There was a time when the question of whether a piece of writing was original was fairly simple. A writer wrote it, an editor reviewed it, and a reader trusted that the words on the page came from the person whose name appeared above them.
Artificial intelligence has made that assumption harder to maintain.
Today, a writer can use AI to brainstorm an idea, organise research, check grammar, challenge an argument or rewrite a difficult paragraph. Another writer can give an AI tool a prompt and publish almost everything it produces. Both have technically used AI, but treating the two situations as identical does not tell us much about who actually wrote the work.
That distinction is becoming increasingly important as AI-generated content becomes harder to separate from human writing. It is also raising an uncomfortable question for writers: if readers can no longer tell how a piece was produced, should writers be expected to disclose their use of AI or even prove that they did not use it?
When does AI assistance become AI authorship?
The difficulty starts with the word use.
A writer who asks an AI tool to suggest five possible headlines or refine paragraphs that appear unclear after writing an article has used AI. So has a writer who meticulously prompts a machine to produce a 2,000-word draft. But the contribution of the technology in those two cases is clearly different.
The first writer, after going through the rigorous process of writing — prewriting, drafting, revising and editing — has used the tool to enhance their work.
The second writer may have handed over much of the actual writing process, particularly the prewriting stage, which can involve researching and outlining, to the machine.
But at what point does that assistance become substantial enough to require disclosure? At what point should a writer add a footnote saying, “This article was assisted by AI”?
We are learning to recognise ‘AI writing’, but that does not mean we can prove it
Ever since the first version of ChatGPT was released by OpenAI in November 2022, the internet has, for the most part, been awash with AI-generated content, much of which has come to be known as AI slop.
A recent study by Pew Research Center, which sampled 10,000 webpages collected in July 2026, found that one in 10 showed signs of being written or substantially edited by AI.
The ubiquity of AI-generated content has forced readers to develop ways to sniff out what machine-written content looks or sounds like.
At first, people claimed that the easiest giveaway that written content was generated by AI, particularly ChatGPT, was the presence of an em dash.
The disdain for the em dash became so widespread, and triggered such a backlash, that writers who would ordinarily use it to spice up their articles began to avoid it. This is despite the fact that some of the world’s greatest writers and publications have used the em dash beautifully for centuries.
As the models got better, people began to flag other patterns as indications that an article had been prompted into existence rather than written by an actual human.
Sometimes it is the structure. The writing moves through perfectly organised sections, with each paragraph neatly leading into the next.
Sometimes it is the language. The sentences sound polished but strangely impersonal, filled with broad statements that could apply to almost any subject.
There can also be repetition, predictable transitions and conclusions that simply restate everything the article has already said.
These patterns have become familiar enough that people now regularly describe writing as “AI-generated” based on how it sounds.
The problem is that sounding like AI is not evidence that AI wrote something.
A human writer can be repetitive. A human writer can use generic language. A human writer can produce an overly polished article. A writer who has spent years producing corporate or SEO content may naturally have a style that resembles the patterns people now associate with AI.
The reverse is also true. AI can be prompted to produce writing that sounds informal, personal or deliberately imperfect.
That makes the identification problem much harder.
AI detection tools have also attracted criticism because determining whether text was generated by AI is not the same as determining whether the text sounds like AI. A system can flag something because it detects patterns associated with machine-generated writing, but that does not establish who actually produced the words.
This creates a troubling possibility for writers.
If ‘AI-like’ becomes a sufficient reason to question authorship, a writer may eventually find themselves defending a piece not because there is evidence that a machine wrote it, but because someone believes the writing does not sound human enough.
That changes the burden of proof.
Instead of asking, ‘Was this written by AI?’, we could end up asking the writer, ‘Can you prove that it wasn’t?’
And proving that is far more difficult than detecting a suspicious-looking paragraph.
So what exactly should writers disclose?
Writers who want to distinguish themselves are already going the extra mile to show that their work is genuinely theirs.
For instance, Sarah Suzuki Harvard, a professional copywriter, told The Wall Street Journal that she sometimes has to go ‘rogue’ to prove she wrote a piece.
“I’ll use aggressively casual language, like ‘hey yo, for real,’ or drop a bunch of exclamation points. It feels so icky to do this, but it’s what you have to do to sound human,” the 32-year-old said.
While Harvard tried to use casual writing to show that her work was original, podcaster Steven Bartlett once declared in a LinkedIn post, ‘This was not written with AI.’ He also advised companies to ban team members from using AI to write important ideas.
What this shows is that there is no blanket rule writers can use to disclose or prove that an article was generated entirely by AI, written entirely by a human, or merely assisted by AI.
For all we know, Harvard’s ‘hey yo, for real’ could itself be AI-generated. Tools such as ChatGPT, Gemini and Claude, among many others, can be prompted to sound exactly like that.
Moreover, a rule requiring disclosure only when AI writes a certain percentage of a piece would create another problem: how do you measure that percentage?
- Is a paragraph generated by AI but completely rewritten by the writer still AI-generated?
- What if AI reorganises a writer’s original work but adds no new information?
- And who decides when the contribution is substantial enough to disclose?
- How should writers disclose their use of AI? Should they say, ‘AI assisted in writing this’, or adopt Steven Bartlett’s approach: ‘This was not written by AI’?
Amid that confusion, another argument emerges.
If the purpose of writing is to inform, entertain and educate — or a combination of all three and more — does it really matter whether a work was written by a human or AI, as long as it accomplishes its purpose?
When you walk into a car dealership, do you ask whether the parts were assembled by robots, or do you ask whether the car will serve the purpose for which you are buying it?
Before the invention of the first mechanical calculator, the Pascaline, in 1642 by French mathematician Blaise Pascal, almost all calculations were certainly done by humans. There was no need to question whether someone had thought through the mathematics to arrive at an answer.
Almost four centuries later, calculators are used without mathematics professors being asked to disclose whether they arrived at a particular answer by punching numbers into a machine.
Also Check: Should I Stop Using Em-dash?
Human writing still counts
None of what has been said above diminishes the importance of writing that is actually done by human beings. In fact, perhaps surprisingly, the demand for storytellers has risen in recent years as AI-generated content floods the internet.
Job postings in the United States that included the term ‘storyteller’ doubled in the year ending November 26, according to a Wall Street Journal article published in December 2025.
This is partly because AI, despite its growing sophistication and its ability to generate content that some people increasingly find difficult to identify as machine-written, cannot yet compete with the finest human writers.
Also Check: When to—and When Not to—Use the Em-Dash
A recent study reported by The Economist appears to support this argument, suggesting that “AI prose is distinguishable by word and punctuation choice as well as sentence and paragraph structure.”
One reason may be that AI-generated content can be formulaic, following familiar patterns, structures and linguistic conventions that can sometimes limit originality or creativity.
Perhaps the best way to avoid sounding like AI — and the headache of having to explain whether you used it — is simply to be yourself.
Study your own writing. Invest in the thinking process. Read voraciously. Let your thoughts flow through your pen without constantly worrying about whether your writing sounds like a machine.
To go further, consider embracing the full range of punctuation available to you. The Economist study found that AI tends to use fewer commas and semicolons than humans, and hardly any parentheses.
So, the next time you feel the urge to declare that your work is not AI-generated, remember this: although there are no universal metrics for proving authorship, there are still things human writers bring to the page that machines have not fully replicated.
The best evidence of your humanity may not be a disclaimer.
It may simply be your writing.
