CLAUDE STARTED WATERMARKING EVERYTHING IT WRITES

What this means for you and using AI. 

In the last week or two the internet lost its ever loving mind about AI watermarking any content, and we've had a bunch of you asking us what it actually means, do you need to stop using AI so I wanted to address this topic.

On August 2nd, Anthropic (the company that makes Claude) started putting an invisible watermark into the text Claude writes. It's part of a European AI law, but they decided to apply it everywhere in the world instead of just the EU, which means it applies to us here too.

You can’t see this water mark and it doesn't change the words or how they read. It's woven into the actual word choices, sort of like a fingerprint in the writing itself, and it stays there when you copy and paste it somewhere else.

Google's already been doing this with Gemini, so this isn't one company being extra, it's where the whole industry is heading. Which I do think some people appreciate. 

WHAT ACTUALLY GETS MARKED

(This was the first thing I wanted to know.)

If I write something myself, my own words, my own thoughts, and I paste it into Claude and say "hey can you spell check this for me," and then I copy it back out... is that watermarked now?

Yes. Probably.

Anthropic says it right in their own documentation, which I appreciated. Their words: Claude may not be the original author. People use Claude to proofread, translate, summarize, and the output can carry a mark even if the ideas and the text came from somewhere else. (womp womp)

So the mark isn't saying "AI wrote this." It's saying "this passed through Claude." Those are really different things and I don't think most people posting about this understand the difference. Or they are totally click rage bating us to click, comment and blow up their post so it goes viral. I also think it’s good to educate more people that just because it has an AI water mark doesn't mean it’s not the author's original idea! 

A few other things worth knowing:

It doesn't survive heavy editing. If you rewrite it, paraphrase it, translate it, or mix it into a bunch of your own writing, the signal fades out (per my understanding). Short bits of text might not have enough there to detect at all.

To my knowledge, there's also no way to check someones writing to see if it has the AI watermark.

Essentially the watermark is a secret pattern of words. Anthropic has a bank of words that it can choose from and so as it chooses those words and creates a pattern (according to AI). Under the EU code they signed, they have to let users and third parties detect their marks, but they haven't released the detection tool yet. So anybody currently selling you an "AI watermark detector" is guessing, and anybody selling you a "watermark remover" that strips invisible characters could very likely be lying. But I am sure they will come.

The best way to “remove” the watermark is to HEAVELY change the AI version it spits out. Change things with YOUR words. Add in YOUR personality. 

And I am also not convinced we need to make SURE our writing doesn't have a watermark on it. It’s not dinging us. 

NO, THE WATERMARK ISN'T TANKING YOUR GOOGLE RANKINGS

There's a study making the rounds right now claiming that watermarked content ranks about five spots lower on Google, and that AI search tools cite it half as often.

We went and read the actual study, and from what I can tell, that is not the case. Could it change. Yup. But for now it’s not an issue. 

Every piece of content they tested was published after August 2nd, because that's when watermarking started. They published their results in August. So their content had roughly a week and a half to rank before they measured it, and brand new URLs don't rank in a week and a half, which means what they measured was mostly just the internet not having noticed them yet. They also say straight out in the writeup that they didn't control for content quality.

Google's position on this hasn't moved in three years. Using AI to make content isn't against their guidelines. What gets you hurt is pumping out hundreds of pages of unedited AI SLOP to gain search results, and that was true long before anybody was watermarking anything.

However, interestingly, Semrush looked at 42,000 blog pages and found human-written content was about eight times more likely to rank first, and that study ran before watermark detection existed at all. Which tells me the water mark is NOT your issue. It’s AI SLOP that is PAINFUL to read. If you have been using AI at al you can TELL when someone is using AI to write and not changing a thing in it. I am seeing it so much more and Claire and I have had conversations about it and how painful it is to read. If your posts SOUND like AI we don’t want to read it. And I have heard other people say it too. One friend even said, "It's SO obvious that she uses AI to write her emails and I refuse to read one more piece of that garbage.” May seem harsh but there is a certain tone to it. And they aren’t wrong. It can get painful.

My best advice if you are going to have AI help you take your thoughts, maybe organize them, but proofread and CHANGE the text to sound how you would talk! 

