At What Point Do We Stop Assuming Polished Writing Must Be AI?
I have been writing for nearly twenty years. It was, for a long time, a professional requirement. When you work as a travel photographer, the images rarely sell themselves in isolation. You write the article that accompanies them.
You learn, over time, to construct a sentence that does something useful, to vary your rhythm, to not begin three consecutive paragraphs with the word “the.” It is not magic. It is just practice, in the same way that a plumber who has been unblocking drains for two decades gets reasonably good at unblocking drains.
On a recent article, several comments left on a social page suggested it had been written by AI. Their evidence? I had used a subordinate clause. I had, apparently, also committed the additional crime of deploying a metaphor. Case closed.
The New Literary Crime: Knowing What You Are Doing
I want to be clear that I understand the suspicion. The internet is awash with machine-generated content. Articles that begin with “In today’s fast-paced world” and end with “In conclusion, it is clear that…” are everywhere.
The concern is legitimate. The problem is that we appear to have overcorrected so dramatically that the mere act of writing a coherent paragraph is now considered circumstantial evidence of artificial intelligence. We have, in other words, collectively decided that competence is suspicious.
A Brief and Entirely Unscientific Observation
I have noticed that the accusation tends to arrive in one of two forms. The first is the single-word verdict: “AI.” It is posted with the confident finality of a judge delivering a sentence, no further elaboration offered or considered necessary.
The second is slightly more theatrical: “This is definitely ChatGPT, I can tell.” This version is my personal favourite.
It implies a level of forensic expertise that the commenter almost certainly does not possess, delivered from an account whose most recent post was a photograph of a plate of nachos.
What I find genuinely fascinating is the implied standard. If polished writing must be AI, then what, precisely, is the acceptable human register?
I have been doing some research, and it appears the answer is: typos, sentence fragments, and the liberal use of “ngl” and “lowkey.” Anything beyond that, any attempt at well-structured text, and you are apparently a language model.
Exhibit A: The Wizarding World Is 91% Robot
Out of curiosity, and a mild sense of grievance, I decided to put this to a practical test. I took a passage from Harry Potter and the Philosopher’s Stone, a book that has sold over 120 million copies and is widely considered one of the most beloved works of children’s fiction ever written, and ran it through a free AI content detector.
The passage in question describes Hagrid knocking on the castle door, Professor McGonagall appearing in emerald-green robes, and Harry entering the Entrance Hall for the first time. It is, by any reasonable measure, some of the most evocative and distinctly human storytelling of the last thirty years.
The detector’s verdict? 91% AI-generated. Zero percent human-written.
The tool’s reasoning was, I must say, delivered with tremendous confidence. It noted that the average sentence length of 29.2 words was “close to the AI average,” and that the use of a multi-clause compound sentence was “characteristic of AI text.” J.K. Rowling, it concluded, writes like a machine.
I sat with this information for a moment. Then I made another cup of tea.
The Detector Has Spoken
What the detector has inadvertently demonstrated, with the precision of a very expensive piece of software, is that it cannot distinguish between human writing vs AI at all. It has simply been trained to flag anything that is grammatically coherent, structurally varied, and descriptively rich.
In other words, it has been trained to flag good writing.
This is, if you think about it, a rather extraordinary achievement. We have built a tool that is functionally useless for its stated purpose, and we are using it to adjudicate the work of living writers. The irony would be delicious if it were not also mildly alarming.
The Goldfish in the Room
There is, I think, a more uncomfortable truth lurking underneath all of this. We have spent the last decade consuming content in formats that reward brevity above all else.
The average social media post is not designed to be read. It is designed to be processed in approximately the time it takes to blink.
We have become, as a species, extraordinarily efficient at absorbing information in small, digestible units. We are correspondingly less comfortable with anything that requires sustained attention.
A long-form piece of writing, properly structured and carefully edited, now reads as alien to a significant portion of the internet. Not because it is inhuman, but because it is unfamiliar.
It is the literary equivalent of handing someone a novel and watching them look for the scroll bar.
The Irony Is Not Lost on Me
I am aware, of course, of the delicious irony of writing a well-structured, carefully punctuated essay about people who make AI writing assumptions. If this post itself gets flagged as machine-generated content, I will consider it a form of performance art and move on with my life.
But I do think it is worth asking the question seriously, even if I am asking it in a slightly frivolous way. When did we decide that quality was a red flag? When did the bar for human writing vs AI drop to the point where a correctly deployed semicolon triggers an investigation?
I have been writing since before most of these commenters were old enough to have a social media account. I have notebooks filled with drafts, crossings-out, and margin notes that would make a graphologist weep.
A Modest Proposal
Perhaps, before concluding that something was written by a machine, we might consider the radical alternative. That it was written by a person who reads books.
A person who has, over many years, absorbed enough language to deploy it with some degree of facility. A person who, in the course of a long career, learned that a sentence can do more than one thing at once.
Or, alternatively, we could all just keep posting “AI detected” under anything longer than three sentences, and see where that gets us. I am sure it will be fine. Ngl.
If you prefer your travel writing to be demonstrably human, occasionally cynical, and written by someone who has actually been there, you might enjoy reading about The Observations of an Introvert in Chiang Mai or my thoughts on The Observations of an Introvert in Kyoto.
Alternatively, if you are looking for practical advice that hasn’t been scraped from a Wikipedia page, take a look at my guide on Solo Travel Over 50: Why Now Is the Perfect Time to Go It Alone.
A note on how I write: Every trip starts with a journal, filled in wherever I happen to be; on a bus, plane, train, or in a backstreet cafe that has clearly seen better days. It’s not tidy, and it’s not always legible even to me. When I get the time, I use AI to help turn those notes into something readable, organising drafts and occasionally rescuing a sentence that has wandered off without me. The experiences, the perspective, the humour and every editorial decision remain entirely mine. As for the photos, I’m not remotely fond of having my picture taken, family portraits included, so I also use AI to help visualise the places and moments I write about.
