I suspect this is a product of differentialy expressed glycosyltransferases. Instead of adding a sialic acid, for example, if you're coding for Enzyme B which is an alpha 1-3-galactosyltransferase, you're going to add a Gal to the glycan chain forming B antigen, and so forth. Sialyltransferases can add sialic acid groups, but to the same structure it looks like there is competition between the glycosyltransferases to add sugar moieties that results in different structures getting produced from backbone structures. Source, pages 5-10
Type-1 units may be modified by glycosyltransferases that transfer sugars to either terminal galactose or subterminal GlcNAc, generating sialylated structures or blood group determinants. The synthesis of type-1 units and expression of the corresponding β1–3 galactosyltransferases are regulated in a tissue-specific manner...
Structures Common to Different Glycans
So that's to say there shouldn't be a sialic acid group on ABO antigens, though sialyltransferases are capable of forming Lewis Antigens (Source). In the anabolic pathway the backbone structure is simply getting attacked by glycosyltransferases and the pathway goes the direction whichever one wins out (level of protein vs level of substrate). Being said, I'm unsure it's proper to say ABO antigens are produced from gangliosides, but if anyone finds some faults in my explanation please let me know!
I'm looking into catabolic pathways that may produce ABO antigens by recycling gangliosides, probably resulting from some cleavages by other glycosyltransferases, I just haven't gotten there yet, I'll try to update if I find anything.