They don’t mention any kind of control—I guess an appropriate one would be having a human interact with the participants one-on-one to see if they were as effective. (Although even if they were, the chatbots would likely be easier to implement in practice.)