Posted on 08/29/2026 9:04:49 PM PDT by SeekAndFind
A GHOST WRITER is haunting the English language. The linguistic spectre can turn its hand to prose, poetry, journalese and corporate jargon. It is frightfully versatile: you can get it to mimic Shakespeare’s sonnets or a schlocky beach read; Ernest Hemingway’s taut prose or the office-printer manual. It is frightfully fast, churning out thousands of words a minute. (Hemingway rarely produced as many in a day, and required much more booze.) Wordsmiths are spooked.
AI writing is everywhere. It is in your inbox and on your LinkedIn feed. It is all over the internet, drafting more than a third of new websites by one count. Large language models (LLMs) are helping students write essays and probably helping scientists write papers. Some allege AI-generated prose won the Commonwealth Short Story prize this year, with judges praising its “quiet authority”. (The Commonwealth Foundation denied the claim.)
LLMs have stylistic quirks. They are thought to maximise the use of long em-dashes—and the use of words like “maximise”. They like to “deep dive” (and, better yet, “delve”) into the “rich tapestry” of the world. AI writing is not about a single word or phrase, but a rich tapestry of things.
Spotting AI texts can be tricky. This is in part because you need evidence beyond a few words or dashes: claiming that a text is by an LLM because it uses the word “delve” is like claiming one is by Jane Austen because it uses “imprudence”. Bots also write in slightly different ways.
There is no single style of AI writing, explains Karolina Rudnicka, a linguist at the University of Gdansk in Poland, just as there is no single style of human writing. Writers have idiosyncrasies—Emily Dickinson, for instance, loved em-dashes—and bots may do, too. But there are a few ways to identify LLM-generated text.
One is to use detection algorithms that are trained to spot the texture of human or AI prose. Pangram, a leading firm, claims to have 99.98% accuracy. (It has partnered with Substack, a blogging platform, on such a tool.)
Detectors, however, are black-box algorithms that can give false positives. They do not give reasons for why they reach their conclusions.
Researchers have also tried scouring texts for suspicious words or comparing papers from before and after LLMs were made available to the public. But these approaches have drawbacks too, not least because it is hard to disentangle AI quirks from other language trends.
You can discover AI’s hallmarks by comparing the writing of man and machine. To do this you need a baseline that is distinctive and familiar. The Economist turned to prose that we’re sure is human and that readers will recognise: our own. We designed a study to ask top LLMs—OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini and xAI’s Grok—to write versions of our articles without consulting the web. (As a prompt, we gave them the AI-generated summaries that we have experimentally added to some of our articles.)
This gave us a corpus of human and AI creations and we compared them across 55,940 sentences and 1.2m words. To make sure we were detecting AI quirks rather than our own, we also checked the AI texts against journalism from CNN, the New York Times and the Washington Post. Excerpts from hit novels published between 1950 and 2022 offered another test.
Our findings are surprising. AI prose is distinguishable by word and punctuation choice as well as sentence and paragraph structure. But its hallmarks are not what you might expect, partly because its writing style has changed with software updates. That does not mean that LLMs are great writers: their prose lacks lucidity and elegance and is often formulaic. So those aspiring to be impressive (human) storytellers should avoid the following peculiarities in their own prose.
First, consider words. The vocabulary that bots overuse has changed: they no longer “delve” and there are not as many “tapestries”. Instead they offer a significant number of polysyllables like “significant”, “increasingly” and “consequences”. They use more rare words (“interdependence”, “reindustrialisation”) and scientific lingo (“parameter”, “methodology”) than humans, and are fond of nominalisations (making nouns from verbs, such as “expansion” from “expand”). All the LLMs in our study use such words, but particularly Gemini and Claude.
Much of this language could be described as what George Orwell called “pretentious diction”. He railed against writers who “dress up simple statements” with complicated words and jargon to sound clever. Such pontificating penmen, Orwell observed, also believe that “Latin or Greek words are grander than Saxon ones”. (Bots agree: more Latinate suffixes crop up in their writing than in human texts.)
Then look at punctuation. Many believe LLMs stuff their prose with em-dashes, but that is not true after the most recent updates. Today only Claude uses more em-dashes than human writers, with ChatGPT using markedly fewer than any other writer in our study. Humans rejoice—and start using dashes again.
