Substack has added a Pangram tool that provides a probabilistic assessment of AI use in posts, notes, replies and comments, according to the platform’s official launch discussion.
A scan can be requested for sufficiently long text; TechCrunch reports a threshold of more than 100 characters. The result is split among AI, AI-assisted and Human categories and is presented as a suggestion, not proof of authorship.
Pangram founder Max Spero acknowledged that false positives and false negatives are possible. The system analyses patterns in the final sequence of words but cannot determine how much human effort went into prompting, editing or rewriting.
Substack does not plan to ban or punish AI-assisted writing. Authors can explain their process, report an incorrect assessment and remove a scan from their work.
“When I used to pitch Substack to writers, one way I would, and to this day, sometimes the way I’ll pitch it is like, we’ll do everything for you except the hard part. Right? You have to have an idea that’s worth reading, that’s worth caring about, that’s worth sharing. And if you can do that, that one thing is very hard and very valuable. And if you can do that, software should do everything else. Software should take care of hosting your website and making your transcript. But I think you do want the person to do the hard part because that’s the point,” — Chris Best, chief executive of Substack.
Best called the detector a good use of AI, but the company did not provide an independent evaluation of its accuracy. The result should therefore be treated as a signal for further review, not a final verdict on how a text was produced.
