We address in this work the question to which extend reaction field schemes for correlated wave function methods give accurate excitation energies and, at the same time, physically consistent potential energy surfaces. The performance of the perturbation on energy (PTE), perturbation on energy and density (PTED), and post-SCF reaction field schemes is compared for the algebraic diagrammatic construction through second-order, ADC(2), as electronic structure and the conductor-like screening model COSMO as solvation model. The conditions on reaction field schemes to give physically consistent potential energies surfaces are discussed at the example of 4-(N,N-dimethylamino)benzonitrile, which is used as a test case to assess the artifacts introduced by state-specific contributions to the effective Hamiltonian. To evaluate the accuracy for excitation energies, we use two benchmark sets with data in gas phase and solution for ππ* and nπ* electronic transitions. The experimental solvatochromic shifts are compared to the corresponding calculated values at the COSMO-ADC(2) level with the PTE scheme within the frozen solvent approximation, PTED with the linear response (LR) and corrected linear response (cLR) and post-SCF with LR schemes and with the approximate coupled-cluster singles and doubles method CC2 combined with COSMO in the post-SCF (LR) scheme. The PTE scheme gives at the COSMO-ADC(2) level less accurate solvent shifts than the PTED(LR), PTED(cLR), and post-SCF(LR) schemes. The most accurate prediction of solvatochromism is obtained with the post-SCF(LR) scheme. In most cases, PTED(cLR) performs similar to post-SCF, although its nonlinear perturbative correction causes problems for potential energy surfaces.