Substack adds a tool to estimate if a post was written with AI

In short: Substack is rolling out a built-in checker that estimates whether text on its platform was written by AI.

What happened

Substack says it will now let readers check whether what they are reading may have been written by AI. The company is adding a new option called “Scan for AI text” to posts and other parts of the platform.

The tool can scan posts, notes, replies, and comments. It gives an estimate of how much of the text could be AI-generated or written with AI assistance. It is powered by Pangram, a company that specializes in AI detection.

Substack says the feature is rolling out on the web and in its iOS app now, with an Android version coming soon. Readers can scan content that is longer than 100 words by opening the three-dot menu near the top right of a post.

Substack CEO Chris Best said the goal is “transparency,” especially as more AI-written text shows up online. He described the main issue as a gap between what readers think they are getting and what is actually there, like reading something that seems personal but has “no human thought on the other end.” He calls that “Claudefishing,” meaning content that looks like it was written by a person but may not be.

Substack is also adding a “How I make this” statement that writers can use to explain their process. Writers will be able to scan drafts too, and report results they think are wrong. Best also noted that the detector can only guess whether AI was used, not whether a human put care into the writing.

Why it matters

For readers, this is like a “nutrition label” for writing, it will not be perfect, but it may help people decide what to trust and where to spend their attention.

Source: The Verge AI