“We shouldn’t be using ChatGPT for this,” said my colleague, glancing at the draft I’d just sent him.
“I agree. That’s my writing.”
“Oh.” He paused and read a bit. “Well, the em dashes and the structured paragraphs make this seem like AI slop, even if the content is there.”
“Thanks for the feedback,” I said. Then I flung my laptop across the room and leaped to my feet. “Those are my em dashes,” I growled, pounding the table. “And I always write in structured paragraphs. I’m an English major.” [Editor’s note: Em dashes are also house style at Stack.]
OK, no laptops had been thrown and no tables had been pounded, however I used to be a bit affronted. It was the primary and solely time somebody had vocalized the idea that my work was AI-generated, and it made me surprise if anybody (incorrectly) perceives the content material I write as AI slop.
Why did the concept somebody may label my work as AI-generated make me really feel each icky and irritated? Why was I so desperate to deny utilizing a device that lots of of tens of millions of individuals are utilizing? Why did I slip in that defensive “incorrectly” just a few sentences again?
Initially, I wished to put in writing in regards to the supposed telltale indicators of AI-generated textual content and whether or not these indicators really reveal something. In different phrases, I wished to defend my em dashes—and I’ll. However as I assumed extra in regards to the topic, a simple weblog submit turned an ontological exploration into how we understand, perceive, and expertise AI-generated content material. As a result of—as practically anybody can let you know even when they’ll’t fairly clarify why—there’s a major distinction between how we expertise AI artwork and the way we expertise artwork (visible, musical, or literary) created by people. What’s the nature of that distinction, and what does it inform us?
First, a disclaimer. I don’t contemplate the articles and different content material I write as a part of my job to be artwork, essentially. However that content material is my physique of labor, the product of my effort and expertise. Like each different advertising and marketing author I’ve ever labored with, I take delight in my writing and its attribution. The road between work product and artwork is just not all the time a shiny one. For the needs of this text, no less than, please forgive some conflation between the 2.
The suggestion that em dashes are an indicator of AI writing doesn’t come out of nowhere. Wikipedia’s intensive field guide to patterns related to AI-generated content material covers fashion (overuse of em dashes, part headings zhuzhed up by emoji), language and grammar (overdependence on the rule of three, weasel phrases), and broader content material points. These points, much less simply named however simply as apparent whenever you’re a Wikipedia editor studying a ton of AI slop, embrace superficial evaluation, overly promotional language, and an undue emphasis on a topic’s symbolic significance or protection within the media.
The sphere information leads with a vital disclaimer: “Not all textual content that includes these indicators is AI-generated, as the massive language fashions that energy AI chatbots are skilled on human writing, together with the writing of Wikipedia editors.” Herein lies the irony: lots of the tells related to AI writing stem from the skilled and educational writing on which these LLMs have been skilled. We taught them to make use of em dashes; to make use of a indifferent, impartial tone; to make use of didactic little disclaimers like, “It’s vital to know.” AI is sort of a school scholar who’s picked up the vernacular of educational writing, however whose output, upon nearer inspection, reveals how little they really perceive.
Most writers I do know could be offended to have their work recognized as AI-generated. Why? Partially, it’s easy: We don’t need our work miscategorized as AI writing as a result of AI writing sucks. However there’s extra to it than that.
Fairly other than the linguistic, stylistic, and substantive issues on full show in any chunk of AI-generated textual content, AI writing feels hole. “Simply product, no wrestle” was my colleague Ryan’s succinct take. It’s akin to how data-driven selections made by LLMs could also be logically sound and affordable, however they are not innovative. AI output is fully data-driven: statistical representations of a mannequin’s coaching information. Therefore the notion of LLMs as stochastic parrots: able to mimicking human speech with out having the faintest notion of its that means.
Stochastic parrot vibes are undoubtedly a part of why we really feel dissatisfied, unsettled, and even disgusted or betrayed after we understand we’re taking a look at AI-generated content material somewhat than one thing created by a human. Buddy request from an AI character? Yuck. Tilly Norwood on TV? Laborious go.
A much–memed comparability between AI artwork and the fae of folks knowledge is revealing: “Recognizing an AI picture is mainly the identical guidelines as recognizing the fae in previous tales. Depend the fingers, depend the knuckles, depend the enamel.” Each AI and the artful fae might be uncovered by telltale deviations from the human physique; each are out to steal one thing important from us. The analogy makes plain our instinctive disquiet with AI output, together with our deeper mistrust of the expertise itself: “Be very, very cautious,” goes one model, “as a result of [AI] is stealing individuals’s faces and voices.”
