As someone born and raised in the US, prior to AI or even widespread personal computers, I also write in a similar style. I'm bracing myself for the day someone accuses me of using AI. I will say, I have a tendency to miss typos, so maybe that's been my saving grace so far.
This article was really eye opening, thank you for sharing.
This is a great piece and an engaging personal reflection on why LLMs produce the output they do. It's something I've been pushing back on for a while now. There's this persistent meme that AI has corrupted the English language with its "tell tale style" but in reality, it has simply mimicked what already existed, as you so clearly explain here. AI cannot have it's own style when its been trained on countless written works. LLMs are, among other things, pattern matching systems, therefore any "style" its developed is a direct result of something it deemed highly recognizable over millions of training instances. Its literally impossible for AI to sound like anything but the data it was trained on, which is human writing. What people are recognizing now, is just LLM use because when everyone uses it, everyone begins to sound a similar way. Most people dont have the interest or frankly, the ability use a wide ranging vocabulary or to write in highly structured ways. But, now, it appears everyone does. In the age of AI/LLMs, this comes down to a couple things: sloppy prompting, poor review of LLM output and the explicit farming out of written word. For those who see themselves are real writers, creating works from scratch and sharing real perspectives and experiences will continue to have value, even if the formal nature of it appears to be AI generated.
I'm pretty sure that an AI can write in any style if you prompt it hard enough. Want to write Shakespearean Omegaverse fics? You can do that at the click of a button, and that should scare everyone. I've heard fiction writers are also getting hit hard by erroneous AI accusations.
I find this fascinating, but I don't agree that the writing you describe (or this essay) "sounds like AI."
In fact, AI writing is full of "burstiness<" with short dramatic (often negative declaratives) followed by cliches of "quiet revolutions" and so on in longer sentences.
"She wasn't an X. She wasn't a Y. She didn't have Z. But she quietly revolutionized how the world's greatest minds think about Q, turning [concrete W] into [tenuously connected abstract value V]."
"But even here, you can’t assume that anyone using the word is being puppeted by A.I. In 2024, the investor Paul Graham made that mistake when he posted online about receiving a cold pitch. He wasn’t opposed at first. “Then,” he wrote on X, “I noticed it used the word ‘delve.’” This was met with an instant backlash. Just like the people who hang their identity on liking the em dash, the “delve” enjoyers were furious. But a lot of them had one thing in common: They were from Nigeria.
In Nigerian English, it’s more ordinary to speak in a heightened register; words like “delve” are not unusual. For some people, this became the generally accepted explanation for why A.I.s say it so much. They’re trained on essentially the entire internet, which means that some regional usages become generalized. Because Nigeria has one of the world’s largest English-speaking populations, some things that look like robot behavior might actually just be another human culture, refracted through the machine."
"If you ask ChatGPT what its own writing style is like, it’ll come up with some false modesty about how its prose is sleek and precise but somehow hollow: too clean, too efficient, too neutral, too perfect, without any of the subtle imperfections that make human writing interesting. In fact, this is not even remotely true. A.I. writing is marked by a whole complex of frankly bizarre rhetorical features that make it immediately distinctive to anyone who has ever encountered it. It’s not smooth or neutral at all — it’s weird."
Thank you for writing this. As someone who studied english literature (and who is not themselves white and from the core) I felt a lot of similar feelings growing up. Despite this being an essay about robotic writing, there is a rich humanity that shines through. :)
As someone with a lot of native English and ESL students I can definitely say by ESL students use LLM much more than the native English. I know because their essays transformed radically and implausibly the moment they encountered LLM.
You reminded me how I learned English initially during two weeks of intense self study in the park, as my fifth language, using two courses, one American and one British, with five hours audio between them, played on a Sony Walkman I earned the money for during the previous summer. Then I acquired the Oxford Advanced Learners Dictionary as well as one more, that was three times larger, and received an American book from a friend. After reading most of those, I conversed with an American missionary teacher who helped to improve the English of teachers and college students in my native town. They all thought that I just came back from America. Same was the impression of the people in the country where I studied English in the park. A year later I talked to the American ambassador in another country where I received my immigration visa and came to America few months after that. I kept reading books even after exposure to the convenience of learning further using television with closed captioning and the internet. The funny thing is that I never mastered the German and French I had in school and high school to the same level, although the people in München assumed I was local, until I opened my mouth. I feel I would be fluent in German if I stayed there for few more weeks though… 🤣
And yes, the Kenyans I played tennis with in the country with the park, all spoke English that was very pleasant to listen to.
