The GA will require a fitness function, which means you need labeled data for comparison.
That conclusion is wrong. Yes, sometimes your fitness function will use labeled data. For example, if you want to train an XOR gate or any other known function.
However, there is arguably no advantage of training a function with neuroevolution versus backpropagation, except for the fact that you might discover some new architectures which solve the solution very well.
You don't always need a labelled dataset for neuroevolution.
Take for example IAMDinosaur, which trains neural networks through a genetic algorithm - however, the optimal solution is not known. There is no labelling of input data, all it does is calculate the fitness from the score.