Across life and business, some people simply think that swindling is the easiest way to get ahead – and with the advent of AI writing, it’s easier than ever to try and pass off computer-synthesised content as their personal work. AI makes new advancements every day, and soon, we can’t deny that it will be able to perfectly emulate the works and writing styles of even the best writers… but luckily for us, today’s AI is still prone to errors and some have even developed peculiar writing habits. Interested in finding out what these are and how you can become more proficient at spotting AI-generated writing? Then this article is for you!
Spotty habits
A recent update to Google Chrome integrates the powerful Gemini AI model into the system – previously only compatible with Gmail, users can now call upon a bot assistant to help them draft all sorts of written content. Given how easily accessible said tools have become, we’re inevitably going to come across AI-generated material more frequently (people are even using AI tools to help them write reviews). Although businesses and users don’t have to disclose whether or not they used AI, a close look at the writing structure and word choices can provide solid clues.
Nick Morrison, a freelance journalist and contributor of Forbes specialising in education, points to a small Cambridge study where three undergraduate students submitted ChatGPT-asissted essays that were pitted against essays written by high school students. “The ChatGPT essays performed better on average, being marked particularly highly for ‘information’ and ‘reflection’. They did poorly, however, for ‘analysis’ and ‘comparison’ — differences the researchers suggest reflect the chatbot’s strengths and weaknesses,” Morrison describes. The researchers also recognised several distinct characteristics unique to the ChatGPT essays:
Latin loaded
Despite the AI’s strengths (which are honestly just regurgitation of the reflections and perspectives of other writers), the AI’s default writing style “echoes the bland, clipped, and objective style that characterises much generic journalistic writing found on the Internet.” Most notably, ChatGPT has a high tendency to use words with Latin roots!
Despite English being the universal language, large chunks of our language are borrowed from other languages – such as Greek and Latin. Here are a few of our favourites:
· Contra (against): contradict / contraband
· Gen (to birth): generate / progeny
· Regul (to rule): regulate / regularise
· Domin (to master): dominate / dominion
· Manu (the hand): manipulate / manufacture
· Infernus (low): inferior
· Bene (good): beneficial / benefit
You might be wondering, “what’s wrong with using Latin-origin words? They might not all be everyday words, but surely there’s nothing wrong with using these!”
And you would be correct! But when you are too heavy-handed with such words, your writing starts getting cluttered. Try reading the example below:
“In the study of eugenics, geneticists believe mankind able to bring dominion over inferior genetics: to seek out and cut inferior traits, to only instill the strong and beneficial into progenies.”
Aside from sounding like a clichéd supremacist super-villain, no one we know writes like that at all – especially not an undergraduate or high school student – it also has the consequence of making the writing sound pompous. So why is it that ChatGPT uses Latin-origin words so much?
Since ChatGPT has access to all forms of publicly online information, it’s possible that the AI is drawing upon the lexicon of historians and academic experts. Don’t forget, unless you set a restriction in the prompt to force the AI to write a particular style or to write like a certain age group, the AI is free to do whatever it wants.
So how about those adverbs?
Researchers also found that the AI had the tendency to begin sentences with adverbs like however, moreover and overall. Using adverbs to link connecting sentences and thoughts are a great way to extend sentences – we do it all the time – however, they work best when you use them sparingly across the page. Excessive use of anything leads to dull and tedious copy, which might explain why some AI-written content reads like it was made using a cookie-cutter.
This and that
Numbered lists with colons as separators are the next culprit. List-making requires exceptional categorisation and organisation, and while not everyone may have those skills, AIs excel at looking for patterns and putting them together – it’s why there have been an influx of AI assistants specialising in task management and to-do lists (like todoist’s AI Assistant).
We will say, using colons to separate entries is rather unusual, as most people we know uses semicolons or foregoes punctuation between entries entirely. Numbers and line breaks are good enough already, so it seems rather redundant. And on the matter of redundancy…
Pleonasms, tautology and repetition o’plenty
Unlike anaphora (repeating the word or phrase at the beginning of several sentences or clauses) or other forms of repetition used in literary devices, pleonasms are purely redundant – most commonly seen with safe haven, RIP in peace and so forth. Tautology is similar, where you say the same thing twice using different words – like new innovations, first and foremost, close proximity etc.
Again, this phenomenon can likely be attributed to the AI picking up the habits of other writers, as best explained by Reddit user Sweet_Computer_7116:
“ChatGPT uses specific words because it’s based on a prediction model that generates text token by token from previous words… ChatGPT was trained on written text, which often includes highly educated and formal language, along with a significant amount of online text, like news articles that use buzzwords.”
If the AI detects the use of safe haven a lot across various channels of content, it may determine pleonasms and tautology to be a suitable and accepted style of writing!
Oxford comma aficionados
Writers tend to have a love-hate relationship with the Oxford comma. Some argue that it’s needed to eliminate ambiguity from lists, whereas others claim it to be akin to the Chinese metaphor “adding legs to a snake”.
In an interesting conversation between ChatGPT and fiction writer Greg Thomas, the topic of Oxford commas is brought up – the AI mentions that it isn’t specifically programmed to use the Oxford comma, but it chooses to use it for the sake of clarity.
Ultimately, whether someone uses the Oxford comma or not is out of personal preference (as we previously covered in an article), but to say that the Oxford comma is the only way to clear up ambiguity is highly debatable. A competent writer knows more than one way to structure a sentence properly… so if a writer is unable to defend themselves for using Oxford commas and/or provide alternate ways of sentence formation, then there’s a possibility they wrote their content using AI.
There’s no doubt that AI can crank out content quickly and (for the most part) effectively, but the logic and learning still has much ground to cover before it can emulate a skilled writer. Even without professional training, a regular person can easily identify the strange habits and tells of AI writing – if this is you trying to pass off synthesised content as the real deal, you better get to editing!