Ethics-in-practice in UX Design: An Empirical Study of an AI-Based Conversational Support Tool

While designers routinely face ethical dilemmas related to privacy, accessibility, transparency, and manipulative design, existing UX ethics resources, such as checklists and static guidelines, often fail to provide timely, context-sensitive support within fast, iterative workflows, limiting their effectiveness as instruments of ethical accountability. This paper examines how conversational and static delivery of the same ethics content differ in supporting ethics-in-practice tasks. We developed a retrieval-augmented agent grounded in a curated UX ethics knowledge base and compared it to an equivalently sourced static guideline in a counterbalanced within-subjects study involving five common UX ethics scenarios. Participants evaluated both tools in terms of practicality, clarity, actionability, and decision confidence; results show significant gains for the conversational agent in practicality and actionability, with smaller improvements in clarity and confidence. Qualitative feedback indicates that the agent supports the translation of abstract principles into situated design decisions, while also revealing risks related to response structure, source transparency, and limited multi-stakeholder framing. Automated evaluation of the retrieval-augmented generation pipeline shows high faithfulness and recall but lower answer relevancy, suggesting alignment limitations at the generation stage rather than gaps in the evidence base. We argue that conversational ethics-support tools can improve access to and operationalization of ethics guidance, but should be treated as complements to human deliberation and static resources rather than as shortcuts to ethical correctness.