What "simple trick"? The original point of FizzBuzz is not to check that the interviewee produces hyper-efficient code that avoids repeating the divisibility checks (or, for that matter, to check that they repeat the divisibility checks and are able to defend why that's simpler). It's to check that they can write down any solution at all (within reason [1]). Doing that doesn't require any tricks; you literally just need to write down the requirements in the form of code. Unless you count for loops as tricks! I know the guy who came up with FizzBuzz [2], and he did so to root people who couldn't code at all, not those who weren't good at coming up with (or memorising) coding tricks. Or, if the "trick" you're talking about is memorising, verbatim, a simple for loop solution to FizzBuzz without understanding it, then that can be countered with small variations of the problem.

I agree that some interviewers pose questions that are too hard, which ending up testing ability to perform in an unrealistic situation more than actual programming competence. Of course this happened before FizzBuzz existed. Similarly some probably set a straightforward FizzBuzz and overanalyse details rather than worrying about whether it works (or worry about silly syntax mistakes, which are perfectly reasonable on a whiteboard).

[1] http://joelgrus.com/2016/05/23/fizz-buzz-in-tensorflow/

[2] https://imranontech.com/2007/01/24/using-fizzbuzz-to-find-de...