Overview
Suggestions
Starter prompt chips and cards to solve the blank canvas problem
onboardingpromptsstarter
Props
| Prop | Type | Default | Description |
|---|---|---|---|
suggestions | StarterPrompt[] | required | Array of starter prompts with id, title, description, prompt, category, and icon |
onSelect | (prompt: StarterPrompt) => void | required | Callback fired when a suggestion is selected |
columns | 2 | 3 | 4 | 2 | Number of columns in the suggestion grid |
variant | "chip" | "card" | "chip" | Visual style — chip for compact, card for detailed |
Docs
Suggestions are the icebreakers of AI interaction. They surface a small set of ready-made prompts — usually three to five — that help users discover what the system can do and overcome the blank-canvas problem. When a user selects a suggestion, it either prefills the input for editing or triggers the action immediately. Beyond onboarding, suggestions keep the conversation moving by proposing relevant follow-ups based on the current context, turning a potentially aimless exchange into a guided experience.
Variants
- Static suggestions — Fixed starter prompts displayed during onboarding or first-run experiences. They may change by mode but aren't personalized to the individual user.
- Contextual suggestions — Prompts that shift dynamically based on what the user is viewing, the active document, or the current mode — for example, offering code-related questions inside an IDE.
- Adaptive suggestions — Prompts that evolve over time as the system learns user preferences, team conventions, or past interaction patterns.
Use Cases
- First-run onboarding to demonstrate system capabilities
- Idle states where the user hasn't interacted for a while
- Context switches such as opening a new document or changing modes
- Follow-up prompts after the AI delivers a response
- Discovery features that surface lesser-known capabilities
Best Practices
- Show three to six suggestions ranked by relevance or engagement. Too many creates scanning fatigue; too few limits discovery.
- Make suggestions actionable — selecting one should either execute the prompt or prefill it for editing, not just display text.
- Ground suggestions in the current context (active file, page content, recent conversation) so they feel relevant, not generic.
- Place suggestions where users need direction: onboarding screens, idle states, and after mode changes. Avoid showing them during focus-intensive tasks.
- For suggestions that trigger data access, publishing, or heavy computation, show a preview or confirmation step before committing.