Sowing is a well-established restoration technique to overcome dispersal limitation. Site-specific seed mixtures are most effective to achieve functional communities. This is especially important if the restored vegetation has to protect critical infrastructure like roadsides and dikes. Here, an improved seed-substrate combination will secure slope stability, reduce mowing efforts, and generate species-rich grasslands. A factorial field experiment addressed this topic on a dike at River Danube in SE Germany in 2018-2021. Within 288 plots, we tested three sand admixtures, two substrate depths, two seed densities and two seed mixture types (mesic hay meadow, semi-dry calcareous grassland) in north and south exposition, and measured the recovery completeness by calculating the successional distance to reference sites, the persistence of sown species, and the Favourable Conservation Status (FCS) of target species. Overall, the sown vegetation developed in the desired direction, but a recovery debt remained after four years, and some plots still showed similarities to negative references from ruderal sites. In north exposition, hay meadow-seed mixtures developed closer to the respective reference communities than dry-grassland mixtures. In south exposition, the sown communities developed poorly which might be due to a severe drought during establishment. This initial negative effect remained over the entire observation period. Sand admixture had a slightly positive effect on target variables, while substrate depth, seed density and mixture type had no effects on species persistence or FCS. Synthesis and applications: Site-adapted seed mixtures make restoration more effective. However, applying several seed-substrate combinations might foster beta diversity. Furthermore, additional management efforts are recommended, as they might be necessary to reduce the recovery debt, as well as re-sowing after unfavourable conditions.