It is well established that social conditions often modify foraging behaviour, but the theoretical interpretation of the changes produced is not straightforward. Changes may be due to alterations of the foraging currency (the mathematical expression that behaviour maximizes) and/or of the available resources. An example of the latter is when both solitary and social foragers maximize rates of gain over time, but competition alters the behaviour required to achieve this, as assumed by ideal free distribution models. Here we examine this problem using captive starlings Sturnus vulgaris. Subjects had access to two depleting patches that replenished whenever the alternative patch was visited. The theoretical rate–maximizing policy was the same across all treatments, and consisted of alternating between patches following a pattern that could be predicted using the marginal value theorem (MVT). There were three treatments that differed in the contents of an aviary adjacent to one of the two patches (called the ‘social’ patch). In the control treatment, the aviary was empty, in the social condition it contained a group of starlings, and in a non–specific stimulus control it contained a group of zebra finches. In the control condition both patches were used equally and behaviour was well predicted by the MVT. In the social condition, starlings foraged more slowly in the social than in the solitary patch. Further, foraging in the solitary patch was faster and in the social patch slower in the social condition than in the control condition. Although these changes are incompatible with overall rate maximization (gain rate decreased by about 24% by self–imposed changes), if the self–generated gain functions were used the MVT was a good predictor of patch exploitation under all conditions. We discuss the complexities of nesting optimal foraging models in more comprehensive theoretical accounts of behaviour integrating functional and mechanistic perspectives.