Advances in natural language processing techniques enhance the capabilities of conversational agents to process input text and generate fluent responses in various genre. In the educational setting, these conversational agents can function as learning companions to promote peer learning, enhance social interaction, and motivate collaborative behavior. In this paper, we investigate how children collaborate with conversational agents in the context of story construction. We conducted user studies and administered survey forms to collect children’s evaluation of their experience, captured the child-agent conversation logs, and used an observation checklist to monitor the interaction. We analyzed the logs to determine how children respond to varying roles portrayed by the agent and the dialogue moves that motivate collaborative behavior during storytelling. We correlated these with the children’s reported perception and our own observation of the interaction based on performance, collaboration and humanity attributes. Drawing on these results, we identified the roles of conversational agents that align with children’s traits and abilities as storytellers and propose dialogue strategies to facilitate collaborative storytelling. Findings from our study can inform future educational technology designers on the roles of conversational agents as collaborative learning companions who can adapt to the individual needs and preferences of the children.