Inputs AI UX patterns
Input patterns shape how users prompt, attach context, select modes, and compose multimodal requests.
Start here
Core patterns for inputs UX.
22 patterns
Tool Switching in Composer
Switch between AI capabilities within composer
Context Chip Management
Adding context sources via menu with removable chips
Input Mode Toggle
Switch between text, voice, and dictation modes
Follow-up Chips
Suggested next turns
Command Bar
Cmd+K for AI
Context Mentions
Reference files via @
Slash Commands
Quick actions via /
Magic Edit
Transform selection
Prompt Starters
Zero state examples
Voice Input
Speech-to-text with visual feedback
Multimodal Input
Images & Text
Prompt Templates
Starter prompts
AI Context Menu
Actions on select
Predictive Type
Ghost text
Tone Sliders
Adjust style
Persona Selector
Change AI role
Dynamic Follow-ups
Suggested questions
Gesture Input
Draw or gesture to trigger AI actions
File Upload with AI Preview
Upload files with AI-generated previews
Voice-to-Action
Voice commands that trigger specific actions
Smart Autocomplete
Context-aware autocomplete beyond text
Batch Input Processing
Process multiple inputs at once
Frequently asked questions
What problems do input patterns solve?
They make intent legible before send: attachments, mode switches, templates, and constraints that reduce malformed or ambiguous requests.
How do command bars differ from slash commands?
Command bars are global, searchable action palettes (often Cmd+K). Slash commands are in-context, usually in the composer—use both for power users at different scopes.
When is multimodal input worth the complexity?
When your users routinely mix voice, files, and images with text—support, field apps, creative tools—not when a single text box covers 95% of jobs.
What should tool-switching UI communicate?
Which capabilities are active, how they change model behavior, and how to turn them off. Silent mode switches cause “why did it do that?” support tickets.
How do tone sliders and persona selectors help?
They set expectations before generation—output style, audience, and guardrails—so users don’t post-hoc fight the model with long correction prompts.
Which input patterns reduce bad prompts?
Templates, smart autocomplete, predictive type, and file-upload previews steer users toward valid structure; pair with empty states that show strong examples.