Artificial Intelligence and language have never been this close before.
And that’s sparking a massive debate around the world, especially among writers, readers, and researchers trying to figure out just how far machine influence reaches into the way we write, read, and communicate every day.
Let me throw a quick challenge at you before we dive in.
Three paragraphs, pulled from three different hotel reviews. Can you tell which one, if any, was written by an artificial intelligence?
The hotel is in a great location for everything. Lots of places to eat and drink. The hotel itself is always buzzing. The tavern on the ground floor is a must-visit. Food, service, prices, and atmosphere were all great.
A good hotel, although the room had the proportions of a well-decorated elevator. Slept well, the shower was excellent, the staff friendly. Breakfast was busy but competent. Would return, though probably not with a very large suitcase.
Excellent base for a trip to London. The room was quiet, the bed comfortable, and everything worked exactly as it should. The staff were helpful without being intrusive. A smooth and hassle-free stay from start to finish.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
How do you think you did on that test?
According to Claire Hardaker, a forensic linguistics professor at Lancaster University, most people get this kind of judgment right only about 60% of the time, which is basically the same hit rate as flipping a coin. Her online test, called Bot or Not, asks users to spot the fake texts in a series of fifteen reviews, and that middling result tends to surprise anyone who swears up and down they can spot AI writing from a mile away.
It was exactly this kind of doubt that blew up on social media when a prize-winning short story by writer Jamir Nazir came under scrutiny. A lot of people rushed to condemn it, swearing they had recognized the hand of the machine in his text. If you know, you know, one user commented with all the confidence in the world. The thing is, Nazir later publicly stated that he did not use any artificial intelligence to create the work. And as for the review test up top: only the first one was authentic. Did you get it right? 😉
The Classic Signs of AI Writing: Myth or Reality?
For a long time, people circulated the idea that texts generated by artificial intelligence had very specific, easy-to-spot markers. According to Hardaker, people tend to rely on a handful of quick, simplistic rules to sniff out AI writing, like the presence of clichés and the frequent use of the em dash. The famous rule of three, where words or phrases are organized into a satisfying trio, is also considered a dead giveaway. People have learned very simplistic patterns and now they go around applying them like crazy everywhere, the researcher says.
The problem is that these supposed telltale signs also show up — a lot — in human writing. After all, large language models were trained on texts written by real, flesh-and-blood people. You could go back to Charles Dickens and accuse him of using AI, because he also loved his em dashes, Hardaker jokes. And public speakers have known the rule of three since Julius Caesar dropped that classic Veni, vidi, vici. In other words, a careful writer with solid experience who revises their own work thoroughly can sound just as polished as a well-tuned model.
Precisely because certainty is so hard to come by, suspicion has become the default setting. In the literary world, accusations of AI use now follow writers around with varying levels of justification. A debut horror novel called Shy Girl was pulled from circulation by publisher Hachette after rumors about AI use spread online — something the author denies to this day. Meanwhile, The Future of Truth by Steven Rosenbaum, a serious study on how AI reshapes reality, turned out to contain several fabricated, or hallucinated, quotes that the author himself acknowledged in a public apology.
Media organizations, including major newspapers, are receiving more and more complaints about texts supposedly generated by AI. These involve gut feelings about certain expressions, but also comments about typos and grammatical errors. In one curious case, the word then appeared duplicated by mistake in a sentence, and a reader wrote in saying they could not imagine a human editor letting something like that slip through — demonstrating a touching faith in the infallibility of copy editors. The truth is that all of this creates a linguistic hall of mirrors: AI trains on human writing, and humans are stylistically influenced by AI. Without a confession from the author, it is nearly impossible to make a definitive call on the origin of any single text, and that uncertainty is a recipe for paranoia. 🧠
Automated Detectors: Useful Tool or Trap?
With the rapid growth of language model use in professional and educational settings, it did not take long for tools to pop up promising to automatically detect whether a text was generated by artificial intelligence. The pitch was tempting: just paste the text in, hit a button, and the machine would tell you if another system had written it. But that scenario turned out to be far more complicated in practice than it sounded in theory.
Hardaker, who has served as an expert witness in court cases, is extremely skeptical about the effectiveness of these tools. She explains that many of us naturally write in a way that would be flagged as AI-like, and cites neurodivergent individuals as an example — people whose style can be unfairly singled out. On top of that, it is perfectly possible to tweak AI output to make it sound more human. When you throw that kind of content into a detector, the results come out completely haywire, the professor summarizes.
