Substack deploys AI detection tool to curb 'Claudefishing' and restore trust
Co-founder Chris Best argues that transparency around artificial intelligence usage is essential to protect reader attention and writer livelihoods on the platform.
Substack has introduced a new feature designed to identify content generated or assisted by artificial intelligence, marking a significant shift in how the newsletter platform manages content integrity. Developed in partnership with AI detection firm Pangram, the tool aims to increase transparency by allowing users to determine the proportion of AI-generated text within posts, comments, notes, and replies.
The feature is now available on the web and iOS app, with an Android version scheduled for release in the near future. To use the detection capability, readers must select the "Scan for AI text" option from the three-dot menu located in the top-right corner of a post. The tool processes text exceeding 100 words to estimate the likelihood of AI involvement, providing a metric that was previously absent from the user experience.
Substack co-founder and CEO Chris Best described the initiative as a response to "Claudefishing," a term he coined to describe the mismatch between reader expectations and reality when content lacks human thought. Best argued that the core issue is not the use of AI itself, but rather the situation where readers unwittingly invest their attention in material with no human authorship, thereby undermining trust in authorship.
Alongside the detection tool, Substack has introduced a "How I make this" statement, enabling creators to explicitly disclose their writing processes and the extent of their use of AI tools. Writers can also scan their own drafts using the Pangram integration and have the option to report inaccurate results, although the source material does not specify the tool's accuracy rates or false-positive metrics.
Best emphasised that the goal is to empower readers to make informed decisions about where to invest their time. He warned that platforms rewarding unattributed fakeness risk creating a "race to the bottom" that threatens the livelihoods of writers, including those who use AI tools thoughtfully to support their creative work.
