Suppose you have two electrons in a state of entangled spin. Suppose also that you can measure the spin for each electron along one of two axes the x axis or the z axis. Regardless of what axis you choose to measure you will get each of the possible results with probability 1/2. If you measure the spin of electron 1 along the x axis and the spin of electron 2 along the x axis, then you get opposite results: if electron 1 has a spin pointing up, electron 2 has a spin pointing down. If you measure the spin of electron 1 along the z axis and the spin of electron 2 along the z axis, then you get opposite results: if electron 1 has a spin pointing up, electron 2 has a spin pointing down. But if you measure electron 1 along the x axis and electron 2 along the z axis then the results will match with probability 1/2.
So the probability of getting a match when you compare the results depends on whether you do the same measurement on each electron. So if the electrons were set up to have the correlations in advance, the process for setting up the electron spins would somehow have to be able to work out what measurements you're going to do in advance, which is impossible. There is some maths for ruling out loopholes in this argument, but you don't need to know that to get the gist of the problem.