What the person is doing
An AI feature accepts open-ended natural-language input.
Foundation · lesson 3 of 23 · AI And Automation UX
Prompt suggestions
Underlying principle: An automated system should expose what it is doing, the evidence behind its output, the limits of its authority, and the action a person can take next.
Offer contextual, truthful prompt starters that reveal useful tasks, load into an editable prompt field, and require the user to intentionally send the final request.
Study the interaction
Follow the task before choosing the component
An automated system should expose what it is doing, the evidence behind its output, the limits of its authority, and the action a person can take next.
What the interface must communicate
Offer contextual, truthful prompt starters that reveal useful tasks, load into an editable prompt field, and require the user to intentionally send the final request.
Initial state with contextual prompt starters and capability or source boundary notes.
What to look for
After selecting a suggestion, can the user inspect and edit the exact text that will be sent?
Selecting a suggestion populates or inserts editable prompt text instead of silently running the request.
Decision first
Choose this pattern when the problem matches
Use when
- The AI surface supports open-ended requests and users need examples of useful, supported tasks.
- The product can generate or curate prompts from reliable context such as page state, selected content, role, workflow, or agent capabilities.
- Teams benefit from saved, shared, or curated prompt starters that can be adapted before use.
Avoid when
- The user must provide fixed required fields that are clearer as a form.
- The action is high-risk enough that a prompt starter would hide necessary review, approval, or structured input.
- The suggestions are generic marketing copy rather than practical task starters.
Problem it prevents
An AI feature accepts open-ended natural-language input.
Test the behavior
Try strong and weak versions of the interaction
Open the working example, try its controls, and compare the behavior against the decision guidance.
In the lab you can
Launch Prompt suggestions in the lab - switch between strong UI, weak UI, strong UX, and weak UX;
- trigger required, empty, loading, error, and recovery states;
- inspect keyboard and accessibility expectations beside the demo.
Pattern anatomy
What a strong implementation has to make clear
An AI feature accepts open-ended natural-language input.
Offer contextual, truthful prompt starters that reveal useful tasks, load into an editable prompt field, and require the user to intentionally send the final request.
Initial state with contextual prompt starters and capability or source boundary notes.
The selected starter sends a prompt before the user can correct missing context.
Expose suggestions as named buttons, list items, or cards with clear keyboard focus and selected state.
Quality bar
The difference between expert and weak execution
Strong implementation
Specific, visible, recoverable
- Each prompt starter shows an action label, full prompt preview, and boundary note such as uses selected thread only.
- The selected suggestion appears in an editable prompt field with Send and Clear actions.
- Unavailable suggestions are disabled with a reason tied to missing selection, permission, or source data.
- Selecting a starter loads a draft, moves focus to the prompt editor, and lets the user adjust scope before sending.
Weak implementation
Vague, hidden, hard to recover from
- A suggestion styled like a final AI answer rather than an editable starting point.
- Prompt suggestions remain visible after the selected document or agent capability changes.
- A vague prompt such as do everything perfectly is auto-filled and sent.
- The selected starter sends a prompt before the user can correct missing context.
UI guidance
- Render contextual prompt starter cards or chips with a task label, editable prompt text, capability boundary, selected state, disabled or stale state, and explicit send action.
- Keep suggestion content visually distinct from user-authored prompt text until the user edits or sends it.
- Place the editable prompt editor close to the suggestions so the transition from starter to owned prompt is visible.
UX guidance
- Help users discover useful AI requests while preserving control over the exact prompt that is sent.
- Use prompt suggestions to reveal supported capabilities and needed context, not to imply guaranteed correctness or hidden data access.
- Let users revise, combine, ignore, save, or share suggestions without making any suggestion mandatory.
Implementation contract
What the implementation must handle
States
- Initial state with contextual prompt starters and capability or source boundary notes.
- Edited draft state after the user changes the suggested prompt.
- Submitted state that shows the exact prompt sent and preserves recovery if the output is weak.
- Unavailable, stale, or limited-capability state when suggestions should be hidden, refreshed, or clearly labeled.
Interaction
- Selecting a suggestion populates or inserts editable prompt text instead of silently running the request.
- The suggestion label explains the task outcome, while the draft prompt contains enough context to be useful after selection.
- Users can ignore, revise, combine, clear, or send suggestions using pointer and keyboard.
- The UI distinguishes system-suggested text from user-authored text until the user edits or sends it.
Accessibility
- Expose suggestions as named buttons, list items, or cards with clear keyboard focus and selected state.
- Keep the prompt editor reachable immediately after choosing a suggestion.
- Tab moves through suggestion controls, the prompt editor, send, and clear actions in a predictable order.
