Vehicular ad-hoc networks (VANETs) inflict smart traffic control using information exchange (i.e., warning messages, routing and network discovery messages) with nearby vehicles immediately in a highly dynamic environment. Prior approaches have proposed several routing computations and discovery techniques using prediction, direction aware, and hello message assisted algorithms. However, most of them do not perform adequately in real-life traffic situations and have various concerns i.e., broadcast storming, unpredictable network dynamics, fixed-route life-time problem, non-linear real-world traffic, high routing overhead, and low packet delivery ratio. Clustering architecture in vehicular scenarios improves resource utilization, network scalability and flow of data between communicating vehicles but stable cluster formations is also a challenging task in real-life traffic situations. Therefore, in this paper, we proposed a cosine similarity based selective broadcast routing protocol, also known as CSBR, which leverages non-linear cluster formation ability using cosine similarity index. Distinct clusters and the coordinating vehicles assist each other in finding the most suitable path to reach the destination. Additionally, a probabilistic forwarding approach is used to disseminate routing messages further in the network. The outcomes exhibit that the proposed scheme improved 5-10% packet delivery fraction (PDF), minimizes average delay approximately 25%, up to 10% low communication overhead, improved throughput upto 5-10%, and less neighbour discovery messages overhead compared to the existing broadcast routing protocols in VANETs. © 2020 IFIP.