When the distribution of species is limited by propagule supply, new populations may be initiated by seed addition, but identifying suitable sites for efficiently targeted seed addition remains a major challenge for restoration. In addition to the biotic or abiotic variables typically used in species distribution models, spatial isolation from conspecifics could help predict the suitability of unoccupied sites. Site suitability might be expected to increase with spatial isolation after other factors are accounted for, since isolation increases the chance that a site is unoccupied only because of propagule limitation. For two native annual forbs in Californian grasslands, we combined experimental seeding and niche modeling to ask whether suitability of unoccupied sites could be predicted by spatial variables (either distances from, or densities of, conspecific populations), either by themselves or in combination with niche models. We also asked whether experimental tests of these predictions held up not only in the short term (one year), but also in the longer term (three years). For Lasthenia californica, seed additions were only successful relatively near existing populations. For Lupinus nanus, seeding success was low and was positively related to the number of conspecifics within 1 km. For both species, a few previously unoccupied sites remained occupied three years after seeding, but this subset was not predictable based on either spatial or niche variables. Seed addition alone may be a limited means of native forb restoration if suitable unoccupied sites are either rare or unpredictable, or if they tend to be close to where the species already occurs.