Substack deploys AI detector to identify content generated by chatbots
As AI-generated content increasingly permeates the internet, Substack aims to equip readers with a tool for transparency. The newsletter platform has announced a partnership with Pangram to incorporate a system that can estimate whether a piece of text was created by a human or a generative AI.
Introduced by Substack CEO Chris Best, this initiative is part of a larger effort to combat what he refers to as “Claudefishing,” a practice where content produced primarily by AI models is published under the guise of being entirely human-written.
A Button to Analyze Text Origins
This new feature allows readers to analyze publications, notes, and responses posted on Substack. It relies on technology developed by Pangram, a company specialized in detecting AI-generated texts.
The analysis is available for all content exceeding 100 words, published since July 21. The results of the analysis remain strictly private: only the user initiating the scan can view the evaluation.
According to Chris Best, while no AI detector is foolproof, Pangram has demonstrated a high level of accuracy in numerous independent evaluations.
The feature is already accessible via the web and on iOS, with an Android rollout planned for the future.
New Tools for Authors
Substack isn’t just offering these updates to readers. Content creators can now add a section titled “How I Made This” to their publications, where they detail their working method and any role AI may have played in their text creation.
Before publication, authors can also use Pangram to analyze their drafts and anticipate the results readers will see.
In cases of perceived inaccuracies in detection, they have the option to report the result and request its removal.
Substack also mentions several upcoming developments, including preferences related to AI usage in response rules for publications, as well as filters allowing readers to shape content recommendations based on their acceptance levels of AI-generated texts.
A Response to the Surge of AI-Generated Content
According to estimates cited by Substack and Pangram, some social networks may now contain as much as 40% of content that is entirely AI-generated. This trend raises concerns about content quality, authenticity in interactions, and the increasing difficulty in distinguishing human-produced works from automated texts.
Thus, Substack’s goal is not to prohibit the use of AI but to foster greater transparency, allowing each reader to form their own opinion.
A Trend Affecting All Platforms
Substack is not the only player addressing this issue. In recent months, LinkedIn has begun limiting the visibility of certain content deemed to be mass-produced by AI. Meanwhile, Meta is also developing its own detection system, though it has yet to be officially launched.
These initiatives reflect a shared concern: to maintain user trust in a digital environment where generative models produce increasingly convincing texts.
Transparency as a New Strategic Focus
The introduction of these tools signifies a significant shift in how platforms address artificial intelligence. Following a rapid adoption of generative technologies, the challenge is now not only to create content more easily but also to clearly inform users about its origins. This pursuit of transparency could gradually become a distinguishing criterion among publishing platforms.
However, an essential question remains: will AI detectors keep pace with the models they aim to identify? As large language models become more sophisticated, the line between human writing and automated generation continues to blur. In this context, detection tools must evolve as rapidly as the technologies they are tasked with monitoring.


