If you don't use this particular mechanism then you could use any optocoupler.
I'd stick with a 'linear' one
The problem is knowing the characteristics and the large variance across devices so each complete unit would have to be individually calibrated.
If the OP only need to make a few, that would be OK, for volume production Id wholehardedly agree.

also, if the range to be measured is e.g. 270-330V the calibration would not be too cumbersome.

another thought: could the feedback not be PWM, capacitor coupled to and rectified in the high voltage box

Erik