A vehicular ad hoc network (VANET) is a type of mobile network whose nodes are traveling cars and which communicate with one another using short-range wireless communications. These cars can exchange and share different information among them, which can lead to the development of interesting applications that require the cooperation of vehicles or using the vehicular network as a distributed computing platform. Along with the opportunities offered by vehicular networks, a number of challenges also arise. In this paper, we focus on information gathering in vehicular ad hoc networks. Until now, only a few research works have addressed this problem. They have lead to solutions relying on push models, where potentially useful data are pushed towards vehicles. To the best of our knowledge, no work has tackled the use of pull models in VANETs. Such models would allow users to send queries to a set of cars in order to find the desired information. In order to propose such a query processing scheme, the main challenge to address is to route the different results towards their recipient in a highly dynamic network where the nodes move very quickly. To solve this issue, we propose GeoVanet, a DHT-based geographic routing protocol which ensures that the sender of a query can get a consistent answer. Our goal is not to compute the query result "instantaneously" but to ensure that the user will be able to retrieve it within a bounded time. To prove the effectiveness of GeoVanet, an experimental evaluation is provided in the paper. It shows that up to 80% of the available query results are delivered to the user.