Biodiversity offsets are used to mitigate the residual impacts of development on biodiversity. However, their ability to achieve no net loss is rarely evaluated, and factors leading to their success are mostly unknown. Here, we modelled the biodiversity outcomes of averted loss offsetting—in terms of vegetation extent and habitat quality—in the endangered brigalow woodlands of central Queensland, Australia. We found that biodiversity outcomes were highly sensitive to the time period used to inform counterfactual scenarios and to large differences in clearing pressures among vegetation types used for offsetting. Our results reveal major challenges for achieving no net loss of biodiversity in dynamic landscapes globally. Offsetting policies must develop plausible counterfactual scenarios—a difficult task in a volatile regulatory context—and allocate offsets according to spatially-explicit counterfactual biodiversity losses and gains. Failing to do so may drastically overestimate the expected outcomes of offsets and thus result in large net biodiversity losses.