AI Detection Tools, Watermarks, and Slop Buttons, Oh My!

“We are dead smack in the middle of an AI Witch Hunt right now.”

This was the opening line from one of my favorite email newsletters this weekend, The Digital Contrarian, written by Ryan Levesque.

He’s not wrong.

We’re all sick of the AI-generated slop showing up everywhere, from our social media feeds to the ads we’re seeing, and yes, in our emails.

I wrote about the AI slop problem in the second issue of the Email Marketing Ecosystem newsletter.

Since then it’s become impossible to get away from. Worse, it’s getting harder and harder to tell whether it’s actually AI or not.

Given enough information, the LLMs, like Claude, ChatGPT, Gemini, etc. are now good enough to write something that seems like it was genuinely written by a human. Whether it’s any good is still a matter of personal opinion.

This has gotten to be such a big problem that various platforms are cracking down on AI slop, and the LLMs themselves are actually making changes to their outputs, so humans (and other AIs) can verify how something was created. By a human or by a robot?

Is It Live, or Is It ChatGPT?

In the late 1970s, early 1980s, there was an ad campaign with the tagline, “Is it live, or is it Memorex?” The idea being that Memorex, a brand of cassette tape, was so good, that you couldn’t tell whether a performance was live or pre-recorded. (Yes, I’m old. Get off my lawn.)

Now, the question has become, “Did a human create this, or was it AI?”

More and more, we’re realizing that it was AI. And humans hate it.

To the point that people are being lambasted if anything they produce even has a whiff of AI about it, whether or not AI was used in its creation.

You may have heard about a couple of high-profile publishing deals that fell through after the publishers discovered that the authors had used AI in their writing processes. Both publishing companies were concerned about the taint of AI affecting book sales.

Regardless of whether either author actually used AI to write their books, both claim they didn’t, the stigma of using AI is becoming a thing.

It’s also impacting social media.

I regularly see long, interesting posts about all sorts of topics, politics, history, daily life, etc. come across my feed. A lot of them are shared by friends. People I know in real life, not just online.

Somewhere in the process of reading these posts, I either get the feeling the text is AI-generated myself, or I see a comment that says, “This is AI slop.”

About half the time, I’ll search for the topic or story myself to verify whether the post is real. I usually find some reference to whatever the story is about. However I also find a lot of references indicating the post is AI slop or the story is real, but there are several related, inaccurate, AI-generated posts or memes circulating on social media.

I expect you’ve seen the same thing in your feeds.

FYI, I learned a new term for this phenomenon while I was researching this week’s issue. This behavior of creating a “…mismatch between the reader’s expectation and reality, especially when they unwittingly invest their attention in something with no human thought on the other end…” is called “Claudefishing.”

LinkedIn Adds an AI Slop Button

LinkedIn in particular has become a hotbed of AI slop and accusations of AI slop, because of this very problem.

A lot of people are trying to “be seen” on LinkedIn, whether they’re looking for a new job or trying to sell whatever it is that they sell (present company included). A simple way to up your LinkedIn profile used to be to post regularly and comment on other people’s posts.

As of July 30^th, LinkedIn has added a “Seems like AI Slop” option to their reporting button. You find these options when you click on the 3 dots at the top right of any LinkedIn post.

A list of the LinkedIn post options with "Seems like AI slop" in a red rectangle.

Too many people were using AI to write their posts and to comment on other people’s posts. LinkedIn even had their in-platform “Enhance with AI” feature which helped users “improve” their posts. That’s been removed.

LinkedIn is now claiming that “AI Slop is a top priority for all of us” and they want “real voices” back on their platform.

This is what happens when you become the butt of too many jokes.

Now, people will actually have to write their own posts and their own comments if they want to be seen on LinkedIn. Or at least, that’s the hope. We’ll see how fast people figure out a way around this and which tech platforms help them do it.

Substack Has Added an AI Detection Tool

Last month, Substack added an AI detection tool to its app. Apparently Substack is just as concerned about the amount of AI slop showing up on its platform as LinkedIn is.

They partnered with an AI detection tool called Pangram to integrate this tool into the Substack system. You have to check the article you’re reading to see if it’s AI generated or not.

Three mobile screenshots of the new Pangram tool on Substack, showing how to use the tool. (Please click through to the linked article for more details.)

FYI, the picture of the AI detection tool was taken from a Substack article about the Pangram tool. This is the mobile app version. The quote about Claudefishing was taken from the same article.

I used this tool myself on the web browser. If you go to the dots in the top right corner of your browser window, you’ll see this menu.

The Substack post options with "Scan for AI text" highlighted with a red rectangle.

When you run the tool on a Substack Article, you’ll get the results instantly. (Or at least I did.)

Results of the Substack/Pangram tool, declaring an article "Fully Human Written."

