I think they key idea in this is the Ward-Takahashi identity. For correlation functions, this comes as consequence of the invariance of the functional measure of the path integral under gauge transformations. The Ward-Takahashi identity is basically quantum version of the classical Noether's theorem with a current, but now you also have contact terms.
In the context of renormalization, one direct consequence is the equality Z_{1}=Z_{2}: charge renormalization comes entirely from the photon field renormalization and the charge current is NOT renormalized. Additionally, this implies that the ratio of the charges of a quark and an electron for example are not renormalized.