For a given OTU (call it X), a "dominant OTU" (call it D) is an OTU with much higher abundance and high sequence similarity. If X has a dominant OTU, then X may be a spurious OTU induced by bad sequences where the correct sequences belong in D.
This heuristic is similar to the method used by denoising algorithms to identify probable bad sequences; see UNOISE algorithm.
See Tolstoy's paradox for
discussion of why bad sequences and spurious OTUs can be common even with very
low error rates.