Your AI Is About to Start Labelling Its Own Work

Last week Anthropic announced it will add invisible text watermarks to all Claude text output worldwide, and C2PA signed provenance data attached to the images it generates. Ostensibly this helps meet the requirements of the EU AI Act.

While many of us think that this might be a good thing, there are a few things to be aware of.

Running your copy through Claude for a spellcheck is unlikely to mean that the whole thing could be labelled as AI. Leaving aside that this is a very expensive spell check (the one in MS Word is free), it should be worth pausing for thought before proceeding. The same will also apply to formatting it.

AI is now such a ubiquitous tool in researching, analysing and drafting content that it is unrealistic to revert to everything being hand-crafted by a human. You can, and probably should, use AI to help develop your thinking. If you have an article which you plan to publish that has been completely or mostly produced by AI, then it will likely be labelled as such.

It will be read as your thinking, so it needs to be meaningful and able to land with your audience. Not just generating a piece of content for the sake of it.

Some questions to consider include:

1. If a customer knew AI helped write the article, would it matter? If the answer is no, then it has enough value.
2. Where do you need the human touch, and where can AI be used to leverage your time? Some emails and chunks of proposal can indeed be automated, but it will never replace human judgement. Taking time to test and evaluate this is important.
3. Does the content sit outside your domain, where you don't feel confident telling good from bad? If that's the case, then not sending may be the right thing.

That last question is the same judgement call that decides what an AI agent should be allowed to do on its own. It is the argument for keeping a human checkpoint on anything that is hard to check or hard to undo, applied to your writing instead of your workflows.

The human voice keeps its premium

There has been tooling around to try to make AI more human, and it is probably getting better, but I don't think it's there yet. My wife is in film and TV and there is a lot of angst about AI in that industry. I think this development from Anthropic, and the others soon no doubt, means people will continue to put a premium on the human voice. Whether it's a movie or a LinkedIn post, how can we best create a connection to our reader?

If you want a hand mapping where AI touches your customer-facing work, and where a human checkpoint belongs, email me.