We show that a simple first-principles correction based on the difference between the singlet-triplet CO excitation energy values obtained by density-functional theory (DFT) and high-level quantum chemistry methods yields accurate CO adsorption properties on a variety of metal surfaces. We demonstrate a linear relationship between the CO adsorption energy and the CO singlet-triplet splitting, similar to the linear dependence of CO adsorption energy on the energy of the CO $2\ensuremath{\pi}*$ orbital found recently [Kresse et al., Phys. Rev. B 68, 073401 (2003)]. Converged DFT calculations underestimate the CO singlet-triplet excitation energy $\ensuremath{\Delta}{E}_{\mathrm{S}\ensuremath{-}\mathrm{T}},$ whereas coupled-cluster and configuration-interaction (CI) calculations reproduce the experimental $\ensuremath{\Delta}{E}_{\mathrm{S}\ensuremath{-}\mathrm{T}}.$ The dependence of ${E}_{\mathrm{chem}}$ on $\ensuremath{\Delta}{E}_{\mathrm{S}\ensuremath{-}\mathrm{T}}$ is used to extrapolate ${E}_{\mathrm{chem}}$ for the top, bridge, and hollow sites for the (100) and (111) surfaces of Pt, Rh, Pd, and Cu to the values that correspond to the coupled cluster and CI $\ensuremath{\Delta}{E}_{\mathrm{S}\ensuremath{-}\mathrm{T}}$ value. The correction reproduces experimental adsorption site preference for all cases and obtains ${E}_{\mathrm{chem}}$ in excellent agreement with experimental results.