Aug 17, 2026
8 min
The Invisible Pattern: How AI Watermarking Actually Works
Last week, Anthropic started watermarking everything Claude writes. Google has been doing the same thing to Gemini's text for over a year.
The first question everyone asked was fair: does this make the AI worse?
The answer is no - but almost every explanation of why gets the mechanism wrong. Including the one you've probably read. Let me walk through what's actually happening, in plain English.