Three accounts of common ground maintenance make different assumptions about speakers’ responsibilities regarding listener-privileged information. Duplicated responsibility requires each interlocutor to assimilate the other’s knowledge before designing appropriate utterances. Shared responsibility appeals to least collaborative effort [Clark, H. H., & Wilkes-Gibbs, D. (1986). Referring as a collaborative process. Cognition, 22, 1–39.], requiring each interlocutor to report her own privileged knowledge. Cognitive load [Horton, W. S., & Gerrig, R. J. (2005b). The impact of memory demands on audience design during language production. Cognition, 96(2), 127–142.] assumes duplicated responsibility curtailed by processing limitations, so that simpler cues to listener knowledge should be preferred. Three experiments track genuine gaze of instructors at simulated projected gaze of confederate followers whom they guide along a map route. Though instructors can correct off-route gaze, a simply interpretable cue to listener knowledge, they habitually rely on inaccurate, underspecified verbal feedback instead and begin corrections without first checking the follower’s gaze. Time pressure discourages corrections accompanied by gaze checking. The results argue for shared responsibility attributable to a limitation in capacity to seek and integrate listener knowledge.