One of the big AI items that is getting a lot of press over the past few weeks is the news that Claude is watermarking their text output so it can be identified as AI when published. Specifically, they say:
When a supported Claude model generates text, it weaves an imperceptible watermark directly into the text itself. You won’t see it, and it doesn’t change the meaning, quality, or readability of Claude’s response.
They can’t hide code directly inside of text, so they’re really just putting a few words in a particular order to identify them later. It’d be like if I said “the first six words in this sentence start with IBLIIS, so we can identify that later”.
As you might expect, this means that the watermarking will fade if you make edits to the text, as they explain:
Because the watermark is part of the text, it will travel with the text when it’s copied and pasted elsewhere, and may persist through some editing.
Does Google care?
In a word, no. Back in early 2023 Google made it clear that they just want to see high-quality content, and they don’t care if was made with the help of AI or not.
As Stephanie Richards points out, the bigger concern would be if you claimed to write by hand but were caught using AI. She shares:
So what do you do if you have produced content using AI and are worried that you might be “found out?” Honesty is the best policy in my book. A disclaimer on your website saying how and when you use AI alleviates a lot of distrust. You can also label content, such as images, as being generated with AI. It is better to be upfront than get called out.
The other angle, as I first shared nearly two years ago, is that AI detectors simply don’t work. This tool may work slightly better than tools from the past, but “watermarking” text is a very tricky thing to do and won’t hold up to any decent degree of editing.
My overall two thoughts on this:
- Be honest about what you do. Using AI to generate some content can be ok, but be up front about it. Everything on here is written 100% by hand, by me, but I openly use AI in other areas of my life.
- Search engines don’t care. My guess is that that’d prefer if everything was hand-written (and therefore more likely to be helpful as they grow their models), but they know it can’t be reliably detected so instead they just focus on quality.
For more, Clifford Almeida has a nice write-up that gets into more details on what’s happening here.