The part where you're wrong is: overusing groups of three, "Not [just] X. Y.", or chains of simple sentences - the usual "ChatGPT tells" - don't come from academic papers, encyclopaedias or legal documents. You'd be hard-pressed to find an encyclopaedia written in this style; if you think about it for a minute, I'm sure you'll agree. Instead, they're associated with marketing copy (the likely origin of the style and the only context where it makes sense) and, in recent times, social media, especially LinkedIn. That's where ChatGPT got these "habits" from - writing that tries to be superficially engaging, but quickly becomes annoying when the same 4-5 patterns are used over and over.
"Not A. Not B. C." looks like it comes from a billboard advertising a new model of SUV, not from an academic paper (oh no - I didn't put the "not" part first - a crime against ChatGPT).
I like your writing and understand your frustration with stuff like being called out for using a dash. I just disagree with the premise - I think you're giving ChatGPT's writing ability way too much credit.
"Reinforcement Learning From Human Feedback" - how ChatGPT fine-tuned its style - results in convergence towards mediocrity and endless repetition of rhetorical structures that the average person finds compelling.
Personally, I don't pay mind to these microscopic rules for LLM detection. If my conversation partner does not express themselves concisely and with purpose, they are wasting my time. At that point, it does not matter if they use generative AI or decided to personally go on a tangent. ⌛
Less is more. I vastly prefer sloppy informal remarks over bloated speculative rambles. I prefer an e-mail without Hello or Regards over an e-mail with uncalled-for headers and sentence-long bullet points. I value the meaningful information more than the structure. If a text structure fails to convey the content any better than a plain stream of words would have, then the structure should not have been used. At best, unnecessary formalities and structures are thoughtless habit. All too often, they are deliberately used to disguise the absence of meaningful information. That may be called a deception, a ruse, or slop.
I can tell this article is authentic because every single sentence is making a point, and every single point ties into one coherent narrative, and that narrative perfectly fits into the context. The formatting is minimal: Italics for two asides, and some horizontal rules to break the narrative into beats. Only what makes sense in this context, and it helps the reader immensely. In other words, thank you for writing such a compelling article! I picked up some writing tips from this, too! ❤
Heh. Saw this being shared on the fediverse and… (sigh).
You ever notice that when the colonized get good at something, colonizers just… decide to rug-pull the thing we trained our entire fucking childhood for with naught but a few cutting words? “You write like ChatGPT”— motherfucker I WRITE like I was supposed to be good at it, what's your excuse, Brittneigh from Bumfuck, Connecticut, your lips burned while drinking seltzer water, was it?
I miiiiiight be a little bitter. I spent most of my life being immersed in English — not as intensely as you, I think, but where you had the KCPE, we had standards that were based by the old Cambridge 1119, and I think that drill exists still for younger students in my country in the IGCSE English syllabus (not the English as a Second Language, no sir, those are for parvenus). But, like… god. I spent YEARS getting good at this fucking language and I fucking DESPISE the fact that now, all it takes for someone to invalidate my argument online is “lol yOu suOnD LiKe cHaTGPt huRR” as if that's a gotcha. What do you say about that? What CAN you say?
And the other galling thing was that argument where folks who were colonized “imitated the colonizers” so it's perfectly fine for the colonizers to dress up and appropriate the cultures of the colonized, right? Because having it be drilled into you that this was your one way to success, to having a future, was totally the same as putting on someone's culture as a costume and dressing up as a culturally illiterate, unfunny, offensive stereotype of a colonized person for fucking clout on your stupid-ass social media platforms. Just get dressed as a Nazi, why don't you. That's your culture, isn't it — taking the bones of people — WHO ARE STILL ALIVE! — and act like they're dead and gone forever so that you can use the power of their departed spirits — WHO ARE STILL HERE! — for your meaningless games of clout-chasing.
God, I'm going to be cranky this morning. Not because of you, though. Great essay.
Really interesting essay, thanks. My only (small!) comment or criticism — doesn’t every English speaker complete “The cat sat on the…” with “mat”? 😅
(I guess that’s wrong! It must be culturally-specific… A sentence which appears in specific books for teaching children to read, and then gains extra currency by being quoted in movies and TV.)
This was so insightful, thank you so much. I will say, you can still tell a very well educated human wrote this because there is soul in the writing. While you and gpt follow many of the same writing conventions, a lot of the time generative AI just rambles without a point, saying the same thing for 3 sentences. Can't say the same about you, and at least we have that going for us!
As someone born and raised in the US, prior to AI or even widespread personal computers, I also write in a similar style. I'm bracing myself for the day someone accuses me of using AI. I will say, I have a tendency to miss typos, so maybe that's been my saving grace so far.
This article was really eye opening, thank you for sharing.
