An Experimental Comparison of AI-Enabled Semi-Structured Interviews and Fixed Surveys: Respondent Satisfaction, Response Patterns, Quality, and Representation

Artificial intelligence (AI) interviewing offers the potential to enhance the depth and quality of survey research, yet it may also introduce respondent fatigue and unintended consequences for the measurement of political attitudes. From the standpoint of respondents, more extensive interviews impose greater cognitive and temporal burdens, which may reduce engagement or increase satisficing. From the perspective of political behavior researchers, deeper probing may shift the subset of beliefs or arguments that respondents access when forming opinions, thereby altering the distribution and apparent stability of measured attitudes. To investigate these dynamics, we conduct a large-scale, fully randomized experiment. We find that AI-mediated interviewing significantly increases both the volume and specificity of respondents’ articulated rationales, but any decline in participant satisfaction is minuscule and we detect no increase in participant rolloff. However, AI interviews appear to shape subsequent responses: After offering justifications through AI interaction, respondents exhibit greater polarization in follow-up items. This result suggests that AI interviewing may alter the cognitive architecture of opinion formation. These findings both highlight the potential for AI interviewing and underscore the need for studies of how AI interviewing may become a source of attitudinal change.