AI Markers in Content Writing…and what marks an article as human-written
This list includes what I think are recognisable AI markers in writing, as well as what makes an article uniquely human and nonreplicable by AI.
Pepita Maiden
10/2/20267 min read
For my series on AI in Science Communication, I’d like to tackle a topic that has been discussed a great deal this year. There have been notable instances of titles and articles pulled from publication for being AI-produced. Authors, including myself (as a test), have been putting their own, sweat-of-the-brow, creative writing into AI checkers and receiving scores that are well over 50% AI. Panic has set in, with authors wondering how they will be able to prove that their content is unique and authentic. (Nb. My recent article discusses why tech companies are dropping their AI checkers due to inaccuracy).
There’s no doubt that picking an AI-written article is getting harder. But while I struggle to know if a video with puppies is AI or not, it is still obvious to me when I look over written content. I have included what I think are recognisable AI markers in writing, as well as what makes an article uniquely human and nonreplicable by AI, at this time.
Remember that just because an article is written by a human, it doesn’t mean that it’s good! Sprinkling in mistakes and vernacular, or making an article less polished, might help the piece evade obvious AI markers, but it could still be boring, pretentious or even illogical. This particular article can’t help with poorly written, or thought-out, content!
Note that these examples work more for technical and business writing rather than fiction.
RECOGNISABLE MARKERS (THE GIVEAWAYS)
The binary formula
‘It’s not just a software update, it’s a fundamental change’.
‘It’s not just an investment, it’s a commitment’.
This binary reframe creates a false sense of depth in writing – it makes a sentence sound weighty. If this type of formula is in the article more than once, it’s not just AI-researched, it’s AI (see what I did there?) Like many of these rhetorical techniques, humans write this stuff too. Look out for these sentences and examine them. Is the meaning really as profound as it sounds? Are there more than one?
Jargon
Buzzwords and filler words tend to be overused in AI. Delve, underscore, tapestry, leverage, elevate, unlock…this sort of hyperbole makes ideas sound impactful and sophisticated while actually saying very little. Do you really need to tell your clients about Ecosystem Synergy? System, network and data sharing work just as well…and you won’t lose your audience.
Broken or fictional references
Check the sources – if there are broken links (particularly more than one), it’s the easiest way to tell an author hasn’t actually researched the article.
Generic introductory formulations
Particularly obvious throat-clearing includes:
"In today's rapidly evolving world..."
"As technology continues to transform..."
"In an increasingly complex landscape..."
"X is no longer simply a trend; it is becoming..."
Sound familiar? Humans write this crap too, but in synthetic articles you often see it in every other paragraph.
Signposting (telling then re-telling)
An AI article will often tell the reader what it’s going to say next:
"There are three key reasons why this matters."
"Let's explore what this means in practice."
"The implications are significant."
"This raises an important question."
Of course, humans do this too – it’s a good rhetorical technique - but AI does it excessively, and it really isn’t necessary. In fiction writing, your editor will cross these out with a big red metaphorical pen.
Symmetry and perfection
AI loves a perfectly symmetrically article; neat headings, every section the same length, orderly arguments. Humans are messier; they linger on interesting points and make mistakes in grammar, spelling and structure. Dot points might be missing, paragraphs will be different lengths, and punctuation will be incorrect/.
Rule of Three (tricolon)
AI compulsively uses three-part constructions. Humans like to do it too…there are three examples in the paragraph directly above (I did it deliberately, people). We write in triplets because three items connect to each other better than two or four, and because it sounds good; the sentence is balanced and it can help build intensity. But AI paragraphs use this technique over and over and over! (and it's possible that AI has never used a preposition at the beginning of a sentence - my chat search couldnt' confirm though).
Interchangeability
Read an article, for example AI in Mining, and try replacing the industry with ‘healthcare’, ‘accounting and finance’, or ‘energy’. Does the article still work? That’s a red flag. Good specialist writing contains themes and ideas that can’t be transplanted without causing logical breaks.
POSSIBLE MARKERS - THE JURY IS STILL OUT
Em-dash
I once paid an American editor to look over a piece of fiction I wrote. She spent the entire budget changing all my dashes (-) and semi-colons (;) to em-dashes(—). I complained that we don’t really use this punctuation in Australia — we don’t even have a key for them on a standard keyboard (I had to go online to find the em-dash in this sentence, and I should, in fact, have used a semi-colon or comma). She didn’t offer a refund.
American authors and content writers do use a lot of em-dashes, they seem to love them, but the AI modules go over-the-top. An amusing fiction author I follow on twitter recently joked that AI would take her em-dashes from her cold, dead hands. So overuse of the em-dash might be AI, but it also might just be American.
