The consensus sequence is generated by creating a multiple alignment of the cluster and taking the majority symbol (letter or gap) from each column. If the majority symbol is a gap, the column in skipped. Terminal gaps are ignored unless the cons_truncate option is specified. If ‑cons_truncate is given, and the majority symbol is a terminal gap, then the column is deleted. In other words, with -cons_truncate, all gaps are counted, even if they are terminal gaps.

The consensus sequence may be a better model of a cluster than the centroid sequence. For example, if you are clustering next-generation reads, then the longest sequence in the cluster tends to have more errors, and taking the consensus tends to correct the errors.