Substack adds a tool that discovers AI-generated content

Substack introduces a new tool to help readers identify whether content was written by a human or generated by AI. Announced in a blog post on Tuesday, the new self-publishing platform feature is powered by AI discovery tool Pangram, and is available for Substack posts, notes, replies, and comments.

“Not everything made with AI is slop, and not all slop is made with AI,” wrote Substack founder and CEO Chris Best. “The problem is when there is a conflict between the student’s expectations and reality, especially when they unknowingly focus their attention on something that has no human thought on the other side.”

Substack’s AI discovery tool is now available on the web and as an iOS app, with Android to follow. To scan Substack content for AI contributions, simply tap or click on the “…” meatball menu, then select “Scan AI text.” A popup will then appear stating what percentage of the text Pangram has determined is AI-generated, AI-assisted, or human.

Substack’s new AI detection feature has some limitations. Specifically, it will only work on Substack posts longer than 100 words, and only those published after 8:30 am PT / 11:30 am on July 21. Attempting to scan Substack content published before this results in a popup stating that it is not eligible for AI detection.

Attempting to scan passages shorter than 100 words also results in a popup stating that there is not enough text for accurate AI detection. A popup on Best’s own post also offers an additional message: “I’m still considering all of this. (For now these words are mine, for better or worse.).”

Pengram will only scan the text of any AI identity if the reader requests it. Therefore, it is up to users to take action and investigate the basis of what they are reading to the best of their ability. The best notes the tool can’t detect if, for example, a person has used AI to do research before writing a post.

Substack’s AI-detection tool adds new features to authors. Authors are now able to add a statement explaining how they wrote the content, which will appear when readers use the AI ​​discovery tool on their posts. Authors can use Pengram’s AI detection tool on their drafts and submit a report if they believe it has edited their content incorrectly. Interestingly, authors can also remove the flag themselves by completely disabling AI detection of their content, regardless of whether Pengram’s detection was correct or not.

Pangram itself uses AI to find AI-generated text. However, instead of generating AI, it uses a neural network model trained to recognize and distinguish between human and AI-generated text. The company says its human-authorized text algorithm data was collected from sources published before 2021, before AI-generated content was widely used online, to ensure that no AI-generated text gets in its way.

Best of all, Substack is looking to add AI-focused features in the future, depending on user interest and feedback. This includes allowing users to set preferences for the content Substack recommends to them, suggesting we may eventually see an option to filter all AI-generated content.

A recent study by digital marketing company Graphite found that the number of online articles generated by AI now equals the number written by humans. Many publications have been cited for using AI to generate articles over the past few years, several of which have been riddled with errors. In some cases, these AI-generated articles are said to be written by fake, AI-generated authors or “staff” of the publication.