This is a great piece and an engaging personal reflection on why LLMs produce the output they do. It's something I've been pushing back on for a while now. There's this persistent meme that AI has corrupted the English language with its "tell tale style" but in reality, it has simply mimicked what already existed, as you so clearly explain here. AI cannot have it's own style when its been trained on countless written works. LLMs are, among other things, pattern matching systems, therefore any "style" its developed is a direct result of something it deemed highly recognizable over millions of training instances. Its literally impossible for AI to sound like anything but the data it was trained on, which is human writing. What people are recognizing now, is just LLM use because when everyone uses it, everyone begins to sound a similar way. Most people dont have the interest or frankly, the ability use a wide ranging vocabulary or to write in highly structured ways. But, now, it appears everyone does. In the age of AI/LLMs, this comes down to a couple things: sloppy prompting, poor review of LLM output and the explicit farming out of written word. For those who see themselves are real writers, creating works from scratch and sharing real perspectives and experiences will continue to have value, even if the formal nature of it appears to be AI generated.
I'm pretty sure that an AI can write in any style if you prompt it hard enough. Want to write Shakespearean Omegaverse fics? You can do that at the click of a button, and that should scare everyone. I've heard fiction writers are also getting hit hard by erroneous AI accusations.
I find this fascinating, but I don't agree that the writing you describe (or this essay) "sounds like AI."
In fact, AI writing is full of "burstiness<" with short dramatic (often negative declaratives) followed by cliches of "quiet revolutions" and so on in longer sentences.
"She wasn't an X. She wasn't a Y. She didn't have Z. But she quietly revolutionized how the world's greatest minds think about Q, turning [concrete W] into [tenuously connected abstract value V]."
The best description of AI-cliche writing is still from Sam Kriss in the NYT Magazine:
https://www.nytimes.com/2025/12/03/magazine/chatbot-writing-style.html?unlocked_article_code=1.jVA.866A.u82hipd2ppKv&smid=url-share
But here Kriss is on the same wavelength as you:
"But even here, you can’t assume that anyone using the word is being puppeted by A.I. In 2024, the investor Paul Graham made that mistake when he posted online about receiving a cold pitch. He wasn’t opposed at first. “Then,” he wrote on X, “I noticed it used the word ‘delve.’” This was met with an instant backlash. Just like the people who hang their identity on liking the em dash, the “delve” enjoyers were furious. But a lot of them had one thing in common: They were from Nigeria.
In Nigerian English, it’s more ordinary to speak in a heightened register; words like “delve” are not unusual. For some people, this became the generally accepted explanation for why A.I.s say it so much. They’re trained on essentially the entire internet, which means that some regional usages become generalized. Because Nigeria has one of the world’s largest English-speaking populations, some things that look like robot behavior might actually just be another human culture, refracted through the machine."
From the above:
"If you ask ChatGPT what its own writing style is like, it’ll come up with some false modesty about how its prose is sleek and precise but somehow hollow: too clean, too efficient, too neutral, too perfect, without any of the subtle imperfections that make human writing interesting. In fact, this is not even remotely true. A.I. writing is marked by a whole complex of frankly bizarre rhetorical features that make it immediately distinctive to anyone who has ever encountered it. It’s not smooth or neutral at all — it’s weird."
Thank you for writing this. As someone who studied english literature (and who is not themselves white and from the core) I felt a lot of similar feelings growing up. Despite this being an essay about robotic writing, there is a rich humanity that shines through. :)
I get this feedback so much on Twitter on my articles! Totally get it.
As someone with a lot of native English and ESL students I can definitely say by ESL students use LLM much more than the native English. I know because their essays transformed radically and implausibly the moment they encountered LLM.
You reminded me how I learned English initially during two weeks of intense self study in the park, as my fifth language, using two courses, one American and one British, with five hours audio between them, played on a Sony Walkman I earned the money for during the previous summer. Then I acquired the Oxford Advanced Learners Dictionary as well as one more, that was three times larger, and received an American book from a friend. After reading most of those, I conversed with an American missionary teacher who helped to improve the English of teachers and college students in my native town. They all thought that I just came back from America. Same was the impression of the people in the country where I studied English in the park. A year later I talked to the American ambassador in another country where I received my immigration visa and came to America few months after that. I kept reading books even after exposure to the convenience of learning further using television with closed captioning and the internet. The funny thing is that I never mastered the German and French I had in school and high school to the same level, although the people in München assumed I was local, until I opened my mouth. I feel I would be fluent in German if I stayed there for few more weeks though… 🤣
And yes, the Kenyans I played tennis with in the country with the park, all spoke English that was very pleasant to listen to.
