The detection of the binary neutron star (BNS) merger GW170817 and the associated electromagnetic (EM) counterpart, the 'kilonova' (kN) AT2017gfo, opened a new era in multimessenger astronomy. However, despite many efforts, it has been proven very difficult to find additional kNe, even though LIGO/Virgo has reported at least one BNS event during their latest run, O3. The focus of this work is the exploration of the sensitivity of the adopted optical surveys searching for kNe during O3. We propose ways to optimize the choices of filters and survey depth to boost the detection efficiency for these faint and fast-evolving transients in the future. In particular, we use kN models to explore the dependence on ejecta mass, geometry, viewing angle, wavelength coverage, and source distance. We find that the kN detection efficiency has a strong viewing-angle dependence, especially for filters blueward of i-band. This loss of sensitivity can be mitigated by early, deep, observations. Efficient gri counterpart searches for kNe at similar to 200 Mpc would require reaching a limiting magnitude m(lim) = 23 mag, to ensure good sensitivity over a wide range of the model phase-space. We conclude that kN searches during O3 were generally too shallow to detect BNS optical counterparts, even under optimistic assumptions.