We model the amplitude line shape and absolute phase of the infrared-visible sum-frequency signals produced by a thiolated polycrystalline gold surface as a function of the visible wavelength. We follow two hypotheses: in the interband scenario, the resonant features are attributed to interband transitions, whereas in the effective surface state scenario, they stem mostly from the excitation of surface transitions. We find that both scenarios lead to a satisfactory account of the experimental data and that only free electrons may spill out of the gold bulk, as expected. For the interband scenario, the balance between free and bound electron contributions to sum-frequency generation has to be adjusted to fit the data. The surface transitions are shown to take their origin inside gold and we investigate the surface states involved in such transitions, with a comparison to the silver surfaces. We finally provide a work program dedicated to discriminate between the two scenarios.