One could require DOIs for each reference. That's both realistic to achieve and easy to verify.

Although then why not just cite existing papers for bogus reasons?

Isn't that what GPTZero does?