We compared three techniques to specify 3D positions for collaborative augmented reality (AR) visualization. AR head-mounted displays allow multiple users to share the same physical space, while keeping seamless social interactions. Interactions being key parts of exploratory visualization tasks, we adapted from the virtual reality literature three distinct techniques to specify points in 3D space, such as for placing annotations for which they cannot rely on existing data objects. We evaluated these techniques on their accuracy and speed, the users subjective workload and preferences, as well as their co-presence, mutual understanding, and behavior in collaborative tasks. Our results suggest that all the three techniques provide good mutual understanding and co-presence among collaborators. They differ, however, in the way users behave, their accuracy, and their speed.