Substack has integrated an AI detection tool from Pangram that lets readers scan posts, notes, replies, and comments for machine-generated text, the company said in a blog post on July 21, 2026. The feature is live on the web and the iOS app, with an Android version to follow. Any piece of content longer than 100 words can be checked by opening the three-dot menu and selecting 'Scan for AI text.'
The scan returns an estimate of how much of a given piece was generated or assisted by AI. Substack is layering it on across the surface area of the platform — long-form posts, short-form Notes, and the comment threads underneath — rather than restricting detection to headline articles. Writers can also run Pangram against their own drafts before publishing and report results they believe are wrong.
Substack co-founder and CEO Chris Best framed the change as a transparency measure rather than a ban. He coined a new term, 'Claudefishing,' for the practice of publishing AI-generated writing where readers expect a human on the other end.
Key facts
- 01Substack integrated Pangram on July 21, 2026 to detect AI-generated text in posts, notes, replies, and comments.
- 02The scanner works on any content longer than 100 words via the three-dot menu 'Scan for AI text' option.
- 03The rollout covers web and iOS at launch, with Android arriving soon after.
- 04CEO Chris Best coined 'Claudefishing' to describe posting AI text where readers expect human writing.
- 05Writers can also scan their own drafts and flag inaccurate results back to Pangram.
Alongside the detector, Substack is rolling out a 'How I make this' statement that lets writers describe their own process — whether they draft in longhand, use Claude or another model as a research assistant, or write with no AI in the loop at all. The pairing is deliberate: the detector tells readers what a machine sees, and the disclosure lets writers tell readers what they actually did.
The design implicitly accepts that AI assistance is now a spectrum, not a binary. A researcher who uses a model to pull sources and then writes every sentence themselves will look different to Pangram than someone who pastes a prompt output straight into the editor, but both may trigger some AI signal. Substack is not automatically penalizing either.
Best acknowledged the limits of the tool directly in the announcement.
“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.”— Chris Best, Substack co-founder and CEO
That caveat matters because false positives on AI detectors are a well-documented problem across the category, and Substack's business runs on paid subscriptions to individual writers. A detector that wrongly flagged a working journalist as an AI-farm operator would cost the platform trust from exactly the constituency it depends on. The self-scan and report-inaccurate-results paths are Substack's hedge against that failure mode.
Substack is not the first platform to grapple with this. Reddit, LinkedIn, and Amazon have all wrestled with waves of AI-generated posts, reviews, and books, and most have chosen to enforce quietly at the moderation layer rather than expose a detector to end users. Substack's decision to put the scan button in the reader's hand is a different bet — one that treats authorship as a signal readers should evaluate themselves rather than a binary the platform enforces.
The 'Claudefishing' framing is also a marketing move. Substack has spent the past two years positioning itself as the home of writing that stands behind a name, in contrast to algorithmic feeds where authorship blurs into the stream. A branded epithet for AI-generated impersonation reinforces that identity, and gives paying subscribers a shorthand for what they are paying to avoid.
The open question is enforcement. Substack has not said whether repeat 'Claudefishing' will affect a publication's monetization, discovery, or standing in Notes, and Pangram scores are estimates rather than proof. Best's post suggests the platform's initial approach is disclosure and reader choice, not takedowns.
For Substack, the Pangram integration is a low-cost way to sharpen its differentiation at exactly the moment when AI-generated newsletters are becoming trivial to spin up. The economic thesis of the platform — that readers will pay real money for a specific human voice — depends on those readers being able to tell that a human is actually on the other end. Making that check a two-tap action, rather than a guess, is a defensible product answer to a problem that is only going to get bigger.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
