If I were them, I would keep some kind of hash of any stuff they have translated, so that when they index it they know if its the output of their own translator.

They probably also have ML models to try to detect machine translations too - ifa human can pick up a machine translation easily, a machine can probably be trained to detect it.

I can see why they want to do that too - multilingual data is very useful for machine language understanding (the computer effectively has two independent ways to understand whats being said), but contaminating the data with machine translations makes it nearly useless.