Content distribution using direct device-to-device (D2D) communication is a promising approach for optimizing the utilization of air-interface resources in 5G network. To achieve the best utilization of network resources, the network should bias the distribution of cached content. The optimal caching distribution depends on the content demand distribution. In this paper, we model the locations of transmitting and receiving devices by a homogeneous Poisson Point Process (PPP), and use results from stochastic geometry to derive the probability of successful content delivery in the presence of D2D interference and noise. We formulate an optimization problem to maximize the total probability of content delivery, assuming user demands are modeled by a Zipf distribution. We also develop strategies to optimize content caching, and bound the total coverage probability for multiple file caching scenarios.