Keep writing, it is beautiful!
The part where you're wrong is: overusing groups of three, "Not [just] X. Y.", or chains of simple sentences - the usual "ChatGPT tells" - don't come from academic papers, encyclopaedias or legal documents. You'd be hard-pressed to find an encyclopaedia written in this style; if you think about it for a minute, I'm sure you'll agree. Instead, they're associated with marketing copy (the likely origin of the style and the only context where it makes sense) and, in recent times, social media, especially LinkedIn. That's where ChatGPT got these "habits" from - writing that tries to be superficially engaging, but quickly becomes annoying when the same 4-5 patterns are used over and over.
"Not A. Not B. C." looks like it comes from a billboard advertising a new model of SUV, not from an academic paper (oh no - I didn't put the "not" part first - a crime against ChatGPT).
I like your writing and understand your frustration with stuff like being called out for using a dash. I just disagree with the premise - I think you're giving ChatGPT's writing ability way too much credit.
"Reinforcement Learning From Human Feedback" - how ChatGPT fine-tuned its style - results in convergence towards mediocrity and endless repetition of rhetorical structures that the average person finds compelling.
Personally, I don't pay mind to these microscopic rules for LLM detection. If my conversation partner does not express themselves concisely and with purpose, they are wasting my time. At that point, it does not matter if they use generative AI or decided to personally go on a tangent. ⌛
Less is more. I vastly prefer sloppy informal remarks over bloated speculative rambles. I prefer an e-mail without Hello or Regards over an e-mail with uncalled-for headers and sentence-long bullet points. I value the meaningful information more than the structure. If a text structure fails to convey the content any better than a plain stream of words would have, then the structure should not have been used. At best, unnecessary formalities and structures are thoughtless habit. All too often, they are deliberately used to disguise the absence of meaningful information. That may be called a deception, a ruse, or slop.
I can tell this article is authentic because every single sentence is making a point, and every single point ties into one coherent narrative, and that narrative perfectly fits into the context. The formatting is minimal: Italics for two asides, and some horizontal rules to break the narrative into beats. Only what makes sense in this context, and it helps the reader immensely. In other words, thank you for writing such a compelling article! I picked up some writing tips from this, too! ❤
Sorry, but I ain't reading all this AI generated slop, but thank you for making a clown of yourself. 🤡
Heh. Saw this being shared on the fediverse and… (sigh).
You ever notice that when the colonized get good at something, colonizers just… decide to rug-pull the thing we trained our entire fucking childhood for with naught but a few cutting words? “You write like ChatGPT”— motherfucker I WRITE like I was supposed to be good at it, what's your excuse, Brittneigh from Bumfuck, Connecticut, your lips burned while drinking seltzer water, was it?
I miiiiiight be a little bitter. I spent most of my life being immersed in English — not as intensely as you, I think, but where you had the KCPE, we had standards that were based by the old Cambridge 1119, and I think that drill exists still for younger students in my country in the IGCSE English syllabus (not the English as a Second Language, no sir, those are for parvenus). But, like… god. I spent YEARS getting good at this fucking language and I fucking DESPISE the fact that now, all it takes for someone to invalidate my argument online is “lol yOu suOnD LiKe cHaTGPt huRR” as if that's a gotcha. What do you say about that? What CAN you say?
And the other galling thing was that argument where folks who were colonized “imitated the colonizers” so it's perfectly fine for the colonizers to dress up and appropriate the cultures of the colonized, right? Because having it be drilled into you that this was your one way to success, to having a future, was totally the same as putting on someone's culture as a costume and dressing up as a culturally illiterate, unfunny, offensive stereotype of a colonized person for fucking clout on your stupid-ass social media platforms. Just get dressed as a Nazi, why don't you. That's your culture, isn't it — taking the bones of people — WHO ARE STILL ALIVE! — and act like they're dead and gone forever so that you can use the power of their departed spirits — WHO ARE STILL HERE! — for your meaningless games of clout-chasing.
God, I'm going to be cranky this morning. Not because of you, though. Great essay.
Thank you.
Brilliant!
Really interesting essay, thanks. My only (small!) comment or criticism — doesn’t every English speaker complete “The cat sat on the…” with “mat”? 😅
(I guess that’s wrong! It must be culturally-specific… A sentence which appears in specific books for teaching children to read, and then gains extra currency by being quoted in movies and TV.)
This was so insightful, thank you so much. I will say, you can still tell a very well educated human wrote this because there is soul in the writing. While you and gpt follow many of the same writing conventions, a lot of the time generative AI just rambles without a point, saying the same thing for 3 sentences. Can't say the same about you, and at least we have that going for us!