In this paper, we present a promising multi-agent self-organizing ant inspired approach applied to unstructured peer-to-peer networks (no global or distributed map). This work shows enhancement resulting from the ability of the system to self-organize, and contribute to understand its intrinsic properties. For the selected self-improving approach, we show how the process of information search builds an inner topology deeply correlated to the network topology. Consequently, the efficiency of search algorithms is considerably enhanced, particularly for rare data. We show how ant inspired algorithms can build well adapted scale-free-like topologies for search in unstructured peer-to-peer network, and show further improvements considering the initial topology of the network.