The effects of human-like social cues on social responses towards text-based conversational agents—a meta-analysis

Humanizing chatbots through social cues is a common strategy to increase user acceptance. However, whether and in which circumstances this strategy is generally effective is still unclear. This meta-analysis thus examines the effect of text-based chatbots’ social cues on users’ social responses and the influence of potential moderators. It includes experimental studies that manipulate human-likeness using social cues and examine their effects on user responses, including attitude, perception, affect, rapport, trust, and behavior. A systematic search for published and unpublished research resulted in a final sample of 800 effect sizes from 199 datasets reported in 142 papers (N = 41,642). Meta-analytic random-effects models computed overall and for each outcome category yielded a small effect of human-likeness on social responses (g = 0.36, 95% CI [0.27, 0.44]). The results further suggested that human-like chatbot characteristics improve user responses to varying degrees and under different boundary conditions. The findings can guide practitioners in designing effective and ethically justifiable chatbots.