Implement filtering tasks such that spurious sequences are sufficiently removed to meet the assumptions of the research question. The quality of metabarcoding results is likely to depend most on the appropriate inclusion of filtering into a pipeline (Calderón‐Sanou et al., 2020; Elbrecht et al., 2018; Zinger et al., 2019), so proper implementation of filtering tasks are critical for robust and harmonised use of COI metabarcoding. In metabarcoding, real amplicon sequence variants (ASVs, Callahan et al., 2017) amplified from target genes are inherently accompanied by spurious sequences, arising from multiple sources. Indeed, taxonomic inflation is a recurring issue demonstrated in communities with known haplotype composition (Creedy et al., 2020; Elbrecht et al., 2018). This can be exacerbated for mitochondrial markers like COI, due to the co-amplification of NUMTs and other non-authentic ASVs that are missed by denoising and require stringent, optimised filtering based on read abundances such as that implemented by the metaMATE software (Andújar et al., 2021). To ensure quality and reproducibility, metabarcoding studies should consider implementing the six most common filtering approaches, i.e Quality, Length, Chimera, Translation, and Frequency filtering, plus Denoising. For each of these tasks, appropriate thresholds should be considered, implemented and fully reported to a level that ensures reproducibility. Given the demonstrated importance of these tasks for most wocDNA metabarcoding studies, if any are not employed by a study the omission should be explained.