Structured Hybrid 3D Virtual Environments are 3D virtual spaces where staff (organisational) software agents support human users in their task achievement. These systems are characterized by: i) being hybrid, so that humans and software agents can interact; and ii) being structured and task oriented, so that interactions are regulated by a subjacent Organisation Centered Multi Agent System (OCMAS)-an Electronic Institution (EI). The contribution of this paper is to include task-oriented conversational staff bots (i.e. the embodiment of staff agents in the 3D environment) that communicate with users by using natural language. With this aim, we extend the Artificial Intelligence Mark-up Language (AIML) with special tags to enable complex task-oriented conversations whose flow needs to consider both the states of the conversation and the ontology related to the task. We evaluate the usability of our conversational proposal and compare it to a previous command-based interaction system. Results show the conversational approach presents a higher user satisfaction than the command-based one. Moreover, in average, it also performs better in terms of efficiency, effectiveness and errors.