Children's social networks are filled with real and imagined relationship partners, and advances in artificial intelligence (AI) have generated a host of possible relationships with virtual characters. However, little is known about how children conceptualize virtual characters, and how they might be viewed as friendship partners. These studies were designed to develop a paradigm for investigating children's concepts of virtual characters. In Study 1, items were developed that could be used in a task with children. Adults (N = 144) attributed a range of properties (biological, psychological, and social) to four child agents: a real child, a child on a video chat program, an AI child, and a doll. Items that captured adult concepts were used in a task with children (N = 30; aged 5 to 8 years old). This task was designed to: (a) control for idiosyncratic features of the four child agents, (b) enable the expression of uncertainty, and (c) assess concepts of reciprocal relationships. In Study 1, adults did not differentiate between the AI child and the doll on the social property; however, a subset of adults favored the AI child over the doll for sharing secrets. In Study 2, a subset of children indicated a preference for the doll over the AI child on opportunities for reciprocal love and companionship. These findings suggest that virtual characters afford select social opportunities that may differ between adults and children, raising further questions as to the design goals for virtual characters and the functions they serve in our day-to-day lives.