Augmentative communication (AC) systems with synthesized speech output have been developed for nonspeaking people. Most AC devices that aim to support social conversation have been designed to help the user generate novel utterances as quickly as possible. However, they remain too slow to support effective, real-time conversation. Preconstructed phrases have been shown capable of supporting socially effective conversations with a succession of new partners, when the phrases are stored within a structure that models pragmatic aspects of natural conversation. The extent to which prestored text can be effective in repeated conversations with the same person, where most phrases are not reusable, was investigated in a single-case experimental design. Results indicated that the AC user's conversational rate with a repeat partner was faster, without any accompanying loss of social effectiveness. Furthermore, the user did not need to resort to on-line entry of text more frequently with the repeat partner than with new partners. Implications for the design of AC systems and for the understanding of the pragmatics of conversation are discussed.