Date created: | Last Updated:

Category: Project

Description: The availability of artificial intelligence (AI) may enable researchers to study preference formation qualitatively through unstructured question formats that used to be too costly for inclusion in mass surveys. We experimentally assess whether AI chatbots and their empathic capabilities affect answers to sensitive and non-sensitive questions in qualitative conversations. Our design randomly assigns respondents to a series of pre-set, open-ended questions, a weakly empathic chatbot interviewer, or a highly empathetic chatbot. We evaluate chatbot performance with respect to the informational value of survey responses, perceived empathy and rapport, and price per complete. Our applications include political science (vote choice) and consumer research (drivers of purchasing behavior) questions. Since qualitative interviews promise to increase respondents’ willingness to provide truthful responses to sensitive questions that are particularly prone to social desirability bias, we also investigate attitudes toward abortion and hiring discrimination. The results shed light on whether AI chatbots are a more useful tool to qualitatively explore the determinants of preferences and overcoming social desirability problems compared to using fixed, open-ended questions determined by the researcher.