Claire even saw a post about big companies trying to make sure that their content does NOT feel like AI so they are infusing what is called naive design. The idea is deliberately unpolished. Wobbly hand-drawn lines, flat shapes with no real perspective, bright unsubtle color, faces and figures that look like a kid drew them. The design comes from naive art, which historically meant work by self-taught artists with no formal training, and modern brands adopted the look on purpose because it reads as warm, human, and a little imperfect. The idea is that you look at it and don’t instantly question if it was created by AI.

Claire has even confessed that she left a lowercase i in one of her emails to a client purposely so they would know SHE wrote the email not AI. I can also say I’m guilty of this, not using perfect punctuation in a post caption. I guess my horrible spelling and lack of proofreading will come in handy cause people won’t think, “did AI write this?”  (Don’t worry, if you hire me to write your website copy, I use Grammarly to catch all the misspellings before we go public) ;)

LET’S DISCUSS CANVA AND AI TAGGING IN INSTAGRAM

I saw a post from a girl who used Canva's background remover on a real photo she took, posted it to Instagram, and Instagram slapped an "AI info" label on it which tanked her views.

This might be a bit of a bigger problem than the watermarks on your writing because Instagram IS saying they are favoring REAL content over AI content. So this is where it COULD start to affect your visibility. Instagram isn't looking at the picture and determining if it LOOKS like AI, it's reading hidden data inside the file, and design tools write "AI was used here" into that data any time you use an AI feature in what you are creating. And Canva has confirmed that features like their background removers, magic erasers etc are in fact AI tools. The crappy part is one tiny edit on a real photo (like using background remover) gets tagged as AI the same as a fully AI generated image.

HOW TO REMOVE THE AI LABEL INSIDE A FILE.

If that happens to you and the photo is legitimately yours, you can strip that data before you post. Re-exporting the file usually does it. Meta has actually stated that this works. Just know the AI label gets decided the second you upload and it doesn't get re-checked, so if it's already live your only option is delete, clean the file, and repost the new clean file.

If you are a paying AI CLUB member, I left more specific instructions on how to "clean the file" in the memebrs only section. It’s really easy.

OUR TWO CENTS

Does this mean you should stop using AI? No.

Some people are going to decide they don't want their work marked, and that's a legitimate choice, we're not going to argue with anybody about it. But we'd want them to at least understand what they're opting out of, because a lot of what gets run through Claude is proofreading and cleanup on writing that was already yours. (At least that is how it SHOULD be)

We have been using and reading AI for a year and a half and it’s our opinion that the watermark is not the thing that's going to hurt you. AI slop is the thing that's going to hurt you.

We can spot AI writing from a mile away now, and so can you, and it's getting kind of painful to scroll through. The tells we see constantly: the choppy little three-word sentences stacked on top of each other for drama. The sentence that announces what the next sentence is about to say instead of just saying it. Two word sentences in sequences of 2 or 3 examples. And using phrases like “here’s the truth,” “here’s what we’ve seen,” “Unlock your,” “the thing is,”  “But here’s the thing,” 

None of that has anything to do with watermarks. It’s just writing that nobody wants to read.

I think after a little bit of time people won’t care whether text has a was watermark on it what they will care about it if it is painful to read or if it was a waste of their time. 

Do I use AI to write this newsletter? YES. but here is how I do it. 

I use AI to do research for me at times so I can see what’s happening out there. What people are talking about and what you and I really need to know.

I think use Wisprflow (BEST freaking app out there) to speak into my microphone and tell AI what I want to say in this newsletter. Then I let AI write it up. (Cause it does that WAY faster than I can) and THEN I spend an hour reading through it and changing about ½ to say what I want it to say the way I want to say it. I add little after thoughts or my personality to it. Some of these paragraphs were completely typed by me after the fact.

The more important part is, when you read this today, did it feel like AI wrote it or did it feel like we were sitting having coffee and I was spilling the tea on the latest AI news? (Well, my hope is the ladder) because I sat here first and just talked into my Claude like I was chatting about it over coffee. 

On a totally side note, I’m not gonna lie, I keep having this fear that if we all keep reading too much AI slop we are just going to start talking and writing like AI. NOOOOOO. So please, add your personality back into it if you are going to have AI write you something.

BONUS FOR AI CLUB MEMBERS

This week I am giving you a list of AI tells and a prompt we have built into some of our writing skills to remove the AI slop from things we want AI to write a first draft on.

I’ll also drop in the steps to remove the AI label on an image before you post it into Instagram so it doesn’t get flagged as AI. 

XO Krissy + Claire