Biodiversity offsetting is usually the last step in the mitigation hierarchy and aims to compensate for impacts of development projects on biodiversity. It is supposed to contribute to the key environmental objective of “no net loss” of biodiversity by delivering gains equivalent to losses. We hypothesize that such gains can only be attained through ecological restoration of degraded sites: the restored ecosystem should not only equal the original or reference ecosystem as usually assumed, but rather the original state of degradation of the ecosystem used for offsetting should be of the same level as the impacted ecosystem after development. We built on this starting assumption to determine whether impacts and gains were considered equally in the offsetting measures of 24 infrastructure projects, and to infer the potential gains in offset sites, based on an analysis of procedure and administrative documents. The analysis showed that impacts were presented in much more detail than the offsetting measures. In addition, out of 577 ha that was intended to offset areas being artificialized, only 3% of the area was artificial prior to offsetting work, i.e. delivering high potential gains, while 81% could be considered semi-natural habitats, thus with lower potential gains. Little information on the ecological quality of offset sites was available. When described, their good quality was used as an argument to justify their selection, resulting in relatively uncertain gains in comparison to certain impacts. Our results suggest that including multiple comparisons of multiple ecosystem states is a way forward to better evaluate the equivalence between gains and losses, and thus would ensure no net loss of biodiversity.