Substackers Say New AI Detection Tool Is a ‘Witch Hunt’

Substack contributors are expressing outrage over the company's new AI detection tool, dismissing it as a "witch hunt." Some creators argue that the policy unfairly targets those who use AI to write or edit their articles, while others fear that Substack may mistakenly flag their non-AI generated content, damaging their reputation.

Mack Collier, a Substack author, stated that he believes using AI improves the quality of his writing, whereas Alice Lemee, a ghostwriter, warned that the tool is "notoriously, wildly inaccurate" and could irreparably harm a writer's reputation with a single false accusation. Substack CEO Christ Best revealed the AI detection features on Tuesday, explaining that the platform is partnering with AI detection tool Pangram, which generates a percentage-based score indicating the amount of human or AI-generated content in a piece.

Pangram's CEO, Max Spero, acknowledged that while the company is working to minimize errors, the false positive rate is estimated at roughly one in 10,000. Sam Illingworth, a professor and author of Slow AI, stated that Substack is essentially asking a machine to determine the presence of a human in the writing process, which raises concerns about the accuracy of the AI detector.

Substack also provided an option for users to add a "How I make this" statement, disclosing their writing process and if they use AI. CEO Best emphasized that the company is not against the use of AI tools and encourages transparency, but Professor Illingworth argues that this approach fails to address the issue and instead creates a "witch hunt" atmosphere by starting from a position of suspicion.

Substack's spokesperson stated that the tools are meant to increase transparency and provide readers with more context, not to penalize AI-assisted writing, and creators can disable detection and report mistaken scans.