ChatGPT is hiding a watermark in your emails

Your customers can already tell when an email from your business was written by ChatGPT, and now there's a hidden watermark in it as well.

OpenAI has announced it is watermarking ChatGPT text across the EU, so if you're in Ireland, it's coming to your account. You can't see it and there's nothing to delete. The mark sits in which words ChatGPT picks, in a pattern only OpenAI's detector can read, and it survives copy and paste. Google has watermarked Gemini text since 2024, and Anthropic does the same with Claude.

Before you panic, your customers can't run your emails through that detector. It isn't public, and only approved researchers and organisations can apply for access.

The bigger problem is the AI email your customers already spot without any detector. Look through your own inbox and you'll find three or four before lunch.

So keep using AI for the research and the first draft. Then read it out loud, and rewrite every line you'd never say to that customer's face.

In OpenAI's own tests, swapping a quarter of the words for synonyms dropped detection from about 92% to 17%. Rewrite it properly and it sounds like you again.