Human perceptual and cognitive abilities are limited resources. Today, in the age of cheap information --cheap to produce, to manipulate, to disseminate--, this cognitive bottleneck translates into hypercompetition for visibility among actors (individuals, institutions, etc). The same social communication incentive --visibility-- pushes actors to mutualistically interact with specific memes, seeking the virality of their messages. In turn, contents are driven by selective pressure, i.e. the chances to persist and reach widely are tightly subject to changes in the communication environment. In spite of all this complexity, here we show that the underlying architecture of the users-memes interaction in information ecosystems, apparently chaotic and noisy, actually evolves towards emergent patterns, reminiscent of those found in natural ecosystems. In particular we show, through the analysis of empirical, large data streams, that communication networks are structurally elastic, i.e. fluctuating from modular to nested architecture as a response to environmental perturbations (e.g. extraordinary events). We then propose an ecology-inspired modelling framework, bringing to light the precise mechanisms causing the observed dynamical reorganisation. Finally, from numerical simulations, the model predicts --and the data confirm-- that the users' struggle for visibility induces a re-equilibration of the network towards a very constrained organisation: the emergence of self-similar nested arrangements.