What Are AI Tells + How Can You Avoid Them?
People have come up with a lot of AI tells that they think immediately give away AI generated text. Whenever they see these things, they think ugh AI slop and scroll past. Even if you don’t use AI in your writing, if you use these “tells,” people will probably accuse you of using AI and stop reading your work.
The interesting, and dangerous, thing about AI for a writer is that these models learned how to write from existing language and lessons we as humans once learned from too. Large language models (LLMs) basically study a ton of written information, analyzing it to learn how to mimic human written word. They learned how to write from us, and they write the way they do because we do too (or we used to).
So what are the specific AI tells in writing that you should avoid? Well, I often think that these “tells” are less accurate markers of AI writing than, for lack of a better word, vibes, but here are a few.
The em dash
Specifically, the overuse of the em dash. Because let’s be real, all of these things predate AI and were used + taught for centuries before generative AI came along and made it a red flag. As a writer who was taught how to use an em dash and has read a fair few classic novels in the literary canon, I’m here to crush your dreams if you think a single em dash in an article or paper means it was written entirely by AI.
But if it’s riddled with em dashes, making it look almost like someone had the strikethrough setting on when they were typing, yeah, I’d say that’s probably AI. One or two properly used ones here and there, though? I’d venture to say it’s probably not AI… That’s just me though.
The rule of threes
As I write this in May of 2026, people have chilled out a little about em dashes, although they’re still broadly considered a potential sign of AI. Now, people have found a new target. They’re starting to recognize that AI writes in threes. A lot.
It is literally a rule in writing that groups of threes are more satisfying and memorable than other numbers. This goes back to ancient Roman times. AI has learned this pattern and decided to run with it, sometimes using it multiple times in a single paragraph. If your brain especially likes lists of 3, try to keep it limited to just a couple times in any one piece of work, but don’t be too hard on yourself. Instances of it here and there are bound to happen in human written text – it’s a habit at this point.
Think “tall, dark, and handsome” or “blood, sweat, and tears.” If you look for it, you’ll see it everywhere. Even perhaps in places you don’t want to if you’re convinced it’s a dead giveaway for AI text (like your own writing and speech patterns).
One thing about people is that we’re often blissfully unaware of our own (perhaps unfavorable) speech tendencies. The valley girl “like”s and teenage boy sounding “bro”s, “dude”s, and, most recently popularized “bruh”s come to mind.
I think the rule of threes and, to an extent, contrastive negation are like those. We don’t perceive ourselves as using it – at least not much – but in reality, we do. For contrastive negation (the “it’s not X, it’s Y” sentence structure) think of trying to explain why you’re upset over a friend canceling plans last minute to someone who just doesn’t get it: “It’s not just that they canceled, that’s fine, it’s that it was so late I was already getting ready by the time they called.”
It’s also not always so standardized in human speech and writing as it typically is in AI, but it’s still a contrastive negation. Some more famous, streamlined examples include Bond’s martini recipe, “shaken, not stirred;” JKF’s most well-known line from a presidential address “ask not what your country can do for you – ask what you can do for your country;” and, a classic, “It’s not a fight/battle/war. It’s a massacre.”
More flags on the field
People have also spotted specific words that they feel are 100% red flag signs of AI generated writing. But are these as accurate + static as some may think?
One thing we have to consider is that AI is always learning and evolving, just like us. And whether we like it or not, these systems are influencing us and our behavior.
On the Magical Overtinkers Podcast with Amanda Montell, she spoke to Adam Aleksic about the Algorithm. He brought up examples of AI language becoming ingrained in our actual human culture. On TikTok and other social media platforms, you can’t write or say “die,” “killed,” “murdered,” “suicide,” and other so-called upsetting words we use to talk about death. So, youths have started to unironically say “unalive” in place of these other words because it’s become commonplace for the more chronically online – even in offline conversation.
Aleksic also brought up the word “delve.” He talked about how ChatGPT (at the time, again, these things change quickly) disproportionately used the word “delve,” leading a ton of people to point out academic papers + studies that were written with AI. It doesn’t stop there though. Aleksic quickly followed that up by saying that more recent studies have shown that people *are* starting to randomly use the word delve in conversation because of the increased use of the word by the AI we’re all constantly being overwhelmed by.
