Conversational agents (CAs) are increasingly being developed on commercial websites nowadays. We tested in two studies whether gender stereotypes apply to non-gendered CAs. In the first study, participants evaluated whether CAs are expected to display more masculine or feminine characteristics in situations designed to be stereotypically male or female. The sexist attitudes of the respondents were also measured. As predicted, participants perceived that a CA should be more masculine in stereotypically male situations and more feminine in stereotypically female situations. Moreover, we found that hostile sexism but not benevolent sexism moderated the effect of the gendered situation. The second study replicated the results while addressing the limits of Study 1, showing the robustness of these effects. These findings are consistent with models of gender stereotypes in humans and robots and show for the first time a moderation effect of (hostile) sexism in a customer service context with CAs. The processes involved in human relationships seem relevant in a digital environment that involves CAs. Researchers and professionals should work together to avoid reproducing and perpetuating gender stereotypes when developing CAs.