Many catalogues of isolated compact groups of galaxies (CGs) have been extracted using Hickson's criteria to identify isolated, dense systems of galaxies, with at least three or four galaxies concordant in magnitude and redshift. But is not clear to what extent the catalogues of CGs are complete and reliable, relative to 3D truly isolated, dense groups. Using five different semi-analytical models of galaxy formation (SAMs), we identify isolated dense groups in 3D real space, containing at least three galaxies. We then build mock redshift space galaxy catalogues and run a Hickson-like CG finder. We find that the Hickson-like algorithm in redshift space is poor at recovering 3D CGs of at least 3 galaxies, with a purity of ∼ 10% and a completeness of ∼ 22%. Among the ∼ 90% of spurious systems, typically 60% are dense structures that failed the 3D isolation criteria, while the remaining 40% are chance alignments of galaxies along the line of sight, nearly all of which are within regular groups, with some variation with the SAM used for the analysis. In other words, while only 10% of CGs are isolated dense groups, as intended, half are dense structures embedded within larger groups, and one-third are chance alignments within larger groups. The low completeness of the extracted CG sample is mainly due to the flux limits of the selection criteria. Our results suggest that a new observational algorithm to identify compact groups in redshift space is required to obtain dense isolated galaxy systems.