• 2 min read
Substack rolls out AI text scans for posts and comments
Substack now lets readers check whether text appears human, AI-assisted, or AI-generated using Pangram. The feature is live on web and iOS.
Image: Mashable
Substack has launched a new AI detection tool aimed at helping readers tell whether text was written by a person or generated with AI. Announced in a blog post on Tuesday, the feature is powered by Pangram and works across Substack posts, notes, replies, and comments.
“Not everything made with AI is slop, and not all slop is made with AI. The problem is when there is a mismatch between a reader’s expectation and reality, especially when they unwittingly invest their attention in something with no human thought on the other end.”
The tool is available now on the web and in the iOS app, with Android coming later. To use it, readers open the “…” menu and choose “Scan for AI text.” A popup then shows what percentage of the passage Pangram classifies as AI-generated, AI-assisted, or human.
There are limits. Substack says the feature only works on passages longer than 100 words, and only for content published after 8:30 a.m. PT / 11:30 a.m. on July 21. Older posts are marked ineligible, while shorter passages trigger a notice saying there is not enough text for accurate detection. On Best’s own posts, the popup also includes an extra line: “I’m still figuring all of this out. (In the meantime these words are mine, for better or worse.)”
Recommended reading
Threads adds parental controls for teens next week
What authors can do with Pangram on Substack
Pangram only scans text when a reader explicitly requests it. Best also said the system cannot detect cases where a human used AI for research before writing.
Substack is adding a few controls for writers too:
- Authors can add a statement explaining how a post was written
- Writers can run Pangram on drafts
- They can submit a report if they think content was classified incorrectly
- They can also disable AI detection on their content entirely, regardless of whether Pangram got it right
Pangram itself uses AI to identify AI-written text, but not a generative model. Instead, it relies on a classifier neural network trained to distinguish between human and AI-generated writing. The company says its dataset of human-authored text came from sources published before 2021, before generative AI became widespread online.
Best said Substack is considering more AI-related features based on user feedback, including recommendation preferences that could eventually let users filter out AI-generated content altogether. That comes as a recent study from digital marketing firm Graphite found that the number of online articles primarily generated by AI is now equal to the number written by humans.
Culture Editor
Maya explores gaming, streaming, and the internet as a place where people actually live. From deep-dives into creator economies to the anthropology of digital communities, she tracks platform drama and cultural shifts so you don't have to. She believes the best tech stories are fundamentally about human behavior.
via Mashable
