Substack Launches AI Content Detector, but Experts Warn of Structural Blind Spots

Substack this week integrated Pangram, an AI content detector, to scan posts, notes, and comments over 100 words and indicate whether text appears human- or AI-written. The platform's CEO Chris Best framed the move as a transparency tool rather than a ban, giving readers more context about content they consume. Critics point out that such detectors are calibrated to identify 'AI-shaped' text, not whether genuine human thinking occurred — meaning polished, well-structured writing can be flagged regardless of its true origin. Non-native English speakers are particularly at risk, as formal or carefully translated phrasing statistically resembles AI output, a bias documented widely enough that several universities have abandoned AI detectors entirely. The Atlantic has also reported that Pangram itself was linked to a false AI-writing accusation that led to a novel being pulled before publication, raising concerns that a mostly reliable detector may be more harmful than an obviously flawed one.