Quantum chemistry embedding methods have become a popular approach to calculate molecular properties of larger systems. In order to account for finite temperature effects, including both configurational and conformational averaging, embedding methods are often combined with molecular dynamics (MD) simulations either in a direct or sequential manner. One of the decisive factors for a successful application of embedding methods is that that the underlying structures provided by the MD simulation are accurate, if not this will result in low-quality prediction of the molecular properties in question. Here we investigate different approaches for generating a set of molecular structures to be used in subsequent embedding calculations ranging from classical MD using a standard molecular mechanics (MM) force field to combined quantum mechanics/molecular mechanics (QM/MM) MD. Overall, we find an intermediate approach relying on classical MD followed by a constrained QM/MM geometry optimization to be a fairly accurate and very cost-effective approach, although this procedure naturally leads to underestimation of, for example, spectral bandwidths.