Personalized services have traditionally been difficult to automate and therefore costly. Modern technology, such as robo-advising, promises to change this but also introduces new challenges. Robo-advisors typically make investment recommendations based on surveys of clients’ preferences. As a result, biases induced by these surveys can become embedded in the algorithmic recommendations and influence investors’ financial decisions. To examine this conjecture, we conduct two studies. First, we administer a nationwide survey of experienced U.S. investors and show that the response-order effect—the influence of the order of listed choices on respondents’ answers—applies to survey questions commonly used by robo-advisors. Second, in collaboration with a leading robo-advisor in China, we conduct a preregistered RCT and show that, despite the large financial stakes involved, the response-order effect significantly alters both the roboadvisor’s risk assessments and its clients’ actual investment decisions, raising new ethical concerns in the FinTech era.