The integration of artificial intelligence technology with tourism has introduced a growing number of virtual humans to tourists. These digital agents serve not only as endorsers for tourist destinations but also as personnel to provide diverse immersive services for tourists. However, limited research has been considered on the varying effects caused by the roles they play. Therefore, this study, based on stereotype theory, constructs a mechanism model where virtual human appearance realism and social roles jointly influence tourists' word‐of‐mouth intention. The model is validated through two scenario experiments. Findings indicate that when virtual humans assume service‐oriented roles, high appearance realism is more effective in stimulating tourists' word‐of‐mouth intention, whereas when virtual humans assume endorsement‐oriented roles, low appearance realism is more effective. Perceived naturalness and perceived comfort serve as mediators and chain mediators in the aforementioned relationships.