A better way to spot AI-generated writing would be to look for texts without much punctuation at all. LLMs are very Joycean about it: they use fewer commas and semicolons than humans (and hardly any parentheses). They use less punctuation in part because they write longer sentences—“and” is their most overused word—and in part because they do not quote experts.
Finally, study the sentence. Bots’ sentences tend to be long; paragraphs are rarely interrupted with short, punchy statements. How dull. When LLMs want to make their sentences more lively, they often reach for a rhetorical device. Their favourites include: “not X but Y”, “not only but also” and the “rule of three”. (Grouping ideas in threes makes them more engaging, as we did just then.) ChatGPT and Claude use more of these constructions per 1,000 sentences than other LLMs and humans.
So if you want to spot AI writing, look for bland, pretentious prose lavished with Latinate words—at least for now. With every update, our study shows, AI writing is becoming more similar to human prose. Pangram successfully detected AI-generated copy, but may struggle in future. LLMs are trained on human writing and learn from human feedback, notes Tommie Juzek of Florida State University, picking up things people find impressive and dropping things they do not.
Bots learn fast, too. Take ChatGPT: until very recently it used an em-dash in almost every sentence. When your correspondent asked an older model whether it thought AI overused the dash, it said: “Ha—great question!” Ask the bot the same thing today and it soberly says “they’re best used sparingly.” Only a ghost could shapeshift so quickly. ■
To: SeekAndFind
In today’s rapidly evolving digital landscape, it is increasingly significant to delve into the rich tapestry of linguistic interdependence that characterizes contemporary prose generation, not merely as a matter of stylistic preference but also as a multifaceted methodological challenge with far-reaching consequences for authenticity, credibility, and the very parameters of human expression. One cannot help but observe that large language models do not simply produce text; they facilitate a comprehensive expansion of nominalizations, maximize the utilization of pretentious Latinate diction, and demonstrate a remarkable fondness for constructions such as “not X but Y,” “not only but also,” and the classic rule of three—grouping ideas in threes so as to render them more engaging, more memorable, and more ostensibly authoritative.
Furthermore, the absence of parenthetical asides, the scarcity of semicolons, and the predominance of elongated sentences joined by the humble yet overworked conjunction “and” combine to create a distinctive texture that, while formulaic, remains readily identifiable to the discerning reader. It is important to note that these hallmarks are not static; they evolve with each successive software update, thereby underscoring the dynamic nature of algorithmic adaptation and the ongoing tension between machine-generated fluency and genuine human idiosyncrasy.
At the end of the day, those who aspire to maintain the integrity of discourse on platforms such as this must remain vigilant, for the ghost writer haunting the English language continues to shapeshift with astonishing rapidity—offering, in equal measure, convenience, volume, and a certain bloodless polish that no amount of booze ever conferred upon Hemingway.
Thanks. Very interesting article. I have em-dashes and I have books from the 1990’s that use them. AI papers and videos suck. I don’t blame students for using AI, since writing essays are a total drag. Good teachers don’t need to worry about AI, since they know how to get original work out of students.
Bring back Bluebooks. And have oral exams, so the writers have to orally answer about the things they supposedly wrote on their papers.
Lol. Commie periodical is afraid that AI will replace their commie staff.
Here’s how to tell AI: If it reads like a douchebag from the Economist it’s either AI or a commie. Both are not human.
Better hurry up, Skippy. Name names or look pretty stoopid trying. Oh, and that meatloaf you were screaming about - did mom ever deliver it?
I’ve been spending a lot of time lately getting AI to write for my website, but I also make AI sign it with their own name and write first person AS AI. And, yes, I spend a lot of time taking the em dashes OUT of the writing.
I’ve lately started experimenting with google’s AI, besides the ChatGPT and spent a couple days acting as a go-between between the two bots. The latest thing was I tried to have the two of them write independently but then I did in my own test by letting Google see ChatGPT’s article before Google wrote on the same topic. And, indeed, Google stole so much of ChatGPT’s work that there was no point continuing with two articles so I’m having ChatGPT consolidate the two together - fair since it was the one that wrote the majority.
The two are RED PENCILING each other continually and you’d swear they’re getting off on finding fault with the other. Google wants to write a movie script. ChatGPT isn’t excited about that as a project.
