diegosouzapw AI Interaction Patterns
AI-specific UX patterns for designing interfaces where users interact with AI models. Covers the full interaction lifecycle: from first prompt to output verification, memory persistence, and trust building.
Source: Based on Shape of AI pattern library (CC-BY-NC-SA) by Emily Campbell.
Overview
This skill provides:
- 60+ AI interaction patterns across 6 categories
- Pattern selection decision framework
- Trust-level assessment for AI interfaces
- Wayfinding strategies for AI onboarding
- Human-in-the-loop governance patterns
- AI identity and personality design guidance
- Memory and context persistence patterns
- Anti-patterns to avoid in AI UX
How This Differs from Other UI/UX Skills
| Skill | Focus |
|---|---|
| 23-uiux-design-principles | Layout, hierarchy, responsive design (framework-agnostic) |
| 21-enterprise-ai-ux | Enterprise context: challenge taxonomy, professional palettes, RBAC |
| 22-conversation-ux | Thread management, branching data model, context switching |
| 20-interactive-widgets | Widget protocols, rendering pipeline, state management |
| 25-ai-interaction-patterns (this) | AI-SPECIFIC interaction logic: how users prompt, control, trust, and relate to AI |
Reference Documentation
Complete Pattern Catalog
- ai-interaction-patterns - Full reference for all 60+ patterns
- Wayfinders (8 patterns)
- Prompt Actions (14 patterns)
- Tuners (10 patterns)
- Governors (13 patterns)
- Trust Builders (7 patterns)
- Identifiers (5 patterns)
Quick Pattern Selection
By User Problem
| User Says/Feels | Apply Pattern |
|---|---|
| "I don't know what to ask" | Gallery, Suggestions, Templates |
| "AI didn't understand me" | Follow-ups, Nudges, Prompt Enhancer |
| "I want alternatives" | Variations, Branches, Randomize |
| "Is this accurate?" | Citations, References, Caveat |
| "This is taking too long" | Draft Mode, Controls, Cost Estimates |
| "I need AI to do something complex" | Action Plan, Stream of Thought |
| "Is this AI or human?" | Disclosure, Avatar, Name |
| "Don't store my data" | Incognito Mode, Consent, Data Ownership |
| "AI forgot what I said" | Memory (scoped/global/ephemeral) |
By AI Product Type
| Product Type | Essential Patterns | Nice-to-Have |
|---|---|---|
| Chat assistant | Open Input, Suggestions, Follow-ups, Memory, Disclosure | Gallery, Voice & Tone, Branches |
| Code copilot | Inline Action, Stream of Thought, Controls, Citations | Action Plan, Draft Mode |
| Image generator | Gallery, Parameters, Variations, Inpainting, Preset Styles | Draft Mode, Randomize |
| Document AI | Attachments, Citations, Caveat, Disclosure, Summary | Transform, Expand, Follow-ups |
| AI agent (agentic) | Action Plan, Controls, Verification, Stream of Thought, Cost Estimates | Memory, Consent |
| Voice assistant | Voice Avatar, Personality, Controls, Disclosure | Memory, Consent |
| Enterprise analytics | Citations, Connectors, Filters, Modes, Disclosure | Action Plan, Memory |
Trust Level Decision
Is this a high-stakes domain (healthcare, finance, legal)?
YES → CRITICAL trust: Citations + Verification + Disclosure + Caveat + Audit
NO →
Is AI output mixed with human content?
YES → HIGH trust: Disclosure + Citations + Caveat
NO →
Could AI output cause harm if wrong?
YES → MEDIUM trust: Caveat + Citations (optional)
NO → LOW trust: Minimal caveat, focus on UX quality
The Six Pattern Categories
1. Wayfinders
Help users construct their first prompt and get started.
| Pattern | Purpose | When to Use |
|---|---|---|
| Gallery | Showcase what's possible | Onboarding, inspiration, capability discovery |
| Suggestions | Context-aware prompt starters | Cold start, idle moments, mode changes |
| Templates | Structured prompt scaffolds | Recurring tasks, complex prompts |
| Follow-ups | Conversation continuations | After every AI response |
| Initial CTA | First-interaction entry point | Landing pages, empty states |
| Nudges | Proactive guidance | When user prompt could be improved |
| Prompt Details | Expose generation parameters | Educational, reverse-engineering |
| Randomize | Serendipity exploration | Creative tools, discovery |
2. Prompt Actions
Different actions users can direct AI to complete.
| Pattern | Purpose | When to Use |
|---|---|---|
| Open Input | Natural language dialogue | Universal starting point |
| Inline Action | Edit within existing content | Document editors, code tools |
| Chained Action | Multi-step sequential tasks | Complex workflows |
| Regenerate | Re-run with modifications | When output needs improvement |
| Transform | Change content format/structure | Data processing, content adaptation |
| Restyle | Change visual/tonal style | Creative tools, voice adjustment |
| Expand | Elaborate on content | Drafting, content generation |
| Summary | Condense content | Reading, research, analysis |
| Synthesis | Combine multiple sources | Research, report generation |
| Describe | Generate from visual/audio input | Multi-modal tools |
| Auto-fill | AI-populated form fields | Data entry, profile completion |
| Restructure | Reorganize content structure | Documents, presentations |
| Madlibs | Guided prompt construction | Onboarding, structured tasks |
| Inpainting | Region-based selective editing | Image/video editing |
3. Tuners
Adjust contextual data and settings to refine the prompt.
| Pattern | Purpose | When to Use |
|---|---|---|
| Attachments | Upload files as context | Document analysis, RAG |
| Connectors | Link external data sources | Enterprise integration |
| Parameters | Fine-tune generation settings | Advanced users, precision tasks |
| Model Management | Select/switch AI models | Multi-model products |
| Modes | Switch AI behavior profiles | Multi-purpose tools |
| Filters | Narrow input/output scope | Search, data analysis |
| Prompt Enhancer | Auto-improve user prompts | Novice users, quality improvement |
| Preset Styles | Pre-configured parameter sets | Quick style selection |
| Saved Styles | User-created parameter presets | Returning users, consistency |
| Voice and Tone | Configure AI personality | Customization, brand alignment |
4. Governors
Human-in-the-loop features for oversight and agency.
| Pattern | Purpose | When to Use |
|---|---|---|
| Action Plan | Preview steps before execution | Complex/expensive tasks |
| Stream of Thought | Show AI reasoning in real-time | Transparency, debugging |
| Controls | Stop, pause, resume, queue | During generation |
| Draft Mode | Low-fidelity preview first | Expensive generation |
| Branches | Divergent exploration paths | Creative exploration |
| Variations | Multiple outputs for comparison | Selection, quality |
| Citations | Source attribution | Factual claims, research |
| References | Link to supporting materials | Evidence, verification |
| Verification | Confirmation before irreversible actions | Destructive operations |
| Memory | Cross-session context persistence | Personalization |
| Cost Estimates | Resource usage transparency | Paid/metered services |
| Sample Response | Quick preview before full gen | Expensive operations |
| Shared Vision | Align user-AI understanding | Complex instructions |
5. Trust Builders
Build confidence in AI ethics, accuracy, and trustworthiness.
| Pattern | Purpose | When to Use |
|---|---|---|
| Disclosure | Label AI content as AI-generated | Blended content, agents |
| Caveat | Warn about AI limitations | All AI outputs |
| Consent | Obtain permission for data use | Recording, training, sharing |
| Data Ownership | Clarify data storage/usage | Enterprise, privacy-sensitive |
| Watermark | Mark AI-generated media | Images, audio, video |
| Footprints | Attribute aggregated sources | Synthesized content |
| Incognito Mode | Non-persistent sessions | Sensitive topics, privacy |
6. Identifiers
Distinct qualities of AI that can be modified at brand/model level.
| Pattern | Purpose | When to Use |
|---|---|---|
| Avatar | Visual AI representation | Chat, voice, multi-agent |
| Personality | Tone, warmth, authority | All AI interactions |
| Name | AI naming strategy | Branding, trust-setting |
| Color | Brand and state signaling | UI theming, state changes |
| Iconography | AI-specific visual language | Throughout UI |
CRITICAL Gotchas
| Rule | Why |
|---|---|
| NEVER use photorealistic avatars unless the AI truly matches that capability | Sets unrealistic expectations, erodes trust when capabilities fall short |
| ALWAYS show Stream of Thought for tasks > 5 seconds | Users abandon or re-submit when they can't see progress |
| NEVER let Memory be a black box | Users must see, edit, and delete what AI remembers |
| ALWAYS offer Controls (at minimum: stop) during generation | Users need escape hatches for wrong-direction outputs |
| NEVER rely solely on Caveats for safety | Caveat blindness is real; design the system to be safe, then add caveats |
| ALWAYS distinguish AI content from human content in blended UIs | Users may unknowingly present AI work as their own |
| NEVER overwrite user work without Verification | Accidental overwrites destroy trust instantly |
| ALWAYS pair Suggestions with the ability to edit before sending | Users need to refine, not just accept blindly |
When to Use This Skill
Use this skill when:
- Designing any AI-powered user interface
- Choosing interaction patterns for AI products
- Evaluating AI UX for trust and transparency
- Building prompt input experiences
- Implementing human-in-the-loop workflows
- Designing AI identity (avatar, name, personality)
- Planning AI memory and context management UX
- Auditing AI interfaces for anti-patterns
Related Skills
- 21-enterprise-ai-ux - Enterprise-specific AI design (RBAC, compliance, professional palettes)
- 22-conversation-ux - Thread management and conversation data models
- 23-uiux-design-principles - General design principles (layout, hierarchy)
- 20-interactive-widgets - Widget rendering and state management
Support
For AI interaction design questions, invoke:
ai-ux-designer- AI-specific interaction pattern selection and designuiux-designer- General layout, hierarchy, and visual designkaizen-specialist- AI agent capabilities informing UX decisionsfrontend-developer- Implementation of AI interaction patterns