Natural products are still vital sources of structurally diverse and biologically active molecules; however, these products are often hard to discover due to the complexity of natural extracts and the isolation of known metabolites. Phytochemical dereplication is an effective tool for identification of known constituents prior to extensive purification and the selection of metabolites to study for possible chemical or biological novelty. The ability for rapid phytochemical investigation has greatly increased in the last few years with the development of liquid chromatography–tandem mass spectrometry (LC–MS/MS), molecular networking (MN), and artificial intelligence (AI). The accurate mass, chromatographic and fragmentation data obtained by LC–MS/MS is then used to correlate similar metabolites into molecular families through molecular networking, and to provide chemical context for unannotated features. Additional support can be provided by machine learning (ML) and AI for spectral classification, chemical-class prediction, candidate-structure generation, similarity assessment and bioactivity based prioritization. The integration then enables metabolite detection, computational metabolite annotation, molecular family analysis, metabolite candidate ranking, targeted isolation and experimental validation to be coupled together in a workflow. However, some important limitations include incomplete spectral libraries, structural isomerism, instrumental variability, training-data bias, limited model interpretability and false annotation. This means that the predictions made from a computer model should be interpreted as a structural hypothesis, unless backed up by suitable experimental evidence. This review explores how LC–MS/MS, molecular networking and AI/ML complement each other in the process of phytochemical dereplication, how they are used in the discovery of natural products, and the challenges and opportunities of explainable AI, multimodal data integration, better spectral databases, and increasingly automated dereplication workflows.