What's Up THIS WEEK?! ⚠️ THE AI WATERMARK PANIC: What’s Real, What’s False & What to Do | Ep. 553

0:00Welcome to Do This, Not That, the podcast for marketers. We share quick tips, things you can do right now, and then we add a little bit of chaos at the end of every episode. We also keep it short, like this intro. Let's check it out. We are back for What's Up This Week from the Do This, Not That podcast, and I wanted to jump in to what is the biggest topic right now in marketing and AI. What's gone on in the last week with Claude and ChatGPT and them embedding these watermarks and these ID markers into any copy that AI helps us write, any copy that AI edits, any images that AI helps us create or edit, it, there's now these identification markers that we can't get rid of, but I'm going to tell you what you can do about it for both Claude and ChatGPT, that now it can be identified that AI helped do this.

0:57And what's going on is people are freaking out. Number one, they don't want their stuff. If they're writing copies, somebody said, oh, use AI to write that. But also if you take this stuff that the platforms are giving it to you, and then you use that to run social media campaigns or YouTube campaigns or LinkedIn campaigns or whatever, that those platforms are then able to identify that AI was used and they are deprioritizing the engagement is much lower. So I want to detail for you what has gone on, what you can do about it. And this is the truth because there's so much rumor out there. There's so much garbage.

1:30So I want to tell you what's actually going on. So first off, what has happened? Okay, so what happened was Claude, which is Anthropic is the company that makes Claude, okay? And on August 2nd, Anthropic put an invisible machine-readable watermark into any text it generates or into any text it heavily edits. And why did they do this? They did this to be compliant with the European Union's AI Act that just went into effect, which is this whole idea of transparency around AI-generated content. But they didn't just have it apply to Europe. They had it apply to the entire world, everything, everywhere, period, hard stop.

2:15Now, here's the deal. No matter what tier you're on, okay, it does not matter. You could be paying for the Pro Max whatever. it's still going to be putting in there this stuff now here's what it is and what it's not claude is not secretly inserting some invisible characters into your paragraph that's not what it's doing what it is and that's where a lot of the rumors are what it's doing is claude is writing something like for example would say if it gave you back some copy maybe it would write something that said this campaign was surprisingly effective but it could also give you back copy that said this campaign performed surprisingly well.

2:59Essentially, they're both saying the same thing. And normally, their model has some randomness in deciding which perfectly acceptable wording to choose. But now what's going on, Claude is basically using what's called a watermark that is using these subtle differences and choice patterns to hide a pattern in what it gives you back. And while to your eyes and to a regular reader, you can't see that there's any difference here, right? But over hundreds of words and enough of these choices and statistical like fingerprints, it is now identifiable by Claude and any of the other AI tools to see, oh, this was not written by a human.

3:45So you can't see it. You can't spell check for it. Copying into a Google doc doesn't remove it. Pasting it into LinkedIn doesn't automatically remove it. The watermark is essentially a pattern of words Claude has chosen. So it's not a hidden object, right? So here's the deal. If you use Claude to write copy, okay, that's going to be a major problem for you. Now, if you use Claude to help you fix your punctuation and grammar, it's not going to be a problem for you. What is essential though, is that let's say you are using Claude the right way and you are writing something, your first draft, and you're not really good with grammar or punctuation.

4:30What you want to do is when you upload that, and this is very important, you want to say, here's something I wrote, fix my grammar and punctuation, but don't rewrite this. And when you say don't rewrite this, Anthropic has actually said themselves, there's going to me no detectable watermark because Claude's not actually writing it for you. So no longer, right, can you use Claude to just be like, write my LinkedIn post, and then you copy and paste it and publish it. No. Now you have to, you can give Claude your ideas, have it organize them with you, pull out what you like, you rewrite it, okay, and then you can say, you know, help me with my grammar and whatever.

5:11And that is going to allow you to not have these watermarks show up. And here's the secret. That's the way we should have been using this stuff the whole time. That's it. Okay. So let's get into ChatGPT because it's a very, very different situation over there. So OpenAI, right, is the company that owns ChatGPT and runs ChatGPT. They have a very, very different thing going on. they have nothing going on related to text okay there is no chat to bt watermark in their text but but there is um stuff going on with their images so their images have two different things going on in their images they have something called a c2 pa uh metadata thingamabob and then they have another thing called their synth id these are basically invisible watermarks and they're they're trying to expand this all over the place so what you want to be thinking about is um when you are having chat chappi help you make images know that they have these two things this metadata thing and this synth id thing so when you have a generated image what you want to do is have it generate the image to actually be t and then the worst thing you could possibly do is just download that image whatever it is and then use it somewhere use it on linkedin use it on on meta somewhere that's not going to work if you take that image and you bring into canva into photoshop okay and then you add a headline and branding and design elements and resize it and crop it and you make it your own and then you export that uh as like a jpeg or whatever the process will very likely, okay, have the metadata disappear.