The biggest question for me is to what extent is it influencing us, rather than us influencing it? People have this weird tendency to mirror each other’s behavior, mannerisms, body language, and, yes, speech. Like when you hang out with a friend who swears like a sailor and you start swearing more than you typically do. AI does that too. The mirroring, not the swearing so much. As we saw with “delve,” sometimes we mirror AI right back. We’re in this icky cycle of it mirroring us and us mirroring it and it mirroring us even more.
Are we feeling rooted + grounded?
Words like “rooted,” “grounded,” and “quietly” feel, to me, like more current examples of the “delve” phenomenon. Except now we’re more aware of what’s going on, so instead of using them more, people seem to be trying to avoid them at all costs so as not to be accused of using AI. I admit that I don’t – to my knowledge – use these words frequently, but If I’m talking about yoga, I’ll likely still use the word “grounded.” For ideas like home and belonging, I may use “rooted” or “roots.”
AI learned it from somewhere, and that source was us. Basically, I’m saying you shouldn’t always make split-second judgments about whether writing is AI or not just based on a single word out of hundreds. Context is important.
When dealing with a system that changes + evolves so much so quickly, it’s hard for me to justify changing my own speech + writing just to avoid these one-word “tells” of AI that change as fast as we can publish content. Every day more words + patterns are added to the list simply because AI suddenly started to use them a little too much.
The point
If you’ve been waiting for the “so what” of this blog, congratulations on finally making it. If you’ve skipped down to this section without reading the rest, touché. The TLDR of it all is this: if you don’t want people to accuse your work of being written by AI, just write like a person. Use emotion and feeling and slang and things that may not be 100% grammatically correct but are 100% human.
In a more corporate, workplace setting, this can be more difficult than in an admittedly pretty rambly blog like this, but it is still possible. There are sentences that you read and feel in your bones that a real person wouldn’t say. The text comes across hollow + unfeeling. Questioning these “tells” and digging a little deeper into the why of it all isn’t me saying they aren’t accurate or shouldn’t be used as ways to spot AI – it’s my way of asking you to look deeper than just a single word or punctuation mark.
As someone who writes a lot as part of my job, I am absolutely certain that I’ve used *literally all* of the AI tells I mentioned here in my writing at some point. It doesn’t matter that I don’t use AI if these “tells” are in my writing and that’s all people go off of. They’d have stopped reading this ages ago when I first used an em dash. They’d never have given my entertaining prose (direct quote from my high school English teacher) a chance.
While I definitely agree that excessive use of these things or their presence in contexts that don’t really make sense can be markers of AI, I don’t think we should be completely writing someone’s work off because they used a list of three adjectives in a row one time. These tells should be taken into account along with the rest of the work, not seen as the end all, be all.
Not to brag, but I still don’t think a robot can replicate my voice. Yet. The day that it can is truly the end for us. Not to be dramatic, but every day we do seem to get closer to that point. Let’s not dwell, though.
Should you never use AI?
Honestly, I wouldn’t for things that you’re writing. Maybe to help you understand a topic or do very basic “no thoughts, head empty” tasks, but not to generate actual text that you’ll actually use. Even if you’re just using it for “inspiration” and planning to change it all, reading it can influence how you start writing right after. That robotic cadence will be at the forefront of your mind. There are definitely pieces of writing that you can tell don’t feel authentic for one reason or another. Especially if you know someone and the way they talk, it’s not hard to spot disingenuity.
To safeguard yourself if you do choose to use generative AI, make sure you always read through it thoroughly, fact-check it, add your own voice into the text, scan through to make sure you aren’t using any of these classic tells where they don’t make sense, and maybe even read it out loud to help you get a better feel for if it sounds weird if you can’t tell at first. Heavy on the fact-checking though because as we’ve discussed, AI hallucinates, i.e., makes up random stuff *very* often.
Working with a studio (like us) that’s dedicated to producing human work for your content creation instead of ChatGPT may be a little pricier, but it’ll lead to more genuine connections in the long run. You’ll be able to connect to us + your customers/users/clients will be able to connect to you better and more authentically.
The truest, most reliable AI tell? Flat, unemotional, grammatically perfect, monotone writing. AI doesn’t have lived experience, feelings, or heart. My best advice is to make sure that your writing does. If you wouldn’t say “tapestry” or ask “the catch?” out loud, (which I seriously doubt you would do if you’re a human living in 2026 excluding very specific circumstances – like talking about Carole King’s album or speaking to a Disney villain offering to make all your dreams come true) then why would you put it in your writing? Being professional doesn’t mean being robotic + unnatural.
