Substack adds AI detection for posts and comments

Substack has introduced an AI detection feature that allows users to estimate whether text was written by a person or with AI assistance.

Developed in partnership with Pangram, the tool can scan posts, notes, comments and replies longer than 100 words that were published on or after 21 July 2026.

Results are not displayed automatically. Users must select the ‘Scan for AI text’ option to receive an estimate of how much of the content is likely to be human-written or AI-assisted.

The feature is available on the web and iOS, with Android support expected later.

Creators can add a ‘How I make this’ statement explaining their writing process, scan drafts before publication and report results they believe are inaccurate.

They can also turn off detection on individual posts or notes, in which case readers will be told that an AI analysis is unavailable.

Substack said the aim is to improve transparency rather than discourage responsible AI use, arguing that problems arise when readers’ expectations about authorship do not align with how content is produced.

The company is also considering community preferences for AI content, recommendation controls and further measures targeting spam, bots and AI-enabled scams.

Why does it matter?

Substack’s move reflects growing pressure on publishing and social platforms to provide clearer information about AI-assisted authorship without treating every use of AI as deceptive. On-demand detection and voluntary process disclosures may help readers make more informed choices. However, estimates can still be disputed or disabled and cannot determine the quality, originality or level of human judgement behind a text. The effectiveness of the approach will therefore depend on how users interpret the results and how reliably the detector distinguishes substantial AI generation from limited editorial assistance.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our chatbot!