Google GeminiOutput & Refinement
Gemini renders long-form output in-thread with inline source chips, comparison tables, section export, follow-up chips, refinement menus, Workspace handoffs, Maps split-pane, and a sources sidebar.
Long-form reply with numbered sections, bullets, and inline source chip with popover
What's happening
Gemini treats many answers as structured output: numbered sections, tables, inline source chips, and contextual follow-ups. Regenerate sits on the action row; length and personalization live in a reshape menu; ⋯ overflow routes work into Google Workspace.
Patterns
Top-level regenerate plus Longer, Shorter, and Don't personalize without leaving the thread
Task-scoped follow-up cards as large clickable rows
Headings, bullets, and multi-column tables for scannable output
Place card plus live map pins for location-heavy planning output
UX Insights
- •Inline source chips keep citations adjacent to claims without a separate tab
- •Export to Sheets on sections turns chat output into spreadsheet workflows
- •Don't personalize in refinement menu surfaces memory as an explicit knob
- •Workspace exports (Docs, Gmail) position Gemini as a drafting front door
- •Maps split pane turns local planning into an in-product personal assistant
Design Decisions
Google keeps answers in-thread but layers native export paths. Tables and section actions signal output is meant to be reused, not treated as ephemeral chat.
More case studies
PerplexityOutput & Refinement
Perplexity renders search answers with Answer/Links/Images tabs, research steps, follow-ups, export, rewrite, and selection refinement.
ChatGPTOutput & Refinement
ChatGPT renders structured answers in the thread with reasoning activity, version navigation, feedback, sharing, and selection-based refinement into writing blocks.
ClaudeOutput & Refinement
Claude keeps short answers in the message stream with per-message actions, non-destructive Try again, and version navigation.
ChatGPTResponse Refinement
ChatGPT provides a contextual menu on AI responses that allows users to refine or modify the output with specific instructions like "Try again", "Add details", "More concise", "Search the web", or "Think longer" without retyping their prompt.
More real-world AI UX in your inbox
Weekly gallery picks, interface patterns, and notes on how products ship AI - no spam, unsubscribe anytime.