PatternFly AI: Conversation and Chatbot Guidelines (Red Hat)
Red Hat’s open-source enterprise design system. Its guideline pages are generated from the patternfly/patternfly-org GitHub repository.
License: MIT License — confirmed by the repository’s own root LICENSE file, which explicitly covers “the Software and associated documentation files,” not only the component code, so the guideline prose itself is safely licensed too.
Key points
- AI overview / Principles and guidelines state five core principles — accountability, explainability, transparency, fairness, and human-centeredness — governed by one rule: “imperceptible AI is not ethical,” meaning a person must always be able to tell they’re interacting with an AI system. The design guidance built on those principles: confirm an AI feature addresses a genuine need before building it rather than adding one for novelty’s sake; design it to enhance rather than replace human ability, including peer review of AI-assisted output; disclose AI usage, capabilities, and limitations up front; and plan for failure by keeping errors explicit and routing to accessible human support.
- AI guidelines restates the same transparency and error-handling points as concrete practices: tell users when AI is being used, set expectations about its capabilities and limits, explain how it reaches a decision, keep users in control of how they interact with it, and be explicit and honest when it fails or hallucinates.
- Conversation design covers writing a chatbot’s own messages: keep a bot message to about three lines and stop after at most three consecutive bot messages before asking for user input; frame a request around the user’s own goal and end a message on its actionable point rather than leading with it; explain why personal information is needed before asking for it; avoid open-ended questions; mix cards and buttons in with plain text; personalize a greeting when a user’s name is available; and let the bot deliver a conversation’s final message rather than leaving the user’s own message as the last word.
- About chatbot lays out a chat window’s standard structural parts — a container, a toggle to show or hide it, a header, a footer, and message components for both bot and user turns, plus attachment support and a scrollable conversation history — and states a chatbot should only be built where it adds genuine value to the user’s experience, not for its own sake.
- Chatbot messages documents the individual message component: a visual and label distinction between bot and user messages (an “AI” label marks bot messages), a loading state for a response still being generated, timestamps, file attachments with a read-only preview and a separate edit affordance, per-message actions (positive/negative feedback, copy, share, and text-to-speech), and — for retrieval-augmented responses — paginated source cards with a one-line title and a short body description rather than a bare link.