The Emerging Use of GenAI for UX Research in Software Development: Challenges and Opportunities

TL;DR

GenAI promises to accelerate qualitative UX research in software development, but its adoption hinges on preserving interpretive rigor and human agency. The authors combine qualitative interviews with a design probe (Study 1) and a bottom-up AI analysis feasibility test (Study 2) to uncover stakeholder needs, trust considerations, and interaction patterns. Key contributions include evidence of role-specific AI requirements, validation of a human-in-the-loop bottom-up topic extraction pipeline, and a comprehensive set of design guidelines for trustworthy AI-assisted qualitative analysis. The work offers actionable, role-sensitive design opportunities to integrate GenAI into agile product development without compromising depth, traceability, or collaboration.

Abstract

The growing adoption of generative AI (GenAI) is reshaping how user experience (UX) research teams conduct qualitative research in software development, creating opportunities to streamline the production of qualitative insights. This paper presents findings from two user studies examining how current practices are challenged by GenAI and offering design implications for future AI assistance. Semi-structured interviews with 21 UX researchers, product managers, and designers reveal challenges of aligning AI capabilities with the interpretive, collaborative nature of qualitative research and tensions between roles. UX researchers expressed limited trust in AI-generated results, while product managers often overestimated AI capabilities, amplifying organizational pressures to accelerate research within agile workflows. In a second study, we validated an AI analysis approach more closely aligned with human analysis processes to address trust issues bottoms-up. We outline interaction patterns and design guidelines for responsibly integrating AI into software development cycles.

Figures (5)

• Figure 1: Combined survey results: (a) overall, (b) Q1–Q3 detail.
• Figure 2: Mock-up of AI-extracted topics with the initiation flow shown in the chat. In the chat, the user provides research objectives of their study, and the AI requests transcripts to be uploaded. After the user adds the transcripts, the AI displays suggested topics and asks them to review and validate.
• Figure 3: Mock-up of a chat flow in which a user requests the AI for insights. In the chat panel, the user asks, "Which participants said something about productivity?" The AI responds with three excerpts that mention productivity, along with the option to see the full transcript for each one.
• Figure 4: Mock-up showing outliers in AI-extracted topics. Outliers are displayed as small, disconnected circles, representing the smaller amount of data backing them up and their disconnect with topic clusters.
• Figure 5: Mock-up showing the transcript view for one participant. In the transcripts panel on the left, P1 transcript has been selected. The panel displays a participant summary, transcript summary, and the full transcript below.