6:56But it doesn't necessarily guarantee that the Synth ID disappears. That being said, we've tested this hundreds of times. And what the big issue is, if you have ChatGPT create an image for you, and then you upload it to different platforms, even after you output of Canva and all this different stuff, and you get rid of the metadata, We actually have not seen it show up as a flagged AI created image, but ChatGPT is saying that they're working on their Synth ID. So that way down the road, it actually will show up as a Synth ID AI created image, which will depress your performance. I don't care what anybody says.

7:35So here's the thing that nobody's really talking about yet. OpenAI, okay, has made a verification tool. Okay, it has made a verification tool. So you can now, okay, go to openai.com slash research slash verify. Okay, and this tool actually allows you to upload an image of any kind into ChatGPT. and then it will tell you, was this created using AI? So if you output your image from ChatGPT that you created, you mess around with it, and then you upload it to the new OpenAI image verification tool, and you can just Google OpenAI image verification tool, you upload the image after you mess around with it, it will tell you, hey, no, this wasn't created using AI, and then you will know that you kind of circumvented it all.

8:31I know that that's bad, but that's the reality of where we're at, And this is all important stuff. All right, some other non-important stuff happened this week. Did you see Instagram change their font or whatever? This should have been a nothing burger. I have to tell you, every company, all these B2B SaaS companies that rushed to mock their own, to take their own logo and change it into the font of the new Instagram font and put it out there. You all are dorks, okay? I'm sorry. That's not fun content. That's not cool. I don't know what to tell you. who cares that Instagram changed their logo?

9:06It doesn't mean you do enough. You didn't jump on a trend. You're a dork. All right. What's not dork is CallRail. They are the sponsor of this thing. And I want to tell you about them before we get to the ridiculous portion of this podcast. CallRail is awesome. If you're spending money on marketing and still saying, I think that the campaign worked, you have a problem. There is a free trial. CallRail.com. It's not about calls. Callrail.com slash do this. They show you exactly what's driving texts, chats, form fills, calls, and uses AI to help you understand what is happening. Basically, you stop guessing where your leads are coming from.

9:44They have a free 14-day trial. No credit card. Go to callrail.com slash do this. All right, let's get into since you didn't ask. What's going on? So another thing going on this week is Express, the clothing store that they have in my mall. listen this doesn't work so this past week express launched a new campaign called too easy to look this good whatever and it's this whole 90s and y2k nostalgia thing where they're trying to kind of make this thing around this beverly hills house and kind of like be like beverly hills 90210 and they think oh if we go y2k and and it's all about jeans that we're going to be able to be in this nostalgia theme enough that's not how it works you have to be cool speaking of express you didn't ask but when I was in high school I had a job I worked at the um the limited the women's clothing store I did and the reason I got a job there was I was like I have no I had no game I was so so not cool being like you know meet girls or whatever so I'm like if I work at a women's clothing store, which it sounds kind of cringe now.

10:51I didn't like stalk them or whatever. I was like, maybe I could like, you know, meet girls and stuff. I was at high school. And so I got a job there and they made me the, um, the stock boy stock person. And I sat in the back and what would happen is, um, the sales people would come running in the back. Like I need a, a blue bodysuit in size medium. I didn't even know what a bodysuit was. And I would run around and the back room of these stores are it's chaos okay you might think it's organized it is not and i had to run around and then they would yell at me go go faster you're terrible you're not finding it and i would i would find it i'll give it to them i get the wrong size the wrong style they would yell at me um i was terrible terrible at this job and not only did i not meet any girls i was completely unsuccessful um i alienated a group of female individuals because I thought I was an idiot because I couldn't figure out where anything was in the sock room.

11:50It was a terrible experience. So yeah, that, that, that, that happened quite a bit. I don't know what I'm talking about. Listen, I would appreciate if you leave the show or review, if you liked it, it does help circulate it, share it with a friend and don't ask me to get you a body suit later. Wait, don't go anywhere. There's two really important things I want to tell you. First off, my book is out. Stupider people have done it. And I don't know how, but this book has made it to the USA Today bestseller list, one of the top books in the entire country. Now, all net author proceeds are going to the V Foundation for Cancer Research.

12:29And between this book and things that we're doing, we've now donated$130 ,000 in the fight against cancer. If you buy this book, if you buy the Kindle, the audio on Amazon, on Barnes & Noble, anywhere. Money is going to fight cancer. We need your help. Please grab a copy of Stupider People Have Done It. The second thing is I want to work with you. I want to amplify your brand. I want my agency to help you get more customers, more subscribers, more whatever. I want to partner with you in so many different ways. All you got to do is go to jschwedelson.com, jschwedelson.com. And then there's a button in the upper right-hand corner that says partner.

13:05Fill out that form. Let's work together. Let's do stuff. And if this wasn't the worst podcast you've ever heard, It would be amazing if you follow the show, if you leave a comment, if you tell your friends about it. I appreciate you and go out there and crush it.