For me personally, AI content material is straight away offputting in a approach that’s arduous to explain however not possible to disclaim. The clean symmetrical faces of AI-generated models, untroubled by quirks of genetics or persona, inhabit an uncanny valley whose emotional and contextual vacancy is repulsive. There’s no there there. You may say AI content material lacks a sure je ne sais quoi, during which the quoi is the ineffable human issue.
My colleague Phoebe is Gen Z, which feels doubtlessly related to me, an elder millennial, as a result of analysis has proven that individuals from completely different generations have a tendency to have interaction with AI in typically wildly other ways. I requested her how she feels when she identifies a chunk of content material as AI-generated.
“My major interplay with AI slop has (clearly) been on social media,” she mentioned, “and it’s sadly reached the purpose the place I see [AI content] and simply skip. With an increasing number of AI movies hitting my feed, my pores and skin begins to crawl as soon as I understand the photographs I’m seeing of a cute canine or bunnies on trampolines that in the first place gave me heat fuzzies are literally generated by a machine. The soul of it simply leaves for me, and it feels unnatural.”
Phoebe additionally studies, “One of many memes on TikTok now’s for individuals to submit ‘Is that this AI, I can’t inform’ underneath movies which might be clearly not AI as a joke.” It’s a stage of irony and abstraction that may not have resonated simply a few years in the past.
I requested Ryan the identical query: How do you’re feeling whenever you acknowledge one thing you’re watching/studying/listening to as AI-generated?
“It relies upon,” he mentioned. “If it’s one thing that’s ephemera across the factor I’m taking a look at, like a weblog header or background in a sport, I’m mildly disapproving, however I get it. If it’s the factor I’m taking a look at itself, then I really feel betrayed and a bit disgusted. I wish to experience another person’s brainwaves after I’m studying/viewing artwork/watching films.”
“Numerous artwork is simply product anyway,” Ryan acknowledged (pondering, I assume, of Tron: Ares), “so it’s soulless crap made by individuals, however AI outputs are by definition soulless. There’s no authorial intent, simply stats.” He additionally referenced a question that’s grow to be a well-known chorus in conversations about AI content material: “Why ought to I trouble to learn one thing you didn’t trouble to put in writing?”
As quickly as we had AI textual content turbines, we had AI textual content detectors. These instruments (powered, naturally, by AI) promise to find out how a lot of a given textual content is AI-generated. Roughly one minute later, we had AI humanizers like UnAIMyText, which promise to make your AI-generated textual content sound like one thing written by an precise respiration particular person.
As you’d count on, many customers of AI detectors are lecturers making an attempt to find out whether or not their college students really did the homework. And lots of customers of AI humanizers are college students making an attempt to get an AI-generated paper previous those self same lecturers.
However many individuals, not least the scholars themselves, see a basic contradiction at play right here. Whilst faculty insurance policies talk to college students that AI writing instruments are to be prevented or, failing that, used surreptitiously, children and younger adults soak up the message that they should use AI to be aggressive in a frightening job market. They’re conscious of the ubiquity of AI instruments and the paucity of entry-level roles; from their perspective, in the event that they don’t work with AI, they’ll lose their future job to somebody who does.
Instruments constructed to detect or disguise AI reveal our societally blended emotions in regards to the expertise itself. On one hand, AI instruments promise that with no funding, no expertise, and solely a bit time, you can also create textual content, pictures, or immersive movies to go your class, promote your product, or impress your boss. In fact individuals are going to make use of them. Then again, detectors and humanizers underscore our persistent discomfort with the entire idea: We wouldn’t be constructing instruments to disclose or conceal AI-generated content material if we noticed it as an uncomplicated constructive.
All that mentioned, we buried the lede a bit right here. The very fact is that AI detectors don’t work. Numerous instruments have decided that Jane Austen’s Pleasure and Prejudice (1813) was AI-generated; ditto the USA Structure (1787).
And detectors aren’t simply ineffective; they’ll do incalculable skilled, educational, and reputational hurt to individuals accused of utilizing them. Throughout quite a few research, AI has demonstrated bias towards non-native English speakers, Black students, and neurodiverse people, amongst different populations.
Universities together with MIT and the University of San Diego have mentioned plainly that “AI detectors don’t work” and “AI detectors are problematic and never really useful as a sole indicator of educational misconduct.” A information for instructors at Northern Illinois College referred to as AI detectors an ethical minefield as a result of accusations primarily based on false positives can wreck college students’ educational careers, whereas an article in Inside Larger Ed explores why many professors are “apprehensive that new instruments for detecting AI-generated plagiarism might do extra hurt than good” — primarily as a result of analysis has proven that AI detection instruments are “neither correct nor dependable.”
For all these causes, groups that work on Stack Overflow’s PubPlat don’t use AI detectors.
Let’s return to the query of how individuals expertise AI-generated content material, whether or not they learn it, watch it, or hear it.
