Table of Contents
If the tech evangelists are to be believed, large language models are the great democratiser. Suddenly anyone with a half-formed idea and an internet connection can publish fiction, opinion pieces or website copy at a brisk pace.
The real question is whether you should bother reading something nobody bothered to write. And if not, how would you even tell? The first part is for you to wrestle with. This article will make you wise in the ways of the second.
None of what follows is to say that AI has no business near your writing. Used well, AI platforms can be genuinely useful.
You might use an AI agent to spitball ideas when you are staring down a blank page. Or as a second pair of hands when a paragraph has refused to come together for an hour. Or to summarise source material you would rather not lose an afternoon to.
The difference lies in who is doing the thinking. Use the tools to sharpen your own ideas, and the result is still yours. Hand them the reins, and you get everything described below: competent, inoffensive and hollow.
The tool isn’t the problem. Letting it do your job for you is.
With that said, here is how to tell when a piece of writing is completely and obviously machine-generated, with little to no human involvement.
Unless you are reading generic corporate copy, where blandness is part and parcel, one quick test is simple. Does the writing have any personality at all?
AI writes everything with the appearance of competence. It also goes out of its way to stay as inoffensive and non-confrontational as possible. Left to its own devices, it avoids anything that sounds too emotionally charged or too absolute.
Its sentences are clear and balanced, yet they rarely feel like something a real person would commit to the page. It also loves to explain why something is “significant” or “transformative”, often a sentence after it already made the point.
Human writers vary their sentence structure, and they know there is power in it. A short sentence can land hard. A long, carefully built sentence can become a memorable release.
AI is far more measured. Its sentences settle into a subdued, even rhythm, neither too long nor too short. That metronome quality often makes readers skim, even when the subject should grip them.
There isn’t a writer alive who hasn’t padded a word count with fluff now and then. AI takes it to another level, where you can cut every other sentence and still keep the gist intact.
Models struggle to separate crucial information from filler. They are superb at producing grammatically immaculate text that reads smoothly. Substance is still for human writers to supply and human readers to discern.
Perhaps the most dangerous tell is the confident statement of nonsense. That is harmless when you are reading a bedtime story to a sleepy toddler. It is a real problem when you are trying to learn the finer points of the Civil War.
People, dates and figures can all be fabricated, especially when the model needs them for its narrative but lacks reliable data. Nothing about the text signals the deception, which is exactly what makes it tricky.
AI: “The Treaty had a significant and transformative impact on the region, reshaping its future in profound ways.”
Human: “The Treaty redrew three borders and left two million people on the wrong side of them.”
One asserts importance. The other shows it.
When ChatGPT first arrived, you could spot its output from a mile away. The infamous em dash. The “it’s not X, it’s Y” construction. Words like “delve” and “tapestry”. Many articles outed themselves by “In today’s digital world” alone.
These tells still appear, though developers have curbed the worst of them. A single well-placed em dash now signals literacy more than machine use. But a cluster of these markers in one short text should still raise an eyebrow.
AI prioritises fulfilling the prompt with the resources you give it. It is poor at taking initiative or introducing ideas you did not ask for, even when doing so would strengthen the piece.
Ask it to write about its own transformative role in business, and you can predict the shape. It states the premise in the intro, works through a formulaic middle, then restates the obvious in the conclusion. It rarely raises a counterpoint or points you towards anything new.
| Trait | AI tends to… | Humans tend to… |
|---|---|---|
| Tone | Polished, neutral, impersonal | Have a point of view |
| Rhythm | Even, metronome-like | Vary short and long |
| Detail | Fill space with fluff | Choose specifics |
| Facts | State them confidently, right or wrong | Hedge or check |
| Ideas | Stay inside the prompt | Go off-piste when it helps |
Run a suspicious text past these questions. The more you answer “yes”, the more likely a machine did the heavy lifting.
Not sure about a piece of writing? Put it to the test.
Our free AI Detector gives you a quick second opinion, and the AI Humanizer and Paraphrasing Tool help you make AI-assisted drafts sound like you again.
No. Detectors are a signal, not a verdict. Even the better ones are known to be inconsistent, and they can flag genuine human writing as machine-made. Treat the score as one data point among several.
Not inherently. Using AI to brainstorm, unstick a paragraph or summarise reading is reasonable. The line is crossed when the tool does the thinking for you and the finished work is no longer your own.
Confident, specific claims that fall apart when you check them. Fluent text with fabricated facts is both the hardest tell to catch and the most important one to act on.
Some will fade as models improve, which is why no single marker is proof. The deeper test endures: does the writing have something real to say, or does it only sound like it does?
Telling AI-written stories apart from human ones is getting harder, and imperfect detectors do not make it easier. Paste your own writing into a few of “the most accurate” checkers and watch the confidence scores wander.
Human writing is imperfect too. It may carry a stray typo or an odd stylistic choice. It can be painfully amateur or polished to a shine. But at its best it has something meaningful to say, born of genuine experience — and that is still the hardest thing for a machine to fake.
You May Also Like