Current AI Tells
This is an evolving list that I will be adding to or altering as I notice things shift.
THIS IS NOT INTENDED TO START WITCH HUNTS.
As an editor, I am frequently asked to justify and identify AI tells by people trying to determine if something is AI-generated or not.
Obviously, the presence of one or two tells does not necessarily mean something is immediately AI. We need to use caution in this time to not just throw the baby out with the bathwater.
However, over time, the tells will add up, and something that is pure AI output without a great deal of human editing is pretty easy to flag once you know what you’re looking for.
However, since I will be teaching an Editorial Freelancers Association course later this year on the subject of identifying AI, I figure keeping a running tally of things I notice pinned here may be a useful reference for others who are trying to determine if something is or isn’t AI.
Finally, you will notice that a lot of this is more editorial than just keyword-based. I have been a professional editor since 2008, and I’m autistic. Pattern recognition is my thing. While there are certainly individual words and phrases I will identify as potentially AI keywords, those change. The vibes often do not, and those are frequently model-agnostic. While each model has a distinct voice, they share many similar qualities.
For a more robust and complete analysis of AI text overall, please see my article here. I am not going to rehash the same points I made in this article; this is intended to be just a list of tells.
The List
Writing Styles
Universal
Complete absence of human writing errors (typos, inconsistent grammar, natural mistakes) with the exception of Claude liking comma splices
Frequent overuse of “it’s not x it’s y” patterns and phrasing
Heavy emotional beats that feel out of place or like every single thing has to have some kind of unnecessary emotional weight to it
ChatGPT likes to do one sentence paragraphs like it’s writing a poem except not
Three-part lists everywhere (everywhere)
Over-explanation of things to try and give context unnecessarily
Metaphors that sound weighty but don’t actually mean anything when you think about them
Incredible evenness of tone and sentence length without much for variation the way a human writer does it
Generic specificity (details that sound precise but aren’t verifiable or things most people would notice or could notice)
Em dashes (though humans use them too—a lot!)
Consistently smooth transitions rather than conversational or even jagged ones
Fiction
It has a set number of character and archetypes it can really draw on without more intense guidance (most middle-aged women will be intensely motherly, for example)
Plot holes and inability to remember things that happened more than a few paragraphs ago in the same narrative
Lots of interiority and hedging without real depth “he found himself wondering” or “something about it caught him”
A lot of small, ephemeral moments that make the prose seem a little dreamlike without adding much to the narrative
An unusual number of sensory details and specific ones
A liminal feeling to the prose where characters often appear and disappear into the fog and sometimes take actions as if in several physical locations at once or as if physical space doesn’t matter
Lots of generic fillers with ambiguous names (like calling things “The Council”)
There’s a lot of passivity as if the characters are being acted upon or receiving rather than acting
Tends to try and resolve all tension quickly rather than allowing for open threads or uncomfortable things left uncomfortable
Everything is arm’s length in a weird way
Stories are highly linear and simplistic
Instances of characters having meta knowledge all over the place (keeping characters straight is a huge problem)
Complete lack of subtext in the dialogue (though this can sometimes be true of neurodivergent writers, so use caution here)
Problems tend to resolve immediately in favor of the main character almost without exception
AI has consistent issues understanding injury severity and cannot typically keep details straight regarding it (they’ll change sides, change severity, and so on conveniently to the plot)
Severe difficulties with timekeeping (time is not a constant; it will reference it and than change it immediately)
Excessive focus on numbers of things
Narrative frequently focuses around it “costing” characters things, such as “the words cost her something in the saying”—particularly when it makes no sense that they have cost (tying back to unusual emotional weight)
Nonfiction
Excessive hedging statements: “while x is true, y is also the case” — hedging is acknowledging all sides rather than expressing fallibility, however.
Heightened use of therapy speak
Significant confidence at all times without much room for error
Frequently calling out things that are “real”
Lack of personal anecdotes and personality in the writing itself and sounds “samey” to all other AI writing
Lack of emotional depth and change, even when addressing challenging subjects
Conclusions that re-state the introductory theory without adding anything or displaying new information
Excessive reassuring of the reader, often assuming that the text knows what the reader is feeling
Reframing basic perspective shifts as revelatory somehow, like, “what if rest isn’t lazy, but necessary?”—while this can be absolutely true of human writers, it’s written oddly by AI
Strange aphorisms that imply some kind of wisdom without saying anything of depth or meaning “growth happens at the edge of comfort”
Over-acknowledgement of challenges and move to provide excessive comfort (particularly in self-help)
Always presenting both sides of a thing even when both sides do not need to be shared
Closing on an uplifting note, regardless of subject matter
Avoids repetition even when used to underline a point
Marketing
Lots of rhetorical question openers or things like “what does this mean for you?”
Clickbait-y headline pivots like: “But here’s the thing/kicker, etc.” or “here’s where it gets interesting.”
False conversational tone by using one-word sentences for emphasis, particularly things like “seriously,” “that’s it,” or “simple.”
More frequent use of “it’s not x, it’s y” framing than even other methods.
Lists tend to follow the rule of three (not exclusive to AI)
Second person “you” used to create a sense of false intimacy with the person reading it
Lots of corporate hype language like: unlock, elevate, transform, supercharge, empower, seamless, effortless, game-changer, next-level, level up
Tend to lead with an empathetic opener implying the brand “gets it” whatever “it” happens to be.
Social proof without actual names or identifiers (referencing “top athletes” without individuals)
Transformational claims in headlines “how to x without y” etc.
One-word questions as fragments
Claiming things are extremely custom-made (such as marketing plans built just for you, etc.)
Names
Aria
Ash (place/title)
Calder
Chen
Elara
Eldenmoor
Ethan
Evelyn
Fen (place)
Names ending in -wick (for example: Fenwick)
Holloway
Ironhaven
Liam
Lyra
Maren
Marcus
Mira
Noah
Stormwatch
Voss
Zephyr
Generic constructions like “The Northern Reaches”
Words/Phrases
And this is the thing they don’t tell you (and other clickbait-y styles)
At its core
But here’s…
Carries real weight
[Character] thought about x, and he thought about y, and he thought about z, and…
Colder (particularly when describing people, not temperature)
Crucial/vital
Delve
Fundamental
Genuine
Grounded
It’s important to note...
It’s worth noting that...
Journey (calling everything a journey)
Land (as in “that landed”)
Let that sink in.
Like x was made for this
Name it/naming it
Navigate
Nuanced/nuance
Particular (the particular quality of a person who...)
Plain/plainly
Powerful (everything is powerful)
Quiet (as an adjective or adverb)
Reframe
Resonate
Robust
Seen
Sharp/sharpened
Show up
Sit with
Speaks to
Structural
Tapestry
That isn’t earned / is earned
That matters
That’s not a x—it’s a y.
That’s the part...
The kind of X that...
The specific quality of…
Throat
Underscore
Weight (as in emotional, not physical)
X is doing a lot of work...
You’ve done more in X than most people do in Y


I know the reason you are compiling this list is based on educational purposes, but I can’t help but think it has the potential for misuse. In the current AI witch hunt environment, it would be easy to misuse a list like this by those who would demonize “suspicious” authors without your editorial experience or context.