Great AI chatbot UX is about managing expectations and providing value within limitations. Core principles: 1) Onboarding: show users what the chatbot can do with 3-5 starter prompts. Do not present an empty input box - users do not know what to ask. Provide domain-specific examples. 2) Expectation setting: clearly communicate that it is AI, set boundaries upfront ("I can help with X but not Y"), indicate when AI is uncertain. 3) Streaming responses: always show typing indicator and stream tokens. Users will abandon if they wait >3 seconds for response to start. 4) Error recovery: when AI fails or gives wrong answer, make it easy to report and retry. "Was this helpful?" with easy feedback mechanism. 5) Conversation history: users expect to reference previous messages ("what was that recipe you mentioned earlier?"). Implement proper session management. 6) Graceful degradation: when AI cannot answer, provide clear escalation path to human support or documentation. Never leave users stranded. 7) Accessibility: screen reader compatible, keyboard navigable, WCAG 2.1 AA compliance. 8) Mobile-first design: 60%+ of chatbot usage is mobile. Touch-friendly interface, minimize typing with suggested responses. Trust signals: cite sources when providing factual information, show confidence levels for recommendations, be honest about limitations.