Phrases that give away AI text and what to replace them with
Most often, turns of phrase that say nothing: "in today's world", "it is important to note", "plays a crucial role", "delve into" or "in conclusion", along with words such as robust, comprehensive, innovative or seamless. The best replacement is usually none — cut the phrase, or write the concrete thing it covers up. Swapping it for another cliché only moves the problem along.
Which phrases they are
Humanisation in DetekceGPT counts set turns of phrase such as "in today's world", "it is important to note", "it is worth noting", "plays a crucial role", "delve into", "navigate the complexities", "in conclusion", "it can be stated", "pave the way", "shed light on", "a testament to" or "not only…, but also".
It also tracks individual favourite words separately: furthermore, leverage, seamless, robust, comprehensive, innovative and transformative. On their own they are fine, but in generated text they turn up densely. In the measurement they carry half the weight of whole phrases.
Why cut them, not swap them
These phrases mostly carry no content. "In today's world" adds nothing in front of a sentence about technology, and "plays a crucial role" only says that something matters instead of saying why. Cut them and the sentence gets shorter without anything missing.
Swapping in a synonym does not solve the problem: "plays a pivotal role" is the same cliché as "plays a crucial role". It is better to write what exactly is going on — not "plays a crucial role in the economy", but what share of exports it makes up or how many people it employs. The concrete figure has to be true, of course, not made up for the sake of style.
How humanisation works with them
Before the rewrite, humanisation counts the phrases and passes them to the model with an instruction to throw them out and not replace them with other clichés. After the rewrite it counts them again: the phrase check passes only when no set phrase is left in the text and at most two favourite words.
But the list only catches known phrases, and each of the four languages has its own. It does not recognise new or less common phrases, and people sometimes use the same phrases too — one "it is worth noting" does not make a text mechanical, ten do.
FAQ
Am I allowed to use these phrases at all?
Yes, on their own they are not a mistake. The problem is density — when several of them come together in a short text, it reads as generated.
Can a detector recognise a text by its phrases alone?
Not by them alone. AI phrases is one of seven detectors, and the others measure different properties, such as sentence rhythm or the deviation from your own style. A text without clichés can therefore still come out suspicious for other reasons.
Does this apply to other languages too?
Yes, each language has its own list of typical phrases. In Czech, for example, it is "v dnešní době" (in today's world), in German "spielt eine entscheidende Rolle" (plays a crucial role) or "zusammenfassend lässt sich sagen" (in summary, it can be said).
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