Substack announced the launch of a new feature that allows readers to verify whether the articles they are reading were written by a human or using artificial intelligence, in a step aimed at enhancing transparency and limiting the spread of automated content online.

The new feature comes in partnership with Pangram, a company specializing in detecting texts produced by artificial intelligence, where users can check articles, posts, comments, and replies to see how likely they are to be written by chatbots.

Tool for content inspection

Substack CEO Chris Best explained that the feature is part of the platform's efforts to combat what he called "Claudefishing," a term referring to content that relies on artificial intelligence but is presented to the reader as human-produced.

The inspection tool works on any text over 100 words published from July 21 onwards, while the inspection result is only visible to the user who requested it and not to other readers, according to a report published by "Digital Trends" and reviewed by "Al Arabiya Business."

Best pointed out that Pangram's technology is not infallible, but it achieves a high level of accuracy according to independent tests.

The feature is currently available via the web version and the iOS app, and will later be available on Android devices.

New tools for writers

The new additions were not limited to readers, as writers can now use the same inspection tool before publishing their articles to verify the content analysis result.

The platform also added a new section called "How I make this," allowing the writer to explain how the article was prepared, including the extent of their reliance on AI tools, aiming to clarify the workflow to readers from the outset.

If the writer believes the inspection result is incorrect, they can report it and request the removal of the label.

Substack confirmed that it is studying the future addition of options that allow writers to set policies for the use of AI in replies, in addition to giving readers more control over the type of content the platform suggests to them.

Race to limit automated content

This step comes at a time when AI-generated content is increasingly spreading on social media platforms.

The CEO of Substack cited estimates from Pangram indicating that up to 40% of posts on some platforms are now entirely generated by artificial intelligence.

Substack is not the only one moving in this direction; LinkedIn recently began reducing the spread of posts and comments that rely on AI, while Meta is also testing a special tool to detect this type of content, but has not officially launched it yet.

Despite these efforts, the question remains whether content detection tools will be able to keep pace with the rapid development of AI models that have become more capable of mimicking human writing style.

Advertisement

Advertisement