I doubt they'd just add a UUID in a file header somewhere. If they uniquely modify the actual audio samples in a way that is inaudible during casual listening, that would be much harder to "diff", I think.
Can you guys catch up on the absolute basics of the last twenty years of audio watermarking research before continuing this conversation please dear god
Are you saying that two audio files each watermarked for a specific different user will not show a difference? Because then I am genuinely interested in how they achieved that feat