Writing "Tells": is it actually AI? 🤔
Beloved Moot @umber-cinders did a really in-depth and researched overview of common "AI tells" in writing (which I highly recommend you read here) but I had some thoughts I wanted to add without derailing her post, since my own points are sort of tangential to hers.
I've observed a growing preoccupation in writing communities to identify the definitive "tells" of AI, and it's not hard to see where that urgency comes from. Gen AI is basically a big red button for lazy, incurious hacks to push whenever they want the attention and prestige of creating something that normally requires pesky investments like "time" and "effort". AI's function is to plagiarise, its philosophy is to end human thought and creativity and discipline. It is a crisis, I think, so it is important that we all get very good, very quickly, at sniffing out those thieving robots and denying them the thing those "author's" desperately crave: our attention.
That being said, there's an obvious problem at the core of any effort to define the hallmarks of AI, and that is because human writers have these "tells", too! The em dash, the emphasis in thirds, the decadent, beautiful subordinate conjunctions you'll pry from my cold dead hands - these are all common writing conventions. It is because they are common writing conventions that the AI has stolen them! And I think it's worth asking ourselves what it will do to the craft of writing if we allow these things to become "AI tells" in the public imagination. In my opinion, we are paving the way to a paranoid writing community who becomes more concerned with conveying human-ness in their writing than conveying their actual creative vision. I don't want to live in a world where writers are scalping their own work to avoid any potential accusations of AI, all the while the computer giddily vomits up em dash after em dash because that beautiful little line belongs solely to it, now. So, should we stop trying to assign the AI any "tells" at all?
Well, no, I don't think we should ignore the very obvious habits of the slop machine. I think Umbers' post is very valuable, actually, in compiling as much info as it does, but it struck me as a post that is particularly valuable to writers. You see, I don't doubt that a practiced writer, or a widely read reader, would find great utility in a list of obvious tells, because we know what writing normally sounds like. If you spend all day looking at zebras, then a donkey painted black and white is going to stick out like a sore thumb. But to a casual reader, or worse, a Booktok Drama Influencer, we're at risk of them reading about these tells and going "Look! It has black and white stripes! That zebra is AI!"
And I have witnessed this happen. Back when I still used TikTok (forgive me), I saw the absolute dogpiling of an author for "using AI" to write the logline of her novel. The argument was that the sentence was dry and formulaic... but it was a logline. They are supposed to be dry and formulaic, because they have to summarise the story in one sentence. If you're unfamiliar, loglines almost always follow this structure: "When [character] experiences [inciting incident], they must [overcome central conflict] to [achieve goal]." A writer who can encapsulate their story in a clean, tidy logline is a good writer. But to these Booktokers, because I imagine they've never written a creative summary themselves, this formulaic structure was an obvious "tell" of AI, and they harassed an indie author relentlessly over it.
This is the risk, I think, in trying to definitively nail down any "tell" for the AI. Because it steals from human works, it will always have traits overlapping with human authors, and unfortunately, because the type of people who use GenAI to make creative work have no scruples or integrity or personal pride, the casualties of these "tells" are overwhelmingly human. The prick claiming to write a novel that's actually AI is rarely going to be shamed into silence by pointing out these tells, but the human author who spent 18 months pouring their heart and soul into a manuscript will be discouraged. Furthermore, I think juvenile writers are the ones most vulnerable to these "tell hunts." Writers who are just starting the craft, who haven't developed an innate skill yet, and who are themselves overly reliant on convention to elevate their work may see people pointing to the "obvious" signs of AI and retreat away from pursuing their passion. I used to work with teenagers in the theatre, and what struck me is how much their clumsy first drafts read like AI pieces (and yes, I know they wrote them themselves because we wrote as a group in "roundtable" sessions). This is because these young writers were desperate to seem more skilled than they were, so they relied heavily on imitating the styles of the playwrights they admired. This in itself is not plagiarism; it is the foundational steps towards skill. Absorbing the styles of those that came before until your True Voice has a chance to develop. I think it's an important part of being an Older Artist, to guide the juveniles into the safe middle ground between mimicry (which is how they learn) and their true artistic Voice (which is who they are).
And, if I may, I'd like to suggest an addition to the list of "AI tells" that I rarely see included in these online discussions. The AI has No True Voice. This is where I think Umbers' post truly shines in its value, this is why I think the tells there-in stand out to so many fellow writers: The AI has no voice, it only has convention. Human writing has an ineffable earnestness within it; even the most polished, carefully crafted pieces have some residue of the Author dripping through the cracks. Some imperfection in the prose, some impulsive inclusion in a detail, or a character reaction. To borrow a tell back from the AI, human writing has intent - specifically, a human author knows where the story is going, knows how they want you to feel, knows when they need to mislead you, and all those secret intentions bleed into the prose at least once in every story. If, like me, you're really good at guessing plot twists, it's probably because you have a knack for catching these moments of authorial voice, a moment when the strings of the puppet flash visible in the light. AI writing does not have this voice.
When you're subjected to an AI piece, something you may not notice at first (but trust me, once you spot it, it's impossible not to notice) is how sterile the sentences are. There's no subtext, no artful manipulation of emotion, no excitement, no attitude! Even when the AI has been instructed to write in a more "human" way, it often comes across as "hello fellow kids," specifically because it lacks all character. This is probably the hardest AI tell to get good at spotting, but it's also (in my opinion) the most definitive trait of the computer. Even the aforementioned juvenile writers, overly reliant on convention, are so flushed with earnestness that while their work might be clumsy and uncharismatic, it is saturated with character.
And so, I agree with Umbers' post, and I think it's useful to have a catalogue of AI tells, but I also think it's important to look for those tells in context. It's not even that AI uses these conventions, it's that it overuses them, because it lacks the ability to write impulsively, or revise thoughtfully, and so all that's left is sterile formula. So I would ask you that the next time you think you spot an AI tell in someone's writing, zoom out a little. Is it that they've used a tell, or is it that the tells are so frequent, you're reading a repetitive, voiceless, summary of events that's only loosely disguising itself as a creative piece? Instead of looking closely at the black and white stripes, take a step back to see the full animal. It's far easier to spot the donkey that way.