Briefly? We don’t prefer it. Research revealed in Scientific Reviews discovered that “individuals devalue artwork labeled as AI-made throughout quite a lot of dimensions, even once they report it’s indistinguishable from human-made artwork, and even once they imagine it was produced collaboratively with a human.”
On the same notice, a study in Computer systems and Human Habits discovered that “people understand the identical paintings as much less artistic and awe-inspiring when it’s labeled as AI-made (vs. human made).”
Curiously, members in that research overwhelmingly most well-liked artwork they thought was made by people even when the content material was really AI-generated, mentioned Guanzhong Du, one of many coauthors. “Irrespective of which one is definitely made by the human artist, individuals desire the paintings that’s labelled as human,” Du mentioned. “They suppose it’s extra artistic—and once they take heed to music or take a look at work made by human artists, they suppose they’re extra awe-inspiring.”
“It’s pictures,” reads the highest remark in a Reddit discussion of the Computer systems and Human Habits research. “It’s not artwork.” (Since I’m, in any case, an English main, this jogs my memory of Capote’s famous dig at Kerouac: “That’s not writing. That’s typing.”)
Once we take a look at AI output, we all know that nobody dreamed and labored to create this work with a purpose to share it with us. And that, evidently, issues.
An particularly insightful tackle the unsatisfying nature of AI artwork comes from cartoonist Matthew Inman, creator of The Oatmeal. He writes, “Once I noticed the unique Jurassic Park in theaters, I used to be blown away by the dinosaurs. They had been CG, however I didn’t take a look at the Brachiasaurus and suppose, ‘Gross, it’s only a bunch of monitoring dots hooked up to a growth crane.’”
As an alternative, he remembers pondering, “How did somephysique [he emphasizes “body”] make this?” The Brachiasaurus was “an expression of human beings making human selections. It was the product of self-discipline, expertise, and creativeness.”
“Seeing AI artwork,” he writes, “I don’t really feel that approach in any respect.”
Recall the Scientific Reviews study we talked about above: Individuals “devalue” AI-made artwork “even once they imagine it was produced collaboratively with a human.” We understand a yawning hole between people producing visible results with CGI software program and people prompting Sora to generate video clips.
DC Comics president Jim Lee announced this 12 months that the corporate would “not help AI-generated storytelling or paintings”: “not now, not ever.” “Individuals have an instinctive response to what feels genuine,” he advised the viewers at New York Comedian Con. “We recoil from what feels faux. That’s why human creativity issues. AI doesn’t dream. It doesn’t really feel. It doesn’t make artwork. It aggregates it.”
The Computer systems and Human Habits study means that our instinctive response to AI content material goes past easy distaste. Artwork made by AI poses what the research’s authors name “an ontological menace” to the assumption that creativity is a uniquely human high quality. Possibly that’s why many creators reply scathingly to the encroachment of AI on their work: We see AI artwork as infringing on uniquely human territory.
OK, OK, you could be pondering. AI doesn’t create artwork. However can’t I exploit it in my advertising and marketing campaigns?
Solely if you wish to alienate your prospects. It seems that individuals don’t like AI in advertising and marketing and promoting content material any greater than they prefer it elsewhere. A current report on AI-generated advertising and marketing content material discovered that “client enthusiasm for AI-generated creator work has dropped from 60% in 2023 to 26% in 2025, as feeds overflow with what viewers deride as ‘AI slop’—uninspired, repetitive, and unlabeled content material.”
It doesn’t assist that so many AI ads are laughably janky and/or deeply bizarre. A potential buyer who reacts like Phoebe when she acknowledges AI content material (“My pores and skin begins to crawl… The soul of it simply leaves, and it feels unnatural”) is not going to click on by.
Promoting may not be artwork, however AI has but to generate any advert content material that sticks with individuals, that makes a cultural dent, that modifications the way in which we take into consideration and devour a product. There’s no 1984, no Just Do It, no Got Milk?, and no Buying the world a Coke. AI is not any Don Draper at Esalen—although his junior copywriters tasked with writing ten taglines earlier than lunch would undoubtedly have appreciated it.
Maybe there’s an upside to the ubiquity of AI artwork: It’d educate us to acknowledge the worth of human artists. The Scientific Reviews paper discovered that “evaluating pictures labeled as human-made to pictures labeled as AI-made will increase perceptions of human creativity, an impact that may be leveraged to extend the worth of human effort.”
Human creativity and human effort: We’re gonna miss it when it’s gone. Whether or not we’ll miss it sufficient to restrict AI’s skill to cannibalize and mixture the work of human writers, filmmakers, visible artists, musicians, and different creators isn’t clear. Neither is it clear that, in stark financial phrases, the worth of human artists can be sufficient to offset the lure of apparently low cost, just about instantaneous content material to feed the machine.
