What's this about?
Substack integrated detection startup Pangram so readers can scan any post, reply, or comment and get an estimate of how much was written by a human versus with AI assistance.
What exactly is an "AI detector"?
A classifier trained to spot the statistical fingerprints of machine text — telltale word-choice patterns and rhythm. They're probabilistic, not proof: false positives happen (especially for non-native writers), and paraphrasing tools weaken them. Treat the output as a hint, not a verdict.
What's the community saying?
1.4k likes on the announcement
- This is the first major *reader-facing* deployment — detection as a platform trust feature rather than a teacher's plagiarism tool.
- Writers worry about false accusations damaging reputations, given detectors' known error rates.
- Ethan Mollick's wry note — "glad I have always written my own posts" — captures the retroactive-audit anxiety: a decade of archives just became scannable.