From Prompts to Probes: How Large Language Models Improve Response Quality in Open-Ended Survey Research

Probing (i.e., asking follow-up questions to elicit elaboration) is a common method in qualitative research. While effective in human interviews, its benefits in AI-led surveys operated by chatbots remain underexplored. This paper investigates whether follow-up questions generated by large language models (LLMs) can improve the quality of open-ended survey responses. In a between-subjects experiment (N = 151), we compared different probing strategies and measured response quality by word count and thematic richness. Contextual probing significantly increased both response length and thematic richness. These findings indicate that LLMs can emulate key techniques of qualitative interviewing, enabling richer and more informative responses in online surveys. This positions LLM-driven probing as a scalable way to enhance data quality, bridging the gap between automation and qualitative depth. The study contributes to conversational AI research by showing how real-time adaptation fosters user elaboration, and offers practical guidance for integrating LLMs into surveys requiring nuanced input.