On 22 July, Substack launched a feature that estimates how much of a post was written by a person and how much by AI. The scanner runs on the Pangram detector. Readers can request it for posts, comments and replies published since the end of July.

Co-founder and CEO Chris Best introduced it with the words “This is good use of AI”. Substack continues to permit AI assistance. The scanner turns that assistance into information for the reader.

The scanner runs on Pangram and reads texts over 100 words

The feature works on texts longer than 100 words, and for now only on the web and in the iOS app, with Android to follow later. It covers posts, notes, replies and comments, and the result is shown only to the person who requested the scan.

Alongside detection, Substack added a “How I make this” field, where a creator can describe how they used AI while writing. A publisher can also run Pangram on their own draft before publishing. If they consider a scan wrong, they can report it and have it removed. The company Pangram was founded by Max Spero.

Best frames the problem as a gap between expectation and reality

Best builds the whole step on trust in the reader relationship. “When readers have to wonder if what they’re reading is real, it undermines trust in authorship and threatens the livelihood of writers,” he wrote in the announcement. In his view, the problem arises when a reader unknowingly gives attention to a text that nobody stands behind.

“You have to have something: an idea that’s worth reading, that’s worth caring about, that’s worth sharing,” he said, adding that software can handle everything around it, but the hard part has to be done by a human.

For Substack, whose business model rests on paid subscriptions to individual authors, human authorship is exactly what it sells. “Human perspective is the essential ingredient of culture, and this will be true no matter how powerful AI becomes,” Best wrote in the same text.

The detector only recognises the text itself

The tool has a clear limit. It can only tell whether a model was used on a text. It does not measure the quality of human oversight or the value of the idea behind the text, and Substack says so itself. The scan tells you only whether the sentences were put together by a person or a machine.

For this reason, Substack presents the feature as a signal for readers and as room for voluntary disclosure.

Labelling has been running for longer on video and music

Substack is entering an environment where other platforms introduced similar labelling earlier. Since 2024, YouTube has required creators to switch on a disclosure for realistic “altered or synthetic content”, which then appears as a banner under the player. It began enforcing this at the start of 2025.

The French streaming service Deezer began labelling AI tracks in June 2025, presenting it as the first AI music tagging system in streaming. Fully AI tracks are excluded from algorithmic recommendations and editorial playlists, so that they do not dilute the income of real artists.

In April 2026, fully AI-generated music made up 44% of the tracks uploaded to Deezer each day. On 21 July the platform announced that more than half of its daily uploads were AI-generated for the first time, it said.

While platforms introduce labels on their own, the European Union is making labelling a legal obligation. The transparency rules under Article 50 of the EU AI Act apply from 2 August 2026. They require the outputs of generative systems to be machine-readable and labelled as AI-generated, and they require deployers to make AI content available to the public with a clear disclosure.

A practical guide is to come from the Code of Practice on Transparency of AI-Generated Content, whose first draft was published in December 2025. The Commission published the final code on 10 June 2026.

The rules apply to providers and deployers of AI systems, so the concrete impact on a service like Substack will depend on interpretation and enforcement. The direction, however, is the same as Best’s step. The origin of a text is becoming information that the reader is entitled to.

Substack has made this part of the product. Where an author sells a subscription on their own name, the subscriber pays also for the certainty that a person stands behind the sentences.

By Tereza Tiller