Background <p>Shared decision-making is central to the care of patients with rheumatoid arthritis, particularly when adjusting treatment. Conventional decision aids convey balanced information about therapeutic options but often fail to help patients clarify their preferences. We aimed to develop and evaluate a generative artificial intelligence chatbot to facilitate preference deliberation after exposure to standard decision aids during the shared decision-making process.</p> Methods <p>Following the International Patient Decision Aid Standards and contemporary rheumatoid arthritis treatment guidelines, we developed a patient-facing conversational chatbot. The chatbot was designed to guide patients to reflect on treatment expectations, concerns, and daily-life fit. Twenty adults with rheumatoid arthritis and moderate-to-high disease activity completed two iterative rounds of usability testing. Primary outcomes were the System Usability Scale and the Usability Metric for User Experience. Secondary outcomes included five usability domains (usefulness, attractiveness, accessibility, reliability, convenience) and, in phase 2, the Decisional Conflict Scale.</p> Results <p>After refinements based on phase 1 feedback, mean (± standard deviation) System Usability Scale and Usability Metric for User Experience scores increased from 66.0 (± 11.4) to 70.0 (± 10.6) and from 60.4 (± 21.4) to 73.6 (± 16.3), respectively. Among the usability domains, accessibility improved significantly from 3.1 (± 1.3) to 4.6 (± 0.5) (<i>p</i> = 0.007). Mean (± standard deviation) Decisional Conflict Scale scores decreased from 35.3 (± 18.1) to 27.6 (± 9.0) after chatbot use.</p> Conclusions <p>A generative artificial intelligence chatbot showed favourable usability and preliminary evidence of reduced decisional conflict. By helping patients deliberate and articulate their treatment preferences, the system may bridge to value-concordant decision-making while preserving the physician–patient relationship.</p>