- Arrow-key navigation may be used inside a structured suggestion list if active option and selection state are exposed.
Review
- After selecting a suggestion, can the user inspect and edit the exact text that will be sent?
Evidence trail
Source-backed claims behind this guidance
Understand Prompt Gallery in Copilot
Microsoft Learn - checked
Microsoft documents Prompt Gallery as suggested, saved, shared, and team prompts that help users understand Copilot capabilities.
Configure suggested prompts for Teams and Microsoft 365
Microsoft Learn - checked
Copilot Studio documents configurable suggested prompts that appear before a new agent chat and need publish/update handling.
Prompt assistant
Microsoft Learn - checked
Prompt assistant guidance documents draft prompts that users review, keep, edit, and test before use.
Guidelines for Human-AI Interaction
Microsoft Research - checked
Human-AI guidelines support capability clarity, user control, correction, dismissal, and appropriate reliance.
Full agent/debug reference
Problem Context
- An AI feature accepts open-ended natural-language input.
- The system has useful but bounded capabilities that users may not discover from an empty prompt box.
- Suggested prompts may be created by the product, generated from agent configuration, saved by the user, or shared by a team.
Selection Rules
- Choose prompt suggestions when the user needs help discovering useful AI requests, not when the task can be captured by a fixed form.
- Make each suggestion specific to the current object, workspace, agent capability, or role so it does not read like generic filler.
- Load the selected suggestion into editable input before sending when the request can affect data, people, permissions, or consequential decisions.
- Suppress or rewrite suggestions when the system lacks the required data, grounding, language support, tool access, or permission to satisfy them.
- Prefer search suggestions when the goal is to phrase a retrieval query, and prefer a command palette when the row executes an app command or navigation action.
Required States
- Initial state with contextual prompt starters and capability or source boundary notes.
- Selected suggestion state where the draft prompt is visible, editable, and not yet sent.
- Edited draft state after the user changes the suggested prompt.
- Submitted state that shows the exact prompt sent and preserves recovery if the output is weak.
- Unavailable, stale, or limited-capability state when suggestions should be hidden, refreshed, or clearly labeled.
Interaction Contract
- Selecting a suggestion populates or inserts editable prompt text instead of silently running the request.
- The suggestion label explains the task outcome, while the draft prompt contains enough context to be useful after selection.
- Users can ignore, revise, combine, clear, or send suggestions using pointer and keyboard.
- The UI distinguishes system-suggested text from user-authored text until the user edits or sends it.
- The suggestion set updates when the underlying object, agent capability, language, permission, or source context changes.
Implementation Checklist
- Derive suggestions from the current task, selected object, available tools, and known data sources.
- Show short action labels, full editable prompt text, and a boundary note such as needs selected text or uses open tickets only.
- Provide an explicit send action after selection and keep the draft prompt editable.
- Expose selected, focused, unavailable, stale, loading, and no-suggestion states.
- Avoid suggestions that imply hidden data access, guaranteed correctness, unsupported languages, or unavailable tool actions.
- Measure which suggestions are selected, edited, sent, dismissed, saved, or shared without turning analytics into required user workflow.
Common Generated-UI Mistakes
- Showing generic chips such as summarize, improve, or make better with no object-specific context.
- Auto-submitting a suggested prompt before the user can review or edit it.
- Suggesting work the AI cannot ground, verify, or perform with the current permissions.
- Styling suggestions like final answers or mandatory next steps.
- Leaving old suggestions visible after the user changes file, thread, customer, agent, language, or workspace context.
Critique Questions
- Would a user understand what the AI can do from these suggestions without being misled about data access or reliability?
- After selecting a suggestion, can the user inspect and edit the exact text that will be sent?
- Are these prompts tied to the current task strongly enough that another pattern could not use the same copy unchanged?
Accessibility
- Expose suggestions as named buttons, list items, or cards with clear keyboard focus and selected state.
- Do not rely on chip color alone to distinguish suggested text from user-authored prompt text.
- Announce when selecting a suggestion loads an editable draft but does not send it.
- Keep the prompt editor reachable immediately after choosing a suggestion.
Keyboard Behavior
- Tab moves through suggestion controls, the prompt editor, send, and clear actions in a predictable order.
- Enter or Space on a suggestion loads it into the prompt editor without submitting unless the control is explicitly labeled as send.
- Escape may dismiss an open suggestion gallery or return focus to the prompt editor without deleting the draft.
- Arrow-key navigation may be used inside a structured suggestion list if active option and selection state are exposed.
Variants
- Starter prompts
- Contextual prompt chips
- Prompt gallery
- Saved prompts
- Shared team prompts
- Prompt assistant draft
- Prompt template gallery
Verification
Source trail checked: