Standardized, reusable design solutions that define how humans and artificial intelligence systems interact within a product.
These patterns help teams create more predictable and understandable AI experiences by managing behaviors like uncertainty, loading states, generation, and user feedback consistently across the interface.
Use cases
Ignore these and you get: confused users, broken trust, and AI features that quietly get switched off after launch.
Streaming text outputs — Some AI systems generate responses gradually, word by word. Without clear loading or typing indicators, users may assume the interface has frozen or stopped working.
Uncertainty communication — When AI isn't confident, what does it show? Designers who skip this get hallucination UX: users trust wrong outputs.
How it's used in practice
Designate "AI status" states: Define standard UI indicators for when the AI is thinking (loading), typing (generating), or has failed (fallback). Keep them distinct from human-generated content.
Build the feedback loop: Never deploy a generative feature without an explicit feedback mechanism (like thumbs up/down). This data is critical for product managers to improve the model.
Map interaction to "draft" status: Treat every AI output as a temporary suggestion. Use visual markers (ghost text, subtle background colors) to show the user that the content is unverified until they accept it.
🪄
Pro-tip: One important design consideration is what happens after the AI generates a result.
Users should be able to easily edit, copy, move, or reuse the output without unnecessary friction.
Challenges & limitations
The trust gap: If AI systems present incorrect information too confidently, users may lose trust in the product over time.
Latency handling: AI takes time to think. Designers struggle to keep users engaged during multi-second model responses without making the app feel slow.
