AI is all around us these days, so it’s bound to leave a trail of breadcrumbs wherever it goes, and writing is no exception. By now, everyone knows that em dashes are somewhat of a trademark of AI and can make a piece of writing look suspiciously automated, especially when they appear again and again throughout the text. These notorious dashes are now among the easiest signs of AI writing to spot, to the point that newer AI models have become much better at avoiding them altogether.
Many other signs still manage to make it out alive, however, ending up on blogs and websites for everyone to see. Certain sentence patterns and wording choices are much easier to miss and can make copy feel machine-written even when there are no obvious signs in sight. Here are five less obvious signs of AI writing I recommend you check for, so you know what to steer clear of before your next piece goes live.
1. Its sense of humour (or lack thereof)
This is a personal pet peeve of mine, which is why it takes first place on my list, though admittedly, it isn’t the most obscure sign on here. Ask AI to tell a joke and you’ll quickly see where the problem lies. Any humour it can manage on its own is usually formulaic or misses the punchline, and I mean… really misses it. To give you an example, I asked a chatbot to come up with a joke, and this was its answer:
“Okay. I told my keyboard I needed space. Now it won’t stop giving me the silent treatment.”
Good to know comedians aren’t getting replaced anytime soon. If it does manage to churn out something funny, chances are it’s drawing on patterns it has seen before, with very little originality of its own.
The beauty of humans is that they (at least the funny ones) can find humour in nearly anything, be it an awkward conversation or a small or big-sized inconvenience. Add in a talent for writing, and we get brilliant work that can make life’s heavier moments feel lighter, find humour in the human condition, if you will. AI may be adept at learning, but it rarely colours outside the lines, and for our sakes, I hope it never learns how to.
2. It has no personal stories to tell
Picking up from the previous sign, another giveaway is AI’s inability to share real stories. AI can come up with a personal story if you ask it to, but there is no lived experience behind it, so you get something close to the uncanny valley feeling that human-like figures can evoke, only in written form. Obviously, an LLM has never accidentally spilt iced coffee all over itself that one time at Starbucks or felt a genuine sense of awe looking at the works of Van Gogh.
Personal accounts of that nature are difficult to fake convincingly because anecdotes usually include oddly specific details and random tangents. Again, whatever AI comes up with will either draw from patterns it has seen before or feel too immaculate to be a real story. Humans are natural storytellers too, something a non-sentient machine cannot truly emulate. So, when an article reads perfectly well on the page, but there is still something slightly off about it, the absence of these human details may be the giveaway.
3. It cannot help itself with a rhetorical question (or multiple)
Rhetorical questions can work well when there is a reason for them. AI, though, is guilty of using them as a writing crutch, dropping one in whenever it wants to introduce a point or create a little drama. Ever noticed that? There it is.
You will often see something along the lines of, “So, what does this mean for you?” followed immediately by the answer. The same thought could just as easily have been written as a normal sentence without these unnecessary theatrics.
Human writers use rhetorical questions too, of course, but they exercise more restraint. When several appear in the same article, especially in that predictable question-then-answer format, you can bet they leave another little breadcrumb pointing back to AI.
4. It really likes to play favourites
AI seems to have more than a handful of phrases it prefers not to part with, at least until an update comes along and changes its habits. “This matters because…” is one of the biggest offenders. Then there are old favourites like “Here’s the kicker”, “to put it simply”, “the key takeaway here is…” and “at its core…” There is a strange fondness for making things “sit” somewhere, too. For instance, a feature “sits at the heart of” a product, a principle “sits behind” a decision, and suddenly everything is sitting around.
There are a few other familiar constructions too that you will find scattered throughout AI copy in many places. “Whether you’re a beginner or a seasoned professional…” crops up a lot, as does the dramatic “It’s not X, it’s Y” setup. “Not only does X, but it also does Y…” is another regular, and the rule of three, as in “the update is faster, smarter and easier to use,” is a surprisingly persistent nuisance. AI happens to favour phrases like “in today’s fast-paced world” as well, even when the subject hardly calls for that Peter Jackson-level scene-setting.
None of these expressions screams AI on its own because we use them too. The clue here is seeing the same wording turn up repeatedly across completely unrelated pieces of content. Human writers, no doubt, have their own verbal proclivities, but we tend to notice when they become overkill and try to break away from them, whereas AI often returns to the same collection of stock phrases that appear to be imprinted on it.
5. It makes an event out of the mundane
AI has a habit of making perfectly ordinary things sound as though they deserve their own documentary. A minor software update becomes a “significant shift in the way users interact with technology.” A café adding a new sandwich to the menu becomes “a reflection of its continued commitment to offering customers more choice.”
LLMs often inflate the importance of arbitrary details by tying them to grander themes. You will also notice something called copula avoidance, where AI swaps a simple “is” for “serves as” quite a lot, somehow making the subject sound bigger than it actually is.
Human writers can overstate things too, but usually (and ideally), there is a bit of self-awareness behind the exaggeration. Watch for language that turns every change into a milestone or every small detail into evidence of something much bigger. When the writing seems far more excited about the subject than any reasonable person would be, another tell may be staring you in the face.
Now HERE’S the kicker
Any one of these signs, including the more obscure and the painfully obvious ones, can appear in human writing too, so spotting a rhetorical question or an overzealous sentence here and there does not automatically mean AI wrote the piece. After all, the crime scene starts to take shape when several suspicious clues appear in the same place.
I think the better approach is to look for clusters instead of treating one phrase as a smoking gun. When I’m proofreading work, I pay more attention to patterns that start to make the writing formulaic. If stock phrases, inflated claims, suspiciously polished anecdotes or inane phrasing start piling up, consider it your cue to go back through the piece and strip out anything that feels too familiar. A final sweep like this keeps your own voice intact and gives readers far less reason to question who wrote it.