Antimicrobial peptides (AMPs) are known to selectively bind to and kill microbes over host cells. Contrary to a conventional View, there is now evidence that AMP's cell selectivity varies with cell densities and is not uniquely determined. Using a coarse-grained model, we study how the cell selectivity of membrane-lytic AMPs, defined as the ratio between their minimum hemolytic (MHC) and minimum inhibitory concentrations (MIC), depends on cell densities or on the way it is measured. A general picture emerging from our study is that the selectivity better captures peptide's intrinsic properties at low cell densities. The selectivity, however, decreases and becomes less intrinsic as the cell density increases, as long as it is chosen to be the same for both types of cells. Importantly, our results show that the selectivity can be excessively overestimated if higher host cell concentrations are used; in contrast, it becomes mistakenly small if measured for a mixture of both types of cells, even with similar choices of cell densities (i.e., higher host cell densities). Our approach can be used as a fitting model for relating the intrinsic selectivity to the apparent (cell-density-dependent) one.