solution rather than a restriction on AI use by writers.

Limitations and calibration

Best acknowledges the tool’s constraints: Pangram can detect AI involvement but cannot measure writing quality or identify cases where AI tools inform research without generating the final text. According to The Verge, Best warns that “platforms that reward fakeness will create a race to the bottom,” suggesting that AI detection is a baseline measure to prevent deceptive publishing rather than a comprehensive quality guarantee.

Why This Matters

The Pangram integration signals that major writing platforms are adopting detection as a default feature rather than leaving AI disclosure to individual creator honesty. For readers, this shifts the burden of verification from personal skepticism to algorithmic flagging—useful when scanning unfamiliar authors but incomplete without context. For writers, it creates an expectation of transparency that may reshape publication norms on Substack, particularly for creators who use AI assistively. The real test is whether detection accuracy holds up at scale and whether readers actually use the tool to make reading decisions, or whether it becomes a silent metadata field that few notice.