We studied the effect of dyadic interaction on implicit learning using a visual artificial grammar learning task. Specifically, we were interested in metacognitive experiences involved in the social verification of implicit knowledge. The experiment consisted of a learning phase followed by dyadic and individual test phases. The experimental group exchanged initial responses and confidence ratings in the dyadic test phase before revising their judgments. In the control group, participants could revise their responses but had no communication. We expected that sharing metacognitive experiences and initial responses would improve the accuracy of judgments and the underlying implicit knowledge. Our findings revealed the social verification effect in the dyadic test phase (but not in the individual test phase) that occurs selectively due to the mediating role of confidence. We conclude that dyadic interaction can mitigate the detrimental effects of response revision when applying implicitly learned regularities.