ABSTRACT The study of planet occurrence as a function of stellar mass is important for a better understanding of planet formation. Estimating stellar mass, especially in the red giant regime, is difficult. In particular, stellar masses of a sample of evolved planet-hosting stars based on spectroscopy and grid-based modelling have been put to question over the past decade with claims they were overestimated. Although efforts have been made in the past to reconcile this dispute using asteroseismology, results were inconclusive. In an attempt to resolve this controversy, we study four more evolved planet-hosting stars in this paper using asteroseismology, and we revisit previous results to make an informed study of the whole ensemble in a self-consistent way. For the four new stars, we measure their masses by locating their characteristic oscillation frequency, νmax, from their radial velocity time series observed by SONG. For two stars, we are also able to measure the large frequency separation, Δν, helped by extended SONG single-site and dual-site observations and new Transiting Exoplanet Survey Satellite observations. We establish the robustness of the νmax-only-based results by determining the stellar mass from Δν, and from both Δν and νmax. We then compare the seismic masses of the full ensemble of 16 stars with the spectroscopic masses from three different literature sources. We find an offset between the seismic and spectroscopic mass scales that is mass dependent, suggesting that the previously claimed overestimation of spectroscopic masses only affects stars more massive than about 1.6 M⊙.