I admit, I tend to follow people who I know write their own articles, no AI involved.

If you’re publishing on Substack, you can test your Article or Note prior to publishing.

As with all things AI, the problem is that this tool, along with every other AI-detection tool, can give false positive results.

AI Companies Are Recognizing This Problem

Last week, Anthropic, the AI company behind Claude, announced that they would be adding watermarks to their AI output.

Here is how Anthropic explains it in their announcement:

What is watermarking?
Large language models like Claude work by generating one word at a time. Each time the model decides on the next word, it chooses among a list of potential candidates, ultimately selecting the most sensible or likely based on the preceding text. Take the sentence “The weather today was cold and…”. The next word is very unlikely to be “sugary.” But it is quite likely to be “overcast” or “grey.” Under most circumstances, it doesn’t matter much to the reader which of these latter two words the model ultimately chooses—the meaning of the sentence is largely the same either way. In cases like this, the choice is settled by a random number.

Watermarking uses low-stakes choices like these—which occur many times over a piece of generated text—to leave a pattern in Claude’s responses. That pattern is undetectable to the reader, but is detectable to anyone who has a key that encodes it. When watermarking is used, choices are still made at random, but the source of the randomness is different. Instead of using an arbitrary random number generator to pick the next word, watermarking uses the key and a few words that come before to settle what word the model should pick. That is, the words that Claude picks are still random, but now, one can check the sequence of words and see if it’s consistent with the choices Claude would make if it was using the key. If it is, one can assign a probability that the text was generated by Claude.

So basically, if you use Claude to write anything, social media posts, emails, reports, etc., someone or something, like an AI-detection tool, that has the key to the watermark sequence, would be able to detect and report the probability of whether it was written by you or by Claude.

ChatGPT is headed in the same direction. They’ve already inserted a similar watermark feature into all of their AI-generated images. These are part of the meta data and sync ID in the image itself.

By the way, these companies aren’t doing this to benefit humanity. They’re doing it because they have to.

The EU has instituted the EU AI Act, which is forcing the AI companies to add these watermarks. They see the writing on the wall. The AI companies are making this universal because they expect other countries to follow suit.

Again, humans are pushing back against the onslaught of AI slop.

Go us.

There’s a Simple Solution: Write Your Own Words

“But that’s so hard, Tanya!”

I know, I know…

But, as you can see, your people want real human connection, not regurgitated robot speak.

I’m not saying AI isn’t useful. I admit here regularly that I use AI myself, as a brainstorming tool, and for research. I’ve even written an AI Manifesto, so people know how I use AI.

I also give every issue of this newsletter to Claude and ask it to write social media posts, based on that content. I then edit these posts heavily to be in my own voice. Though I may have to rethink this strategy, based on these changes.

The outcry against AI is as much a symptom of a problem as it is the railing against the world’s current super villain. The problem is our lack of human connection. We crave it. So anything that feels artificial is gross.

The biggest takeaway from all of this is simply that people want to connect with real people.

So how are you going to do that, with or without AI?

How will you connect directly and authentically with your subscribers. Your current and potential customers?

I recommend a weekly nurture email. (You knew this was coming, right? Of course you did.)

Like this issue of the Email Marketing Ecosystem newsletter. Written 100% by me, for you, so we can connect over ideas and information that is important to all of us as small business owners.

If you need to use AI to help you write your weekly nurture email, go ahead and do it. Just follow these two simple rules.

Be honest about it. State somewhere on your website or even within your emails that you use AI to help you write.
Don’t let AI do the job by itself. Give it your ideas, let it give you a messy first draft, then go through it and make that draft your own. So you’re sharing your words, in your voice, with your people.

Several members of my Email Writing Accountability Group use AI to write their emails. One of the benefits of being a member of this group is that we go through those emails together and make sure the AI output makes sense to their readers and actually sounds like them, before they hit “Send” every week.

If you’d like some help finding the right words, whether you use AI or you write everything yourself, please join us for a free session of the Email Writing Accountability Group.

You’ll see exactly how the process works and how simple it can be to get an idea out of your brain and onto the page, where it can help your subscribers solve whatever problem their having.

That way you’re serving them every week. You’re building the relationship you’ve started with them as soon as they joined your email list, and you’re showing them why you’re the right choice when they’re ready to buy.

Plus, you’re forming a community of people who are there to support you and your business, as well as each other. Everyone gets more human connection and we all thrive together.

YES! HELP ME CONNECT WITH MY SUBSCRIBERS

Finally, this week’s picture is from the Butterflies Alive exhibit at the Santa Barbara Natural History Museum. I went last weekend with a friend and it was wonderful!

Two Postman butterflies, mating (one right side up, one upside down) sitting on a purple sage flower.

Want to Get The Email Marketing Ecosystem Newsletter Delivered Directly to Your Inbox Every Thursday?

-->