This week I received a comment from a client that said a line in my copy “sounds too much like AI”.
Which hurt. It hurt because the line she was referencing was a line I wrote myself with pen and paper in my garden. I actually wrote the line a couple months ago for a different set of ads, and I was including it here for a bit of continuity (and because I like the line).
And since I got this comment, I’ve been obsessing over two things. The first is why she thought that line sounds like AI. And the other is why it worked two months ago, but not now.
The second one I think is pretty simple. I originally wrote the line as a headline on an image, with a subhead under it. And this time it was the first line of the intro copy. And taking a step back, I can see that it doesn’t quite fit in the same way. It was a pithy marketing headline where the reader expects context. She was right to want it changed.
But the “sounds too much like AI” comment still bothered me.
It’s not just a copywriting problem
In pretty much every creative industry right now, there is a problem where “feels like AI” is thrown around a lot, both accurately and inaccurately. And when you understand how the technology works, this was always inevitable.
Cartoonist Vig Saminathan posted about this on Reddit last year. He’d posted one of his comics, only for it to be removed for being AI-generated. But it wasn’t AI-generated. He even shared links to videos and drawings that prove it wasn’t AI.
The problem is that the LLMs that produce these images, videos, written content, etc are all trained on existing media. The only reason they can generate a cartoon strip is that it’s been trained on hundreds of thousands of hand-drawn cartoons.
So of course some comics will “feel like AI” because that’s what AI is ripping off.
This wasn’t as much of an issue early on. But we’re being exposed to SO MUCH AI-generated content all the time. We see AI-generated writing on LinkedIn, social media is full of AI slop images, and we’re even starting to hear AI-generated music on the radio.
The more we’re exposed to it, the more confident we feel in identifying AI ‘slop’. Which is a useful skill to have. But this is what leads to the “feels like AI” issue with human-created content.
The infuriating case of the emdash
I’ve used the emdash since I was a teenager, but I didn’t know what it was called until I became a freelance writer 8 years ago. Mostly because an editor mentioned it in a call, and I had to calmly but quickly google what she was on about.
It’s a useful grammatical device — giving the writer a way to take a pause or present a thought mid-sentence.
And it’s always been one of my favourite ways — alongside the ellipsis and comma — to control the rhythm of my writing.
So it was very frustrating when it became the default GOTCHA for AI-generated writing. I’m now specifically asked by clients to avoid emdashes, which is infuriating. My writing is forced to be worse because people decided it was sure-fire evidence of AI-generated writing.
And, for the record, it’s just not. Go grab any book written by an even semi-competent author, and see how long it takes you to find an emdash. I just tried it myself to make sure I wasn’t setting myself up for failure, and it was on the first page I turned to in Elmore Leonard’s Riding the Rap, and on the second page I turned to in Charles Dickens’ A Christmas Carol.
But I do understand where it came from. Yes, I’ve used emdashes for ages, but I’m a professional writer. And most books are either written by professional writers, or edited by professional editors.
The problem is that the average person in the 2020s don’t use emdashes. So when they started showing up in people’s school essays, social media posts, emails, etc., it’s fair to deduce that they probably did use an LLM.
And why does an LLM include emdashes? Because it’s trained on Elmore Leonard’s writing, Charles Dickens writing, and maybe even a bit of my writing.
Copywriting is especially interesting
So when you factor all this in, it’s very easy to see how we get to “sounds too much like AI” from a client. LLMs have been trained on over a century of ad copy. And if you look through the history of copywriting, things have changed much less than you’d imagine.
When I did my copywriting diploma just a few years ago, it taught the AIDA (Attention, Interest, Desire, Action) copywriting model. And if you go look at any website, PPC ad, billboard, there’s a strong chance you can see this formula running through it.
In his bestselling book DotCom Secrets, Russel Brunson repackaged the AIDA model as Hook, Story, Offer (cleverly combining interest and desire to make it a bit snappier). The same method for a new audience.
AIDA was invented by Elias St. Elmo Lewis in 1898. That’s a lot of training data sharing the same principles.
So when we look at an ad and it feels like AI, why is that?
This is what I’ve been pondering on the most, and I think the truth is that it will vary from person to person. To some, it will be that it feels too polished. To others it will feel too corporate. Or maybe it just looks like how you expect an ad to look. Short headline, slightly longer subheading, then longer ad copy building interest until you deliver the CTA.
Maybe it’s that the body copy is spread across a few standalone lines rather than one longer paragraph. We do that for easier readability and creating rhythm. We’ve done that for years, so of course LLMs do that now, too.
It kinda doesn’t matter
This article might read like I’m defending copy that could be misidentified as AI-generated. But honestly, I’m not.
Right now, the companies I work with will have a more sensitive AI radar. They’re marketing professionals who are paying good money for good copy, so if their radar beeps, it’s fair to say something. It’s fair to ask for the copy to be changed.
As time goes on, it’s safe to assume more people will develop that same radar. The same things will make them assume something is AI-generated. And studies have already found that this can hurt trust with your brand.
So it sort of doesn’t matter how ‘good’ the copy is. If someone looks at it and immediately assumes it’s generated by an LLM, it’s not good enough.
As copywriters and marketers, it’s now on us to work out what we do instead. Maybe we move away from traditional headlines. Maybe we prioritise a less polished style of writing over readability and standard copywriting beats. It might mean writing weaker copy that just feels more human.
Whatever we do, we can only do it until LLMs start doing it, too.
Then it’s back to the drawing board.