Would you want your writing watermarked because you asked AI to fix your commas?
by•
Both big AI labs quietly switched on watermarking this month — Anthropic marks every text Claude touches, OpenAI marks GPT-Live voice audio (EU AI Act, Article 50).
The catch: a detected mark means the model processed the text, not that it wrote it. Fixing grammar leaves the same mark as full generation. And no mark proves nothing either — edits, translation or transcription wash it out.
For makers this cuts both ways: your hand-written landing copy might get flagged because you ran a grammar pass, while actual AI spam walks through clean. How are you thinking about AI provenance in your products?
7 views
Replies
The gap you're pointing at is the real problem. A mark that fires the same way on a full generation and a comma fix tells you the model touched the text, nothing about who wrote it. For makers that's worse than no signal, it teaches people to distrust the wrong thing entirely. What would actually help is provenance scored per paragraph rather than per document, since most real writing today mixes both within the same page. Doubt many labs are building for that granularity yet.