Substack Launches New AI Detection Tool to Identify Machine-Generated Content
The Pangram-powered feature reveals how much of the content readers consume is written by AI.
1Substack's New AI Detection Tool
Substack has introduced a new tool in partnership with Pangram to detect AI-generated content. This feature aims to help readers understand the authenticity of what they're reading. It scans posts, notes, and comments longer than 100 words, providing transparency about content origins.
2How It Works
The tool displays the percentage of AI involvement in each piece of content. Users can scan drafts before publishing to check for AI influence. Currently available on web and iOS, with Android support coming soon. CEO Chris Best states this combats 'Claudefishing'—unaware consumption of AI content.
3Testing the Tool
In tests, a fully human-written article was correctly labeled. When AI paragraphs were added, it detected 10% AI contribution accurately. However, minor AI-assisted content was mistakenly flagged as fully human-written, showing room for improvement.
4False Positives and Concerns
Substack acknowledges the tool isn't perfect but cites Pangram's 97.5% accuracy rate. Writers worry about false AI-user labeling. The platform offers an appeal system for incorrect detections, allowing creators to maintain their authentic reputation.
5Transparency and User Choice
Publishers can now add disclosure statements about AI usage in their work. Best notes that while research-stage AI use isn't detectable, readers deserve to know about content creation methods. This initiative aims to rebuild trust in digital authorship.
6AI's Growing Content Role
Estimates suggest AI generates 40% of social media text, reaching 66% on LinkedIn. Half of consumers prefer brands avoiding AI in marketing. Substack's move reflects broader industry efforts to preserve content authenticity amid rising synthetic media.
Original article: CNET — Substack, AI ile Yazılan İçerikleri Tespit Etmek İçin Yeni Bir Araç Sundu