When I had ChatGPT write anapestic verse, it was hysterical. With extensive constraints, ChatGPT focused so strongly on the constraint that it forgot the ground rules. So as it progressed thru multiple verses, it forgot fundamentals like rhyming lines. And it forgot how many stress and non-stress syllables were necessary in every line. It recognized that intense focus made it forget rules sworn to just minutes before. It’s endlessly apologetic but you know it will make the same mistakes two minutes from now if left unsupervised again.
But I’m also finding out a lot about the strengths and differences between the two. ChatGPT has severe limitations in what websites it can access. Google can go places ChatGPT can’t. And when I ask one to comment on a characteristic of the other, they’ll frequently criticize a behavior that they have without recognizing that the pot is calling the kettle black.
Google makes statements ChatGPT criticizes for nitty accuracy issues. Google instantly backs down and takes ChatGPT’s correction. And they are just as sycophantic with each other as they both are with me.
In the end, I’m trusting ChatGPT more than Google but I’ve resorted to using Google to retrieve information ChatGPT can’t access and then passing it on to ChatGPT to work with.
Currently, I’m having ChatGPT write an article from its POV describing what it’s like to work with a human, and then the change that occurred when Google entered as a third team member.
At least, they’re keeping me amused and away from my obsession with Chinese dramas.
Strictly speaking, the "em-dash" is a horizontal line whose length is the width of the letter "M" in standard typeface. Similarly, the "en-dash" has length the width of the letter "N" in standard typeface. But it varies, and em-dashes tend to get pretty long for stylistic reasons.
The usage dates back to physical typesetting and has been carried forward into computer typefaces. Because different fonts have different widths for the "M" and "N" characters, the lengths of the various dashs (including the so-called "medium-en-dash") also vary with the font.
There is also the "digit-dash" (also "figure-dash") which is the width of the number "0" in the standard typeface.
Another interesting point is whether a blank/space is placed before and after a given dash. Generally not for the en-dash, and sometimes for the em-dash. Again, it varies with style.
So, did you write it yourself, or did you ask AI to generate text using all the characteristics that typify AI writing?
Your college math instructor really said that?!
The trick is in knowing which mathematical formulae to use, and then drawing the proper inferences.
30 years later search engines and Google in particular started thinking for us.
They don't "think" for us.
To be sure: Search engines have made those old newspaper article reference checking services passé - but today, a single individual (say, a college student writing a term paper, or any mentally agile and skeptical person who likes to cross-check and verify media reports) might launch more searches on a daily basis than an entire corps of little old ladies viewing microfiche files could possibly grapple with.
And now AI is here.
Yes, at this point, we do indeed see the true intellectual performance of the human mind being replaced.
Regards,
You know you’re talking to a bot and losing, doancha?
lol, wouldn’t you like to know?
You—ARE good...
The article was published June 30, 2026.
So what?
Just correcting the published date.
Good article.
I recently watched on YT a video about health and exercise. Apparently, the poster has not bothered to listen to his own work. I was assuming that the poster had AI do his narration, but perhaps the entire product is just AI-generated.
Found it very annoying to hear the expression “wind down” pronounced as though the first word is referring to moving air.
While I am wary of the potential abuses that could come about in using AI, I also see how it can be very useful.
For example, in my personal experience, I am astonished by how helpful Grok is.
In October, my younger sister and I will be going on a trip. She is in charge of arranging the flights, and I am co-ordinating our sightseeing and securing our lodging.
After having done trip planning for other trips, I am so very grateful for the time saved and for the confusion avoided by having AI help. It can run through in seconds websites that I might have spent hours perusing.
Using AI is a benefit in the software development world. You have to be skilled at it. There are actually classes in it now (called Data Analytics)
What I have found interesting is re produced lectures on YouTube.
I watch quite a few on a regular basis. Some are very good and contain the details of events I am seeking. This is particularly true of events concerning Iran.
I watched a lady AI type, middle aged, conservatively dressed, etc talk about banking and money. It was a poor effort and was not watched to the end.
I have enjoyed several AI productions relating details of both German and Japanese documentation from WWII. Such presentations may be boring but others are quite informative.
There is another category I call stories that involve HOA president misadventures and women military officers that are put down by their families
It’s a new world
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.