Substack adds an AI detector to help spot blogs written by no one
Substack is integrating Pangram's AI detection tool across posts, notes, replies, and comments to flag potentially AI-generated text. The platform is also adding an optional 'How I make this' disclosure statement for writers.
Intelligence analysis by Llama
Substack is rolling out Pangram-powered AI detection across its platform, letting readers scan posts over 100 words for likely AI authorship. CEO Chris Best frames the move as fighting 'Claudefishing' — the trust gap when readers unknowingly engage with machine-written content.
Imagine you opened a letter that looked like it was from a friend, but it turned out a robot wrote it. Substack is now adding a tool that checks if a blog post was written by a computer instead of a person, so readers know what they are really looking at.
Analysis
The 'Claudefishing' Problem
Substack CEO Chris Best is putting a name to a phenomenon that has quietly eroded confidence in online writing: 'Claudefishing,' a riff on 'catfishing' for the AI era. The term captures the specific harm Best is targeting, not people using AI to draft or polish their work, but the gap between what a reader assumes is on the other end of a post and the reality. According to Best, the issue arises when a reader 'unwittingly invest[s] their attention in something with no human thought on the other end.' That framing is doing a lot of work here. It draws a line between AI as a legitimate writing tool and AI as a deception engine, a distinction Substack clearly wants to police on the reader's behalf.
How Pangram Fits In
The detection itself is outsourced rather than built in-house. Pangram, an AI detection company, is providing the scoring engine that estimates how much of a given piece of text may have been AI-generated or written with AI assistance. Readers can invoke the scan through a 'Scan for AI text' option in the three-dot menu on any post longer than 100 words, and the tool covers not just main posts but notes, replies, and comments. That breadth matters: much of the AI content on social platforms now lives in low-effort reply sections, not just in long-form posts. Best is careful to caveat the limits, noting that '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.' Writers can also report inaccurate results, an acknowledgment that AI detection remains an imperfect science and false positives are a real risk.
Disclosure as a Parallel Track
Alongside detection, Substack is introducing an optional 'How I make this' statement that creators can use to explain their writing process. The pairing is deliberate. Detection handles the suspicion after the fact, while disclosure gives writers a proactive way to establish trust. It also creates a softer landing for authors who do use AI tools but want readers to know the role those tools played. Whether the statement catches on will depend on how prominently Substack surfaces it and whether readers begin to expect it. If 'How I make this' becomes as routine as a byline, the AI provenance question on the platform may shift from a forensic exercise to a simple disclosure habit, and the detector itself becomes a backstop rather than the primary signal.
Key points
- Substack is integrating Pangram's AI detection tool across posts, notes, replies, and comments, with support for content over 100 words.
- The feature is live on web and iOS, with an Android launch described as 'soon.'
- CEO Chris Best is framing the initiative around 'Claudefishing,' a term he uses for the trust gap when readers unknowingly engage with AI-generated text.
- Writers can scan their own drafts with Pangram and report inaccurate results.
- Substack is also adding an optional 'How I make this' statement so creators can disclose their writing process up front.
If the Pangram integration gains traction, it could nudge other publishing platforms to treat AI provenance as a standard reader-facing feature rather than a niche concern. The parallel 'How I make this' statement gives thoughtful AI users a way to stay transparent without being penalized, potentially preserving trust in writers who use AI as a tool rather than a substitute.
AI detection remains unreliable, and Pangram's results could produce false positives that unfairly flag human-written work, especially from writers whose style happens to resemble AI output. The 'Claudefishing' framing also risks stigmatizing any AI use at all, pushing writers who rely on assistive tools underground rather than toward honest disclosure.
