How to spot AI text by eye: nine signs to watch for
Generated text tends to be even, cautious and general. What gives it away most is sentences of the same length, a missing concrete detail, smooth transitions between paragraphs and a summary that summarises nothing new. None of these signs is proof, but together they form a picture.
Rhythm and construction
Sentences of similar length. Paragraphs of similar size. Triplets in lists, even where two items would do.
Transitions between paragraphs that are smooth but add nothing.
Content
A concrete detail is missing — a name, a date, a number, an anecdote. The text speaks about the topic in general and avoids taking a position.
The conclusion summarises what has already been said instead of closing anything.
Vocabulary
Cautious phrasing such as "it is important to mention" or "plays a key role". Overly balanced sentences where every claim is followed by a counterweight.
Odd combinations: very formal language next to small grammatical slips that a native speaker would not make.
FAQ
Is this enough instead of a tool?
For short texts, often yes. Across thirty papers, a tool tells you which nine to read carefully.
Couldn't a person write like that too?
They could, and that is exactly why none of these signs is enough on its own.
Are the models getting better at this?
Yes. Signs that held a year ago hold less today. A concrete detail and a position of one's own, though, are still usually missing from generated text.
Try it on your own text
Paste a text or upload a file and look at the score and at the specific sentences that came out suspicious.
Run a detection