This paper shows a new concept of a virtual WWW document (VWD), as a set of WWW pages representing a logical information space, generally dealing with one particular domain. The VWD is described using metadata in the XML syntax and will be accessed through a metadata.class file, stored at the root level of WWW sites. We’ll suggest how the VWD can improve information retrieval on the WWW and reduce the network load generated by the robots. We describe a prototype implemented in JAVA, within an application in the environmental domain. The exchanges of such metadata lay in a flexible architecture based on two kinds of robots : generalists and specialists that collect and organize this metadata, in order to localize the resources on the WWW. They will contribute to the overall auto-organizing information process by exchanging their indices, therefore forwarding their knowledge each other.