It’s not an AI tell, it’s a cliché

(Claude wrote that title. The illusion of a sense of humour.)

AI "tells" shouldn’t be about catching someone out. It's not, "Oh look, I caught you using AI to write your blog post," (with the implication that the spotter is savvy and superior). The real problem with AI slop isn't that it was written by AI. The real problem is bad writing, and writing that’s littered with clichés is bad writing.

There’s nothing wrong with the words “quietly” or “shift” or “delve”. There’s nothing wrong with the sentence structure "It's not X, it's Y" or with em-dashes. The problem is overuse. Because AI writes at the speed of light, it also creates clichés at the speed of light.

Now, people who aren't using AI in their writing are getting unfairly accused of using AI because they've used some of those so-called tells. What's happened here is that AI has quickly created a cliché — an overused word, expression, or sentence structure — and, unfortunately, a non-AI-using writer has unknowingly used something that’s quietly become a cliché.😉 It sucks. But that's the reality.

There’s no such thing as an AI tell

So, there’s no such thing as an “AI tell”, just clichés that AI rapidly created. Of course, if a piece of writing has so many of them that you can barely read between them, we can reasonably assume it was AI-written. That’s the worst kind of AI slop. But a few “clues” don’t mean someone used AI. Most tools for detecting AI writing are notoriously poor and cannot reliably detect whether some piece of writing was AI-created or not. (And, let’s face it, unless someone reads a lot in the online business and marketing space, they can’t possibly keep up with all the newly-coined clichés.)

The problem isn't people using AI to write. So we don’t need to try to “catch” them doing so. The problem is, and always was, bad writing. Cliché-ridden AI slop is bad writing.

What actually matters in nonfiction content

I personally don't care if I read an article or blog post on the internet about, say, SEO or content marketing or social media, and it was written by AI. If it was a good article, it was a good article, and that's fine by me. I've got nothing against people using AI in their nonfiction writing process. I honestly don't care if it took you ten minutes to write with AI or ten hours with a quill and parchment.

What matters to me is the finished product. Is the finished article any good? Is it true? Am I learning something? Is it clear? Is it engaging? Does it have interesting insights? Does it provide evidence? Is the reasoning sound? Does it avoid clichés and unhelpful metaphors? Is it succinct?

It doesn't matter if it was written by the CEO, an intern, a freelance writer, or AI. It doesn't matter who — or what — produced it. What matters is simply the quality of the article at the end. (For the record, I don’t think AI is capable of writing a good article by itself. But it also doesn’t have a monopoly on crappy writing.)

Fiction is different. Fiction is art.

Fiction is a whole different category for me. It's art. AI should not have its sticky fingers anywhere near it.

It matters to me whether an author got AI to generate a paragraph in five seconds or whether they struggled with it for five days. Even if the writing in both cases is good, I want those words to come from the author. When I write my own fiction, it's important to me that every single word, every single comma and em-dash, comes from me. AI does not get within a gigaparsec of my fiction. Part of what's important to me, in art, is not just the finished product but the connection with another human being — the artist.

If I read a novel and I enjoyed it, found it beautiful, was maybe even moved by it, and then at the end discovered it was written by a robot — that, to me, would be deeply disturbing. I do not want to be moved by a robot. I do not want fake connection with a robot. I do not want fake insight into the human condition.

The experience machine

This reminds me of a famous thought experiment from way back when I did philosophy. It’s called the experience machine and was invented by philosopher, Robert Nozick.

The idea is a life-support machine that you could program to give you certain experiences. You could step inside for the rest of your life and enjoy the illusion of your ideal life. You could experience the illusion of being an Academy Award–winning actor, or an athlete winning five Olympic gold medals. You could experience the illusion of falling in love with your ideal person, getting married, having children — you could have it all. But none of it would be real. You didn’t really achieve any of those things, you didn’t really have those relationships, because those people weren’t real. You were just kept alive in the machine with a constant illusion.

Would you choose to go into the experience machine?

Some people would. For me, it’s true horror. It doesn't matter how good the experiences are — I don't want fake. I want real. I want real accomplishments, real sunshine and real rustling leaves, real love with my real family, real relationships with real humans. (I'm getting teary as I dictate this with Wispr Flow, because it means that much to me.) Real connection with real people is part of what makes life worth living. If we go around fake-connecting with robots, then what the fuck have we done?

So my point in all of this is that we’re mistaken in trying to “catch” people using AI. That misses the point. The point is that we want — we need — good writing. And that includes writing that avoids clichés (which is what an AI “tell” is: a rapidly created cliché).

Also, when you write, it’s crucial to be clear on what you’re trying to do. Are you conveying information or ideas? In that case, do it as clearly as you can. Be truthful, be specific, be engaging, don’t plagiarize, don’t overuse adverbs, don’t use unhelpful metaphors. Use AI if you want, but, whatever you do, make sure the finished piece is good, true writing that’s worth reading. These are the same rules that have always applied.

On the other hand, are you creating art? Is human connection part of the point? Then, for the love of all that’s human, please keep AI out of it.

P.S.

My process for this article: I typed and dictated a bunch of disconnected, loosely related thoughts into TextEdit. Pasted them into Claude and told it to organize them into an article while keeping as much of my language as possible. I also asked if it had any questions before it started. It did, and I answered them. It produced draft one. It was still very rambly with too many disconnected points crammed into one piece. I asked Claude what it would recommend cutting. They were good suggestions, so I got it to produce draft two. I also asked it to come up with some alternative titles. I picked my favourite, and pasted the whole thing into Word. I printed that on paper, which you can see in the attached images. I made lots of changes with a pencil, and then edited my Word doc (making a bunch more changes along the way) to produce the version I uploaded to my website, where I made yet more changes. I can see how it still should be two separate articles (do you agree?) but I’m impatient to express my thoughts, so I’m going compromise on that.

P.P.S.

When I say “use AI if you want” for nonfiction writing, this does not apply to students. I think it’s essential that students learn how to write well without AI before they learn how to write well with its help. But that’s a whole nother article. The issue of plagiarism and AI is also an article for another day.