Questions (a) How do the facilitative effects of benefactors on understorey species change along a successional gradient; (b) what are the feedback effects of the beneficiary species on the benefactor; and (c) how do these bidirectional interactions drive community dynamics of a shrub‐dominated ecosystem. Location A sand‐fixing community, Mu Us Desert, China. Methods We quantified the effects of the shrub Artemisia ordosica on understorey herb abundance, biomass and richness using an observational procedure at four positions along a successional gradient, i.e., semi‐fixed dunes (D1 stage), fixed dunes (D2 stage), fixed dunes covered with biological crusts (D3 stage) and fixed dunes with herbs (D4 stage). The relative interaction index (RII) was used to calculate the effect of the shrub on the herbs. Additionally, we conducted a removal experiment at the latest successional stage to assess the beneficiary feedback effects on the growth of the shrub's vegetative and reproductive twigs. Results The response of herbs to the effect of the benefactor shrub showed a humped‐back curve for abundance and biomass along the successional gradient, supporting the collapse of the interaction model, but a linear curve for richness, thus supporting the stress‐gradient hypothesis. This difference could likely be explained by a turnover of life‐history groups of understorey species from annuals to perennials along the successional gradient, with the growth of the latter group being more susceptible to shrub facilitation than the former. The herbs had a negative feedback effect on the growth of the shrub's reproductive twigs. Conclusions Our results suggest that the relationship between the shrub A. ordosica and the associated herbs was antagonistic, and the absence of shrub seedling recruitment at the end of succession suggests that this antagonistic relationship was a crucial driver of a successional shift from a shrub‐dominated to a herb‐dominated community.Read less <