Notion AI onboarding UX: prompt starters & first value
Updated December 29, 2025
Notion AI meets users inside the document they are already editing. There is no empty chat screen. Activation surfaces starter prompts tied to writing, editing, and summarizing tasks. The bet is to show concrete jobs like “Write meeting agenda” instead of asking users to imagine what AI can do.
AI at the point of work
What works
- Notion does not route users to a standalone AI tab. Starters appear in the editor chrome where the cursor already is.
- Context is implicit: the current page, block, or selection informs what “summarize this” or “improve writing” will target.
- Embedded AI feels like a command on the document, not a detour to another product.
What we would push on
- First-time users may not know AI exists until they invoke it. No empty-state AI home means discoverability depends on editor affordances.
- Implicit context is powerful but opaque. Users may not realize selection scope changes what AI targets.
Business strategy
Notion’s bet is AI inside the doc tool users already pay for, not a separate chat app. Activation at the cursor lowers switching cost and ties AI usage to page creation, which drives workspace stickiness.
Tradeoff
| Decision | Benefit | Cost |
|---|---|---|
| AI embedded in editor, not standalone chat | Context is automatic; feels like a doc command | Discoverability depends on editor affordances, not an AI home |
Takeaway
Embedded AI should feel like a command on the document, not a detour to another product.
Pattern: Example Prompts Library
Categorized prompt starters
What works
- Examples are grouped by task type so users browse by intent, not by prompt engineering skill.
- Each starter is a complete sentence users can run with one click, then edit. That teaches effective prompts by example.
- Categories scale as capabilities grow without turning the panel into an unfiltered list.
What we would push on
- Category labels need to match how users think about their job. Misaligned buckets push users back to blank prompts.
- One-click starters may skip the “what can AI do here?” mental model if categories are too broad.
Business strategy
Categorized starters replace prompt engineering with job browsing. Notion wants first AI success in one click so users associate AI with concrete doc tasks, not open-ended chat.
Tradeoff
| Decision | Benefit | Cost |
|---|---|---|
| Categorized one-click prompt starters | First value without inventing a prompt | Category quality determines success; bad buckets feel like noise |
Takeaway
Group starters by user jobs (write, edit, summarize), not by model capability or feature name.
Pattern: Example Prompts Library
From starter to conversation
What works
- After “Write meeting agenda,” Notion opens a tighter AI surface with the starter as seed text and room to refine.
- The flow stays in-document: output inserts where the user needs it, not in a side thread they must copy from.
- First value is one click away; depth is still available for users who want to customize.
What we would push on
- Starter-to-flow transition may feel abrupt if users expected instant output without a refinement step.
- No persistent thread view. Multi-turn AI inside one doc block needs clear boundaries vs page history.
Business strategy
Starter → refine → accept keeps users in the doc loop instead of a side chat they paste from. Output lands where the cursor is, which reinforces Notion as the system of record.
Tradeoff
| Decision | Benefit | Cost |
|---|---|---|
| Starter seeds focused follow-up, output inserts in-doc | One-click first value; no copy-paste from side panel | Refinement step before output; multi-turn boundaries less visible |
Takeaway
First value is one click away; depth is still available for users who want to customize before accept.
Pattern: Follow-up Chips
How it fits together
The pattern
- Invoke AI in the editor → pick a categorized starter → refine → accept output into the page.
- Context is implicit from the current page, block, or selection; no empty chat home.
- Starters teach prompts by example; categories scale as capabilities grow.
Where it varies
- Scope changes with selection vs whole page vs block, but the UI may not make that explicit.
- Compared with chat-native flows: Notion optimizes for users who already have a doc open, not cold-start chat.
- Starter categories must stay aligned to real jobs as AI capabilities expand.
Business strategy
Notion treats AI as an editor feature, not a separate product. Onboarding is activation at point of work with job-shaped starters, which drives AI usage inside paid workspaces instead of free-standing chat sessions.
Tradeoffs
| Decision | Benefit | Cost |
|---|---|---|
| AI embedded in editor, not standalone chat | Context is automatic; feels like a doc command | Discoverability depends on editor affordances, not an AI home |
| Categorized one-click prompt starters | First value without inventing a prompt | Category quality determines success; bad buckets feel like noise |
| Starter seeds focused follow-up, output inserts in-doc | One-click first value; no copy-paste from side panel | Refinement step before output; multi-turn boundaries less visible |
Takeaway
Steal categorized one-click starters if your AI is embedded in a host surface (docs, design files, tickets). Compare with chat-native empty composers: Notion wins when users already have context open.
Pattern: Example Prompts Library
Pattern: Follow-up Chips
Steal this
- Prompt starters at point of work, not in a separate chat home
- Categories aligned to user jobs (write, edit, summarize)
- One-click runnable examples users can edit before send
Skip this
- Blank AI chat as the only entry for embedded products
- Generic “Ask anything” with no doc context
- Making users paste results back from a side panel manually
How others getting to first value
Same job, different product bets, and what each tradeoff reveals.
Original gallery pages: Prompt Starter Examples