So, you’ve saved money on an experienced professional copywriter by getting Anthropic’s Claude to write it all for you?
Great. Claude is excellent, but with precise prompting you can just as good a result from ChatGPT, Perplexity, and Gemini.
However, if you’re simply cutting and pasting text direct from Claude, do you know that they’re watermarking that text? According to the official announcement;
Future Claude models will generate text that contains a watermark. This is a way of determining the likelihood that Claude was involved in writing the text, and we, along with several other major AI providers, are implementing this change to comply with the EU AI Act.
So there you have it. There’s no shame in using AI to help you generate your website text, Google can still rank you very highly, but this is for compliance with the EU’s AI Act that explicitly states that if your site provides public information viewable by audiences in the EU, then you need to label your content if it is AI generated.
This new Claude watermark deals with that.
Meat Proxy
I recently saw a new term “meat proxy” with the definition of it being a person who uses raw AI-generated content without checking, validating, or even understanding what they’re copying and pasting.
This is an interesting development in the pushback against “AI Slop”. It’s not just in videos or images, but can be almost undetectable to the untrained eye in the written word too.
So a person who blindly pastes AI content into websites is a simplistic proxy doing the work of the AI without any expertise, unfortunately.
I’ve seen something similar recently where a business had used a third party to build their website, resulting in a misspelling of their brand, not just in their H1 tags but in the footers and all through out the body copy. They even cocked up the location of the company with a glaring typo in the city name.
Now, AI should have flagged this up, but someone blindly pasted the wrong name and the wrong location in a professional website, and that would be embarrassing for the client, but amateurish for what I suspect is an outsourced overseas provider. Without the experience, the expertise, the quality control or even an understanding of the English language, they butchered what was an otherwise well-designed website.
Back to the Watermark
Unlike images that have a literal watermark in them, Claude’s text watermark will not be visible or detectable to humans.
How it works is that AI-generated text does follow patterns, experienced eyes can see that. But going forward, Claude will have “keys” that align with patterns of text, and so by using raw, unfiltered AI-generated copy, you’ll be indicating to the world that you’ve used it without thinking.
This complies with the EU AI Act and there is a code of conduct that something like 190 signatories have already signed up to. So expect the same of OpenAI’s ChatGPT, Googe’s Gemini, and Perplexity, etc.
Also, these technologies will start to filter out into public use, so detectors will probably have access to new APIs so that they can interrogate bodies of text.
Of course, editing can reduce the level of signals to the key, with light editing to a lesser extent, and heavy editing to the extreme. Ultimately, LLMs are trained on human language anyway, so at what point do we have this schism between flagging up human-written content as AI and machine-generated output as being human?
Finally
So make sure you have an expert looking at, editing, publishing, and managing your content.
If you’d like me to audit your website for typos, grammar, and look to improve your written copy, then I’d be happy to do that.
Call me on 07730 499 539 and let’s talk humanising your AI-generated content so that you connect better with your audience.