Substack adds an AI detector to help spot blogs written by no one
Substack integrates Pangram’s AI detection tool to estimate the proportion of AI-generated text in posts, notes, replies, and comments. The feature aims to restore reader trust by addressing "Claudefishing," where audiences unknowingly invest attention in content lacking genuine human thought. Creators can now use a "How I make this" statement and scan their own drafts for transparency, though they can report inaccurate results. Detection covers text over 100 words via a new menu option, with pl
Analysis
TL;DR
- Substack integrates Pangram’s AI detection tool to estimate the proportion of AI-generated text in posts, notes, replies, and comments.
- The feature aims to restore reader trust by addressing "Claudefishing," where audiences unknowingly invest attention in content lacking genuine human thought.
- Creators can now use a "How I make this" statement and scan their own drafts for transparency, though they can report inaccurate results.
- Detection covers text over 100 words via a new menu option, with plans to expand to Android shortly.
Why It Matters
This development highlights the growing industry pressure to maintain authenticity and trust in digital publishing as AI-generated content becomes indistinguishable from human writing. For AI practitioners and platform developers, it underscores the critical need for transparent disclosure mechanisms rather than just binary detection, balancing creator freedom with reader expectations.
Technical Details
- Detection Engine: Powered by Pangram, an AI detection company, which analyzes text to provide an estimate of AI involvement.
- Scope: Scans posts, notes, replies, and comments exceeding 100 words.
- User Interface: Accessed via a "Scan for AI text" option in the three-dot menu on web and iOS apps.
- Creator Tools: Allows writers to scan their own drafts and add a "How I make this" statement to clarify their process.
- Limitations: Explicitly states it detects AI usage but cannot assess the level of human care or distinguish between AI as a source versus a drafting tool.
Industry Insight
- Platforms must evolve beyond simple content moderation to include transparency features that help users make informed decisions about their attention and trust.
- The rise of "disclosure-first" models suggests that future AI integration in publishing will likely focus on labeling and provenance rather than outright bans, mitigating the risk of a "race to the bottom."
- Developers should consider implementing similar feedback loops where users can challenge or report detection inaccuracies to improve model reliability and user acceptance.
Disclaimer: The above content is generated by AI and is for reference only.