Overview
- Substack rolled out a Pangram-powered scanner on Tuesday, July 21, 2026, that lets readers request an analysis showing what share of a post, note, reply, or comment reads as AI-generated, AI-assisted, or human.
- The tool works only on text passages longer than 100 words and only on content published after the rollout time, and it is live on the web and iOS with Android coming soon.
- Creators can add a “How I make this” statement, run the detector on drafts, report disputed classifications, and in some cases disable detection for individual posts.
- Pangram says its classifier was trained on human-written material published before 2021 to avoid contaminated examples, but detectors cannot prove whether a human did the thinking and are known to misclassify careful, heavily edited, or non-native-English writing.
- The move responds to studies showing large shares of AI-generated longform content on other platforms and could protect reader trust while also risking reputational harm for creators if false positives or policy incentives push authors to hide disclosures.