ESRA 2027 sessions by theme

Back to Overview of Sessions

Session Details

Since the beginning of survey research, live-interview surveys have represented the gold standard of data collection. The advent of artificial intelligence (AI) tools, chatbots, and personas could substantially disrupt this consensus across the survey lifecycle. From questionnaire design and pretesting, to fielding survey interviews, to conducting in-depth qualitative studies, AI methods present opportunities for rapid and efficient data collection that (at least potentially) costs less than a team of live interviewers. Despite these potential advantages, the impact of these tools on interview (in-depth, cognitive, or survey) quality and data quality has barely been examined. At a higher level, the integration of a non-human “persona” into a formerly human-to-human interaction raises ontological questions about what a survey interview is and epistemological questions about how data are gathered and generated by the “black box” of an AI model.

This session invites contributions assessing the impact of AI on interviews across the survey research lifecycle, including its impact on interviewers and respondents. In particular, panelists may consider:
- The impact of AI on respondent and interviewer engagement: how does respondent engagement differ when interviews are conducted by AI and human interviews, and does chatbot quality degrade or improve over the interview duration?
- The quality of AI interviewers as compared to human interviewers, particularly related to leading questions or AI/human misreading of questions
- The impact of fraudulent AI respondents (bots, respondent use of AI assistance) on survey quality
- Human “jailbreaking” of AI interviewers to conduct tasks unrelated to the survey interview and ways to prevent this
- Data quality of AI interviews and approaches to measuring it, e.g., the balance of AI interviewer and respondents within interviews
- Privacy and safety of AI versus human-led interviews
- Comparing Large Language Model (LLM) performance in conducting or assisting survey interviews