How to spot AI writing

What it means here:The article's point is that every writer — human or bot — has personal quirks of style. Emily Dickinson scattered em-dashes everywhere; each AI model has its own habits too. So there is no single "AI style" you can hunt for.

In general:A small habit, preference, or way of doing things that is distinctive to one particular person — a personal quirk.

What it means here:For all its long, impressive-sounding words, AI prose doesn't actually make ideas easy to grasp — the article says it lacks the clearness that marks genuinely good writing.

In general:Clarity — the quality of being easy to understand, or of thinking clearly. From the Latin for "light": lucid writing lets the light through.

What it means here:Orwell's "pontificating penmen" are writers holding forth self-importantly — dressing simple ideas in grand words to sound wise and authoritative.

In general:To speak or write in a pompous, know-it-all way, as if your opinions were official truth. It comes from pontiff — the pope — so to pontificate is literally to talk as though you were issuing decrees from the throne.

Pretentious diction is a famous idea from writing and rhetoric, named by George Orwell in his 1946 essay Politics and the English Language. It means dressing up a simple statement in big, important-sounding words — wearing a tuxedo to gym class. A pretentious writer never uses a thing, they utilize it; nothing starts, it commences; rain doesn't fall, there is significant precipitation.

Orwell noticed that weak writers reach for long Latin-derived words (utilize, facilitate, commence) because they sound grander than plain old English ones (use, help, start). But the plain word is almost always stronger: it reads faster, it's harder to hide behind, and it proves you actually understand what you're saying. Compare "I facilitated the distribution of nutritional resources" with "I handed out lunches" — the second writer sounds like a human who was actually there.

The article's twist is that AI models are champion offenders — stuffing prose with significant, increasingly and methodology — which is exactly why AI writing can sound impressive at first glance while saying very little. The lesson for your own essays: wherever a short word could replace a long one, the short word wins.

Corpus analysis is how linguistics borrows the tools of statistics: instead of judging writing by feel, you treat text as data. A corpus (Latin for "body") is a huge body of text — thousands of sentences — in which you can count things: how often each word appears, how long sentences run, how many commas show up per paragraph.

The key move is comparing your mystery text against a baseline — a sample whose origin you already know. One clue proves nothing: a text isn't AI-written just because it says delve, any more than a novel is by Jane Austen because it says imprudence. But if a pile of text uses significant three times as often as a million words of known-human writing, consistently, across thousands of sentences — that's a fingerprint, not luck. That is exactly what The Economist did: it had chatbots rewrite its own articles, compared the two piles across 1.2 million words, and even cross-checked against CNN and the New York Times so it wouldn't mistake its own quirks for AI's.

The same logic runs everywhere: plagiarism checkers compare your essay against millions of documents, forensic linguists identify anonymous writers by their word habits, and scholars use it to argue over which plays Shakespeare really wrote. The transferable rule: before you can call something unusual, you have to measure what usual looks like.