High-fidelity content distribution and other emerging applications of 5G and beyond-5G mobile broadband networking can put massive load on the core and radio access network (RAN). To address this, direct device-to-device (D2D) communication has recently become a first-class citizen of these networks. While information-centric vehicular networking (ICVN) based on fog computing can exploit such D2D links to alleviate the load on the RAN by proactively seeding content in the network, it has been shown that such seeding can cause even more load if performed where not needed. In addition, trying to determine where to seed content often causes additional load, negating the benefit of seeding. Therefore, in this work, we propose to adaptively seed fog nodes based on a purely virtual clustering approach. Here, vehicles are unaware of clustering decisions, so that an explicit exchange of control messages is no longer required. We show that the benefit of such an adaptive approach goes beyond simply being able to flexibly trade off one performance metric for another; in fact, it can consistently lower the load on the RAN link. We also demonstrate that this property holds even if the only available node location information is as coarsely grained as macroscale grid cells.