American spelling
This is a dead give-away for any English-language writing NOT written by an American. If you see color, honor, center and metre, all through the article, it has likely been mostly written by a US-based deep learning module.
Some spelling will be less clear; your word processor may have changed to the American spelling of defence (defense), travelled (traveled) or organise (organize). I write content for a living and I often don't pick these up.
There is also some spelling that is now generally accepted as either, such as program/programme or judgment/judgement. If you are writing, or editing, using AI, a simple hack is to simply remind your program to use non-US spelling. But the article is not necessarily AI-derived if there are just one or two incorrectly spelt words.
Conjunctive adverbs and transition words
However, importantly, notably, furthermore, additionally, ultimately. I don’t know why AI loves these so much, but they are mostly unnecessary. Humans use these too, but AI overuses them and it makes the writing feel mechanical. Here is an AI-written article in my generic website: I add these for keyword (no-one ever reads them). Take a look at the beginning of each paragraph in an article - if you notice a conjunctive adverb in literally every other paragraph, it's probably AI.
Generic expert quotes
“AI-generated quotes are semantically appropriate but evidentially empty”
Quotes within an AI-written article make it sound authoritative. Sometimes a person might be named (“Dr Sarah Thompson, a leading researcher in AI…”) but either she doesn’t exist or didn’t actually say that quote! However, many authors might quote themselves in this way to break up realms of text or emphasise a point, so its not always a marker. Writers need to make sure the quote actually means something substantive, or they need to offer a named expert with a real quote.
WHAT AI DOES NOT HAVE
Voice
Publishers will tell you they are looking for ‘voice’ in writing – both fiction and non-fiction. It’s a nebulous, inchoate thing that is hard to describe or even pinpoint. It is the signature style of the author, their personality on the page, and is shown in word choices, syntax and tone. All writing needs a narrative voice, no matter how dry your subject matter. A flat, robotic voice can be improved with emotion, opinion, sensory detail, and humour (along with bias - that’s a whole other article)!
Vernacular
In my experience, AI never uses local vernacular. In Australia, even professional language can include idioms and colloquialisms. In business, we often use terms like ‘touch base’, ‘play it by ear’ and ‘have a crack’. In Britain you might use ‘chin wag’, ‘snowed under’ or even ‘bits and bobs’. You won’t find these terms in a synthetic post, paper or article, unless they are quotes or examples.
Anecdotes
Telling a story about yourself, a friend, or a colleague, is a great way to improve personality and voice in a piece of writing. It also helps your readers connect and empathise, and it’s a simple way to show that your article is a human one. They are noticeably absent from AI written articles, but very easy to add into business writing. Eg. ‘We recently had a client who…’ or ‘Our CEO was surprised by…’
Emotion
An article written by a human will often show what the author feels. They might be irritated by a falsity, surprised by research, skeptical of a claim, or uncertain about an outcome. In AI, everything is measured, reasonable and polished, and the machine will always have the same register.
Humour
AI can add jokes to an article, and programmers are trying to teach advanced learning modules about what humans find amusing. But AI, at this time, simply cannot be sarcastic, facetious, snarky, or even genuinely funny. It doesn’t understand cultural or emotional nuance, complex irony or multi-layered wordplay. Dark and edgy humour is very human - like gallows humour, as are absurdism and inside jokes. If you want your writing to sound more human, add a joke or two, or an amusing comment. If you’ve even just smirked, once, at a comment in an article, it was probably human-written.
Unusual punctuation and filler words
The ellipsis (…) is great to show an unfinished thought and I have never seen it in AI. Same with filler words and interjections like ‘um’, ‘you know’, and ‘like’ (eg. “It isn’t true. Like, not at all”).
Swear words
I’ve never seen profanity in a synthetic article, unless used in an example. Obviously, there is a need for polite form in business and scientific writing, but adding the odd ‘hell’, ‘damn’ or ‘crap’ is a great marker for a human-written piece, and adds humour and voice to your writing.
Point of view / opinion
AI can’t create these spontaneously. An observation - and subsequent conclusion - must come from someone’s experience. If you want to make sure your writing sounds human, don’t insert mistakes. Instead offer specific knowledge, genuine editorial judgement, unexpected connections, concrete evidence, and a point of view shown with clarity and emotion:
“I think this article offers a unique perspective on AI markers in writing, coming from a communications expert and professional writer who is concerned for her future career. I wanted to show that, with a few useful examples, readers can more easily spot an AI-written article...and a human one.”
Disclaimer: Did I use AI in this article? Yes…I asked chatgpt for examples of AI tropes, and used some of them here. As always… AI is a tool, and a useful, time-saving one at that.
What do you think of my list? Did I miss any important, easy to see markers that show an article is written by AI, or a human?