DNA metabarcoding can accelerate research on insect diversity, as it is cheap and fast compared to manual sorting and identification. Most metabarcoding protocols require homogenisation of the sample, preventing further work on the specimens. Mild digestion of the tissue by incubation in a lysis buffer has been proposed as an alternative, and, although some mild lysis protocols have already been presented, they have so far not been evaluated against each other. Here, we analyse how two mild lysis buffers (one more aggressive, one gentler in terms of tissue degradation), two different incubation times, and two DNA purification methods (a manual precipitation and an automated protocol) affect the accuracy of retrieving the true composition of mock communities using two mitochondrial markers (COI and 16S). We found that protocol-specific variation in concentration and purity of the DNA extracts produced had little effect on the recovery of species. However, the two lysis treatments differed in quantification of species abundances. Digestion in the gentler buffer and for a shorter time yielded better representation of original sample composition. Digestion in a more aggressive buffer or longer incubation time yielded lower alpha diversity values and increased differences between metabarcoding results and the true species-abundance distribution. We conclude that the details of non-destructive protocols can have a significant effect on metabarcoding performance. A short and mild lysis treatment appears the best choice for recovering the true composition of the sample. This not only improves accuracy, but also comes with a faster processing time than the other treatments.