Natural products (NPs) research is changing and rapidly adopting cutting-edge tools, which radically transform the way to characterize extracts and small molecules. With the innovations in metabolomics, early integration of deep metabolome annotation information allows to efficiently guide the isolation of valuable NPs only and, in parallel, to generate massive metadata sets for the study of given extracts under various perspectives. This is the case for chemotaxonomy studies where common biosynthetic traits among species can be evidenced, but also for drug discovery purpose where such traits, in combination with bioactivity studies on extracts, may evidence bioactive molecules even before their isolation. One of the major bottlenecks of such studies remains the level of accuracy at which NPs can be identified. We discuss here the advancements in LC-MS and associated mining methods by addressing what would be ideal and what is achieved today. We propose future developments for reinforcing generic NPs databases both in the spectral and structural dimensions by heading towards a virtuous metabolite identification cycle allowing annotation of both known and unreported metabolites in an iterative manner. Such approaches could significantly accelerate and improve our knowledge of the huge chemodiversity found in nature.