Substack launches AI detection tool powered by Pangram
Feature update ● Confirmed 95% confidence first seen
Substack has rolled out an AI detection feature powered by Pangram that analyzes text to estimate how much of newsletter content was written by humans versus AI-generated. The tool is available on web and iOS (with Android coming soon), works on content longer than 100 words, and aims to increase transparency about content provenance while helping readers identify potentially AI-written material.
Decision brief
- What changed
- Substack has launched an AI detection feature powered by Pangram that scans newsletter posts, notes, replies, and comments (content over roughly 100 words/characters) to estimate how much text was human-written versus AI-generated. It is live now on web and iOS, with Android support coming soon, and creators can add optional disclosure notes about AI use.
- Why it matters
- This is a notable move by a mid-sized content platform to operationalize AI-content transparency at the point of consumption, which could influence reader trust, creator behavior, and monetization dynamics on Substack. For media and content-driven businesses, it signals growing normalization of AI-detection tooling as a distribution-layer feature rather than just an enterprise or education compliance tool, and raises questions about detection accuracy and its effect on creator reputation and reader engagement.
- Evidence
- Three outlets—The Verge, TechCrunch AI, and The Neuron—independently and consistently report the same core facts: Pangram-powered detection, availability on web/iOS with Android pending, and the 100-word/character threshold, indicating solid confirmation of the feature's existence and basic functionality.
- What remains uncertain
- None of the coverage details Pangram's detection accuracy, false-positive/negative rates, or how Substack will handle disputes or reader/creator backlash if scores are wrong; it's also unclear how this affects Substack's algorithmic ranking, monetization, or subscriber trust metrics. TechCrunch's note that the tool 'may inadvertently' have side effects is truncated in the summary, leaving unspecified risks unaddressed.
- Monitor next
- Watch for creator and reader reactions—especially disputes over detection accuracy or unintended flagging of human-written content—in the weeks following rollout, as well as whether Pangram's detection scores influence Substack's content ranking or recommendation algorithms.
Analytical support, not advice — assumptions and open questions stated above.