Abstract Gaze cues are important in communication. In social interactions gaze cues usually occur with spoken language, yet most previous research has used artificial paradigms without dialogue. The present study investigates the interaction between gaze and language using a real-world paradigm. Each participant followed instructions to build a series of abstract structures out of building blocks, while their eye movements were recorded. The instructor varied the specificity of the instructions (unambiguous or ambiguous) and the presence of gaze cues (present or absent) between participants. Fixations to the blocks were recorded and task performance was measured. The presence of gaze cues led to more accurate performance, more accurate visual selection of the target block and more fixations towards the instructor when ambiguous instructions were given, but not when unambiguous instructions were given. We conclude that people only utilize the gaze cues of others when the cues provide useful information.