The abyssal seafloor covers more than 50% of planet Earth and is a large reservoir of still mostly undescribed biodiversity. It is increasingly targeted by resource-extraction industries and yet is drastically understudied. In such remote and hard-to-access ecosystems, environmental DNA (eDNA) metabarcoding is a useful and efficient tool for studying biodiversity and implementing environmental impact assessments. Yet, eDNA analysis outcomes may be biased toward describing past rather than present communities as sediments contain both contemporary and ancient DNA. Using commercially available kits, we investigated the impacts of five molecular processing methods on eDNA metabarcoding biodiversity inventories targeting prokaryotes (16S), unicellular eukaryotes (18S-V4), and metazoans (18S-V1, COI). As the size distribution of ancient DNA is skewed toward small fragments, we evaluated the effect of removingshort DNA fragments via size selection and ethanol reconcentration using eDNA extracted from 10 g of sediment at five deep-sea sites. We also compare communities revealed by eDNA and environmental RNA (eRNA) co-extracted from ∼2 g of sedimen at the same sites. Results show that removing short DNA fragments does not affect alpha and beta diversity estimates in any of the biological compartments investigated. Results also confirm doubts regarding the possibility to better describe live communities using eRNA. With ribosomal loci, eRNA, while resolving similar spatial patterns than co-extracted eDNA, resulted in significantly higher richness estimates, supporting hypotheses of increased persistence of ribosomal RNA (rRNA) in the environment and unmeasured bias due to overabundance of rRNA and RNA release. With the mitochondrial locus, eRNA detected lower metazoan richness and resolved fewer spatial patterns than co-extracted eDNA, reflecting high messenger RNA lability. Results also highlight the importance of using large amounts of sediment (≥10 g) for accurately surveying eukaryotic diversity. We conclude that eDNA should be favored over eRNA for logistically realistic, repeatable, and reliable surveys and confirm that large sediment samples (≥10 g) deliver more complete and accurate assessments of benthic eukaryotic biodiversity and that increasing the number of biological rather than technical replicates is important to infer robust ecological patterns.