Don’t Filter Email Based on AI Detection

Summary

This opinion article criticizes Pangram’s Gmail integration, which lets users automatically archive or send emails flagged as AI-written to spam. The author argues that Pangram detects signs of AI use, not whether an email is low-quality “slop,” and that a detection result says little about the human effort or legitimacy behind a message. Pangram reportedly claims 99.98% accuracy, while the article notes that even correctly classified messages can be ordinary, useful correspondence. The author gives two examples drawn from public AI-chat research: an email asking a professor to resolve missing course recordings and a request to an academic adviser about teaching-assistant opportunities. Both were rewritten with AI assistance and flagged by Pangram, despite being clearly legitimate messages. The article also explains why people may use AI to write email, including time pressure, clarity concerns, and writing in a non-native language. It warns that automatic filtering could hide such messages from recipients, while people seeking to evade detectors may adapt their writing. The author says further research is exploring detector evasion and models trained to write more like humans.