Context. The evolution of massive stars is not fully understood. Several physical processes affect their life and death, with major consequences on the progenitors of core-collapse supernovae, long-soft gamma-ray bursts, and compact-object mergers leading to gravitational wave emission.Aims. In this context, our aim is to make the prediction of stellar evolution easily comparable to observations. To this end, we developed an approach called “spectroscopic evolution” in which we predict the spectral appearance of massive stars through their evolution. The final goal is to constrain the physical processes governing the evolution of the most massive stars. In particular, we want to test the effects of metallicity.Methods. Following our initial study, which focused on solar metallicity, we investigated the low Z regime. We chose two representative metallicities: 1/5 and 1/30 Z⊙. We computed single-star evolutionary tracks with the code STAREVOL for stars with initial masses between 15 and 150 M⊙. We did not include rotation, and focused on the main sequence (MS) and the earliest post-MS evolution. We subsequently computed atmosphere models and synthetic spectra along those tracks. We assigned a spectral type and luminosity class to each synthetic spectrum as if it were an observed spectrum.Results. We predict that the most massive stars all start their evolution as O2 dwarfs at sub-solar metallicities contrary to solar metallicity calculations and observations. The fraction of lifetime spent in the O2V phase increases at lower metallicity. The distribution of dwarfs and giants we predict in the SMC accurately reproduces the observations. Supergiants appear at slightly higher effective temperatures than we predict. More massive stars enter the giant and supergiant phases closer to the zero-age main sequence, but not as close as for solar metallicity. This is due to the reduced stellar winds at lower metallicity. Our models with masses higher than ∼60 M⊙ should appear as O and B stars, whereas these objects are not observed, confirming a trend reported in the recent literature. At Z = 1/30 Z⊙, dwarfs cover a wider fraction of the MS and giants and supergiants appear at lower effective temperatures than at Z = 1/5 Z⊙. The UV spectra of these low-metallicity stars have only weak P Cygni profiles. He II 1640 sometimes shows a net emission in the most massive models, with an equivalent width reaching ∼1.2 Å. For both sets of metallicities, we provide synthetic spectroscopy in the wavelength range 4500−8000 Å. This range will be covered by the instruments HARMONI and MOSAICS on the Extremely Large Telescope and will be relevant to identify hot massive stars in Local Group galaxies with low extinction. We suggest the use of the ratio of He I 7065 to He II 5412 as a diagnostic for spectral type. Using archival spectroscopic data and our synthetic spectroscopy, we show that this ratio does not depend on metallicity. Finally, we discuss the ionizing fluxes of our models. The relation between the hydrogen ionizing flux per unit area versus effective temperature depends only weakly on metallicity. The ratios of He I and He II to H ionizing fluxes both depend on metallicity, although in a slightly different way.Conclusions. We make our synthetic spectra and spectral energy distributions available to the community.