Overview
How might we help people continue a conversation without staring at a blank composer?
When the answer opens obvious next steps
- Research and chat products where answers naturally branch into follow-up questions.
- After substantive replies where users might not know what to ask next.
- Gemini related questions, Perplexity related research, ChatGPT suggested follow-ups.
- Surfaces where 2-4 high-relevance options beat a generic "Tell me more" nudge.
When chips compete with primary actions or feel generic
- Dense professional tools where chips fight with primary actions (copy, cite, regenerate).
- Answers that already embed a clear checklist of next steps as the main content.
- Multi-turn agent runs mid-flight, where chips interrupt an active plan.
States
Design the suggestion lifecycle, not only static pills parked under every answer.
Answered
The agent reply is complete. Suggestions should reference what was just said, not the whole product.
Suggested
Two to four chips appear under the answer. Each names a concrete next task tied to the content.
Selected
The person taps a chip. It becomes the next user turn, optionally editable before send.
Loading
The follow-up generates from the chip prompt. Prior turns stay visible for context.
Continued
A new answer appears. Fresh chips may replace the old set based on the latest reply.
Dismissed
The person ignores chips and types freely. Suggestions should not block the composer.
Key UX elements
The parts that must be present for suggested next turns to feel useful.
Offer 2-4 contextual options.
Each chip should read like a real next question, not a filler prompt. More than four competes with copy, cite, and regenerate, so keep the row light.
Tie each chip to the answer.
Name a concrete next ask from the answer (compare options, dig into one claim, export). Avoid generic "Explain more."
Send the chip as a user turn.
Tap should append the prompt to the thread so the conversation history stays coherent.
Replace chips after new answers.
Regenerate or a new reply should produce fresh suggestions, not repeat stale options.
Anti-patterns to avoid
Generic chips (“Tell me more”) that ignore the answer’s actual content.
Too many chips (6+) that recreate decision paralysis.
Chips that submit without allowing edit, forcing unwanted side quests.
Repeating the same chip after every regenerate with no novelty.
How products use it
| Product | Implementation |
|---|---|
| Gemini | Related question chips under answers to continue research. |
| ChatGPT | Suggested follow-ups after many replies to drive the next turn. |
| Perplexity | Related questions that deepen or pivot the research thread. |
| Claude | Contextual next-step prompts after substantive answers. |
Real-world examples
How shipped products implement follow-up chips, from our teardown guides.
GeminiGemini output UX: guides, tables & Workspace handoffs
Output, artifacts & refinement
Google AI ModeGoogle AI Mode composer UX: Canvas, Create & Lens
Design the composer
Notion AINotion AI onboarding UX: prompt starters & first value
Getting to first value
PerplexityPerplexity output UX: tabs, follow-ups & export
Output, artifacts & refinement
Implementation
Copy this prompt to generate a production-ready implementation in Cursor, Claude Code, Lovable, or any AI coding agent.
Generate a production-ready implementation of the "Follow-up Chips" AI interface design pattern.
Pattern Definition:Frequently asked questions
What are follow-up chips in AI chat UX?
Follow-up chips are suggested next prompts shown as tappable chips under an answer so users can continue the conversation without inventing the next question from a blank composer.
How are follow-up chips different from the follow-up pattern?
Follow-up chips are a specific UI affordance: visible suggestion pills. Broader follow-up patterns also include written “you could also ask…” copy, related questions sections, or auto-proposed tasks without chip UI.
How many follow-up chips should I show?
Usually 2-4 high-relevance options. More than that competes with primary actions and feels like another search page instead of a light nudge.