LLM-assisted hybrid interfaces for remote repertory grid interviews: A usability and expert evaluation
Yunxing Liu, Jean-Bernard Martens · Computers in Human Behavior Reports · 2026
The Repertory Grid Technique (RGT) is a well-recognized method for eliciting personal (bipolar) constructs. At its core is an iterative laddering process which makes RGT interviews labor-intensive and time-consuming. We present the design of LLM-assisted Q-Survey , a web-based tool that automates the entire RGT process while keeping participants in control, logging the entire construct elicitation process, which can be subsequently analyzed for additional insights. To evaluate the Q-Survey we conducted a remote user test with 17 crowdworkers, and expert interviews with 18 experts in user-centered design and AI. Overall, test participants reported a positive user experience, and it took on average about ( ∼ 27 min) to elicit ≥ 6 personal constructs, which supports the feasibility of remotely administering the RGT with Q-Survey. Experts identified advantages of Q-Survey, namely lower moderator workload and increased objectivity and richer wording in the collected constructs. They also highlighted the need for transparency, privacy, and robust error handling when AI suggestions are involved. Taken together, these studies evaluate whether LLM-assisted Q-Survey can support remote RGT interviews in a usable and methodologically promising way from both participant and expert perspectives. Beyond this primary evaluation objective, we contribute (i) a practical demonstration of how to automate triad-based RGT interviews with LLM support, and (ii) empirically grounded guidance and in-UI guardrails (e.g., visible AI provenance, edit/undo, bypass controls) for how to integrate LLMs into iterative qualitative methods.