In the context of automatic speaker verification it is well known that different speech units offer different levels of speaker discrimination. For short-duration, text-dependent automatic speaker recognition, a user’s pass-phrase bears influence on how reliably they can be recognized; just as is the case with text passwords, some spoken pass-phrases are more secure than others. This paper investigates the influence of text or phone content on recognition performance. This work is performed using the shortest duration subset of the standard RSR2015 database. With a thorough statistical analysis, the work shows how significant reductions in error rates can be achieved by preventing the use of weak passwords and that improvements in performance are consistent across disjoint speaker subsets. The ultimate goal of this work is to develop an automated means of enforcing the use of stronger or more discriminant spoken pass-phrases.