<p>Object referencing is a critical component of effective remote instruction. However, instructors and workers often struggle with this process due to a lack of common ground-a shared understanding of how to formulate mutually understandable expressions. This study proposes semantic augmented reality (AR) labels that attach semantically meaningful content to objects to support the establishment of conceptual pacts and common ground about how to refer to the objects between remote instructors and workers. To evaluate the effectiveness of this approach, we conducted a between-participants experiment with 72 participants (18 groups per condition), who completed three trials of a remote assembly task. In the first two trials, participants performed the instruction task using either the proposed Semantic Label or a conventional method, Simple Label, which attaches alphabetical characters to objects. In the third trial, all participants performed the task without AR labels. The results showed that both AR labeling methods supported the establishment of conceptual pacts and yielded similar performance in terms of task efficiency and mental workload. However, in the third trial, only participants who had used Semantic Labels maintained high communication efficiency, suggesting that Semantic Labels also serve as a catalyst for developing lasting common ground.</p>