Users of question-answering systems expect those systems to be responsive to their unstated goals and plans. In order to do this the system should have the ability to recognize the user's plans, and to plan a helpful response. A new system architecture embodying this framework is proposed, and its application in two implemented systems is illustrated.