Context. The neutral intergalactic medium above redshift ∼6 is opaque to ionizing radiation, and therefore indirect measurements of the escape fraction of ionizing photons are required from galaxies of this epoch. Low-ionization-state absorption lines are a common feature in the rest-frame ultraviolet (UV) spectrum of galaxies, showing a broad diversity of strengths and shapes. As these spectral features indicate the presence of neutral gas in front of UV-luminous stars, they have been proposed to carry information on the escape of ionizing radiation from galaxies.Aims. We aim to decipher the processes that are responsible for the shape of the absorption lines in order to better understand their origin. We also aim to explore whether the absorption lines can be used to predict the escape fraction of ionizing photons.Methods. Using a radiation-hydrodynamical cosmological zoom-in simulation and the radiative transfer postprocessing code RASCAS we generated mock C II λ1334 and Lyβ lines of a virtual galaxy at z = 3 with M1500 = −18.5 as seen from many directions of observation. We also computed the escape fraction of ionizing photons in those directions and looked for correlations between the escape fraction and properties of the absorption lines, in particular their residual flux.Results. We find that the resulting mock absorption lines are comparable to observations and that the lines and the escape fractions vary strongly depending on the direction of observation. The effect of infilling due to the scattering of the photons and the use of different apertures of observation both result in either strong or very mild changes of the absorption profile. Gas velocity and dust always affect the absorption profile significantly. We find no strong correlations between observable Lyβ or C II λ1334 properties and the escape fraction. After correcting the continuum for attenuation by dust to recover the intrinsic continuum, the residual flux of the C II λ1334 line correlates well with the escape fraction for directions with a dust-corrected residual flux larger than 30%. For other directions, the relations have a strong dispersion, and the residual flux overestimates the escape fraction for most cases. Concerning Lyβ, the residual flux after dust correction does not correlate with the escape fraction but can be used as a lower limit.