Platform-based content community Substack officially launched its AI content detection feature on July 21, 2026, Pacific Time, aiming to enhance platform content transparency and help readers distinguish between content written by humans and that generated by artificial intelligence. The feature is powered by the AI detection tool Pangram and is now available on the web and iOS app, with an Android version also set to be released soon.

This update covers all scenarios of content on Substack, including articles, notes, replies, and comments. Users can simply click the “…” menu on the corresponding content and select “Scan AI Text,” and the system will display the specific percentages of human-written, AI-generated, and AI-assisted text through a pop-up window provided by Pangram.

It is important to note that this detection tool applies only to texts published after 8:30 AM on July 21 and with paragraphs longer than 100 words. On the underlying technology, Pangram uses a classifier neural network model, whose training dataset was entirely collected before 2021, when generative AI had not yet become widely popular, ensuring the objectivity of the differentiation criteria.

QQ20260722-163101.jpg

While granting readers the right to know, Substack also retains a high degree of operational autonomy for creators. Creators can self-test or add creation notes during the draft stage, and they can also disable the detection feature completely or remove the mark at any time.

Facing the current situation where the threshold for content production has been lowered by large language models and the proportion of AI-generated content in online texts continues to rise, co-founder and CEO of Substack Chris Best stated that this move aims to address the issue of "mismatch between reader expectations and actual content." This action marks a key step for content platforms in balancing the application of AI tools and maintaining a high-quality human original ecosystem.