Substack has introduced a new AI detection tool that flags content generated by artificial intelligence, prompting immediate criticism from writers who call the move a witch hunt.
Mack Collier, author of the newsletter Backstage Pass, stated he would not apologise for using the technology. He noted that his output quality would drop without it and that his posts would become less structured. Collier argued that AI helps improve the overall quality of his writing. He added that he could write without the tool if he chose to, having done so for 20 years.
Alice Lemee, a ghostwriter and digital writing coach, described the detectors as notoriously and wildly inaccurate. She warned that a single false accusation could irreversibly damage a writer’s reputation. Lemee made these comments in a video discussing the new features.
Substack CEO Christ Best announced the feature on Tuesday. He stated that it is becoming harder to distinguish real content online. Best argued that when content made by no one takes over sections of the internet meant for human voices, it pollutes the commons and makes it difficult to discover authentic work.
The platform has partnered with the AI detection tool Pangram to build the feature directly into the site. Users can now scan any piece of writing to receive a percentage-based score indicating whether the text is AI or human generated. Pangram previously purchased advertising on 404 Media. Previous research from the company, and studies relying on its detector, have shown AI writing flooding various parts of the web.
Earlier reports have highlighted that AI detectors are not perfect. Pangram itself can label human content as AI or vice versa. Max Spero, CEO of Pangram, recently told us the company works constantly to minimise errors. He estimates the false positive rate is roughly one in 10,000.
Sam Illingworth, a professor and author of Slow AI, described the system as a mirror image. He wrote on his own Substack that Pangram is a real machine trained on vast amounts of text to make a probabilistic guess. Illingworth noted that Substack asks a machine to decide if a human is hidden behind the writing.
The platform also added a way for users to include a How I make this statement. This section allows authors to explain their writing process and disclose if they use AI.
Best said the company is not against people using AI to assist their work. He added that people should be free to choose which tools they use to express themselves. He noted that some on the platform are human-writing maximalists while others publicly explore AI tools. Best mentioned that Substack uses AI to write software, conduct research, and build product features like clipping and translations. He stated that readers should know what they are getting.
Illingworth told me that inviting further dialogue is the best way for Substack to address the situation. He expressed liking the new feature that lets authors tell readers how they construct their newsletters using AI. He argued this is the kind of dialogue the platform should encourage rather than ascribing a score that is known to be incorrect. Illingworth worried that the detectors generate false flags for non-native English speakers and neurodiverse writers. He also noted that the tool removes any opportunity for dialogue because it starts from a position of suspicion. He said the goal should be developing opportunities for trust between readers and authors.
A Substack spokesperson said the tools are intended to increase transparency and give readers more context. They do not prohibit or penalise AI-assisted writing. The spokesperson added that the feature does not impact discovery on the platform. They encouraged publishers to add the How I make this statement to explain their process directly. Creators can disable detection on their posts pre- and post-publication. They can also report and remove scans on their own work that they believe are mistaken. More information on how the features work is available on the support site.
Some readers celebrated Substack’s attempt to detect and label AI generated content. Many are exhausted by AI slop and other internet platforms that cannot or do not want to label it. A recent story highlighted how much unlabeled AI slop exists on Spotify, prompting people to create their own sites to detect and label it without the company’s help.
The backlash from some Substack users shows that companies cannot just flip a switch if they want to ban or give users the ability to filter out AI content from their feeds. The flood of AI writing and images seen online and in the real world makes clear that generating slop is easy. What we are going to do about it remains an unsolved problem.
What it means
Writers now face a choice between transparency and potential automated penalty. The new system does not ban AI use, but it forces authors to declare their methods. This shifts the burden of proof onto the creator rather than the reader.