One detector that has gained popularity recently, Pangram, boasts false positive rates of about 1 in 10,000 and has shown, in independent testing, to be quite effective at identifying AI writing — even when the text has been run through a humanizer app to disguise its origin. Still, doubts persist. It was possible to fool the tool on the very first try by adopting a pompous, grandiloquent tone that could easily be typical of AI but could also be the work of someone with a naturally over-the-top style, or of a writer steeped in the content that feeds ChatGPT, Claude, and Gemini. And increasingly, that writer immersed in the AI universe is all of us.
Another point worth paying attention to is that these detectors were trained on language patterns that the language models themselves are constantly updating and refining. It is a race between the detector and the generator, and the generator almost always has the edge because it evolves faster. In contexts where an accusation of AI use carries serious consequences — like academic papers or hiring processes — that margin of error is simply unacceptable and can cause real harm to innocent people.
AI Is Changing the Way Humans Write
Perhaps the most fascinating point in this whole debate — and also the most unsettling — is the possibility that artificial intelligence is already shaping human writing even when the writer has not used any AI tool at all. We have known for some time that language models produce text that is slightly different from human writing, on average. Often this only becomes clear when you analyze large volumes of material.
A sharp-eyed researcher linked the sudden popularity of the word delve to language models back in 2024, after combing through a database of scientific articles. Other AI favorites include words like showcase, boast, underscore, and intricate. Interestingly, some researchers believe this phenomenon does not come from the models themselves but from the human workers tasked with evaluating them in a process called reinforcement learning from human feedback. For underpaid, stressed, and time-pressured workers, certain words end up being treated as synonyms for quality, and the model learns to use them more often.
There are other patterns you can spot too. Language models love nouns but use fewer pronouns than humans do — maybe because they do not talk about themselves or other people as much as we do, social creatures that we are. Different models have their own quirks, and you could even call them dialects: Gemini loves to say here is a summary, while Deepseek tends to respond with an enthusiastic Certainly! When asked to edit formal English from other parts of the world, these systems tend to flatten everything toward an Anglo-American standard, in a process researchers have dubbed cultural erasure.
What Writers Think About All of This
Writers who use AI assistants regularly — even if only for editing suggestions — start to internalize certain structural and vocabulary patterns that these tools favor. Over time, those patterns begin showing up spontaneously in these people’s writing, even without any technological assistance. One study analyzed thousands of spontaneous conversations and found that words like delve and boast spiked after the launch of ChatGPT, proving that AI influence has already leaked into the real world.
Novelist Gary Shteyngart, who teaches creative writing at Columbia University, noticed an intense reaction among his students at the idea of AI-made literature. When a graduate student said he was going to write part of an assignment with AI as an experiment, the others were so furious they sent letters to the professor complaining. There is a kind of implicit pact between writer and reader, where you know the work you are receiving was generated by a human being, he explains. Reading literary fiction is an incredible fusion of minds with another person, a dive into someone else’s consciousness. With AI, you enter only the simulacrum of that consciousness.
Peter Stockwell, a professor of literary linguistics at the University of Nottingham, believes AI handles the basics just fine but cannot reach the heights. If you want something very familiar, very average, and totally functional, it is surprisingly good at that, he says. He describes language as layers, with words at the base and narrative structure at the top. AI is great at the lower levels, but the higher you go, the worse it gets. The arc of a story is particularly hard for the machine to nail convincingly, because it lacks the body and the social world that give rise to typically human motivations.
The Threat of Linguistic Leveling
There is also a collective dimension to this process worth considering. When millions of people use the same language models to draft emails, posts, articles, and reports, there is a natural tendency toward homogenization of writing style. This leveling effect is a legitimate concern, though it is not exactly new. People have long worried about the standardizing effects of American movies and television on accents and vocabulary. The important detail is that this kind of influence usually generates a counter-reaction, and there is no reason to think this time will be any different.
Novelist Jennifer Egan takes originality so seriously that she has completely isolated herself from the technology. I feel a danger of contagion, to use a loaded metaphor, she says. I know they have stolen some of my work, and there is nothing I can do about it, but I am not going to give them even one more word voluntarily. She even admits to a recent bout of paranoia: she has started interrogating every em dash and every trio of words she writes, precisely because the goal is to not write something that anyone else could have written.
Not everyone is that extreme, though. Jeannette Winterson, who has written extensively about AI and art, argues that every writer should make their own choice. Humans are tool-using animals, and that has been our success story. Right now, all AI, including generative AI, is a tool. Would I work with a language model? Sure! Why not?, she says. Still, she makes an important caveat: the machine’s linguistic competence does not mean it can match or surpass human expression, because machines do not share our reality, nor do they even have a limbic system that allows us to truly feel.
At the end of the day, maybe it is precisely our capacity for innovation — born from the body, from emotions, and from social life — that continues to set genuinely human writing apart from what machines produce. This is a debate that is just getting started, but one that more and more researchers, writers, and language experts consider urgent and necessary to follow closely. 🧠
