Aims. This work addresses the AGN IR-selection dependency on intrinsic source luminosity and obscuration, in order to identify and characterise biases that could affect conclusions in studies. Methods. We study IR-selected AGN in the Chandra Deep Field South (CDFS) survey and in the Cosmological Survey (COSMOS). The AGN sample is divided into low and high X-ray luminosity classes and into unobscured (type-1) and obscured (type-2) classes by means of X-ray and optical spectroscopy data. Specifically in the X-ray regime, we adopt the intrinsic luminosity taking the estimated column density (N H ) into account. We also take the opportunity to highlight important differences resulting from adopting different methods of assessing AGN obscuration. Results. In agreement with previous studies, we also find that AGN IR-selection efficiency shows a decrease with decreasing source AGN X-ray luminosity. For the intermediate-luminosity AGN population (43.3 < ∼ log(L X [erg s −1 ]) < ∼ 44), the efficiency also worsens with increasing obscuration (N H ). The same sample also shows an evolution with cosmic time of the obscured fraction at the highest X-ray luminosities, independently of the adopted type-1/type-2 classification method. Conclusions. We confirm that AGN IR-selection is genuinely biased towards unobscured AGNe, but only at intermediate luminosities.At the highest luminosities, where AGN IR-selection is more efficient, there is no obscuration bias. We show that type-1 AGNe are intrinsically more luminous than type-2 AGNe only at z < ∼ 1.6, thus resulting in more type-1 AGN being selected when the IR survey is shallower. Based on this and other studies, we conclude that deep hard-X-ray coverages, high-resolution IR imaging, or a combination of IR and radio data are required to recover the lower luminosity obscured AGN population. In addition, wide IR surveys are needed to recover the rare powerful, obscured AGN population. Finally, when the James Webb Space Telescope comes online, the broad-band filters 2.0 µm, 4.4 µm, 7.7 µm, and 18 µm will be essential for disentangling AGN from non-AGN dominated SEDs at depths where spectroscopy becomes impractical.