I have a general question regarding the type of data to be used for generation of the Venn diagram: would it make sense to use the genes identified via Gene Ontology? Or should I always go back to the DESeq2 comparisons? Because I have several comparisons being done AvsB and BvsA, or AvsC and CvsA, and BvsC and CvsB.
Could the gene list from these comparisons be used to check how similar samples A, B, and C are to each other?
It depends on the exact question that you're trying to address, but I'd argue that both approaches are useful. I would generate a 3-way Venn diagram for all genes that pass the differential expression cutoff (FDR < 0.05, or whatever cutoffs you apply), that will tell you how similar the samples A, B, and C are.
A different question would to overlap the significant GO terms. In some cases a treatment will have little overlap on the genes that are changed, but still affect the same pathways. Using both Venn diagrams (from the gene level and from the GO term level) will tell you if this is the case.
I'm a proponent of combining as much data as reasonable into a single graph. Rather than doing multiple Venn diagrams I would produce a single three-way Venn. If you end up with too many samples in this experiment or a different one, then it's worthwhile looking into "Upset plots" when there are too many categories for a single Venn diagram.