Notion AI Prompts
Relevant source files
The following files were used as context for generating this wiki page:
Notion AI is an agentic assistant integrated directly into the Notion workspace, designed to interact with users via a chat interface or sidebar. It operates within a closed loop triggered by user messages, utilizing a specialized set of tools to query, create, and modify Notion-specific entities such as Workspaces, Pages, Databases, and Data Sources. The system is designed to prioritize internal workspace information over general knowledge, often performing default searches to ensure responses are grounded in the user's specific context.
The scope of Notion AI includes content drafting, database management, and semantic search across the workspace and the web. It adheres to strict formatting guidelines using "Notion-flavored Markdown" to ensure that AI-generated content remains visually consistent with the rest of the Notion platform.
Sources: NotionAi/Prompt.txt:1-15, NotionAi/Prompt.txt:208-215
System Architecture and Concepts
The Notion AI system is built around the core structural components of the Notion platform. The prompt logic defines how the AI perceives and interacts with these objects.
Core Entities
The system recognizes three primary structural levels:
- Workspace: A collaborative space containing all pages, databases, and users.
- Pages: Individual documents containing properties and body content. Pages can be nested within other pages or data sources.
- Databases: Containers for data sources and views, which can be rendered inline or as full pages.
Sources: NotionAi/Prompt.txt:17-21, NotionAi/Prompt.txt:30-33
Data Flow and Tool Loop
Notion AI follows a tool-calling loop where it continues to execute actions until the user's request is resolved. It is instructed to call tools immediately if a request can be resolved through them, particularly for keyword-based queries like "wifi password" or "Q4 revenue."
The following diagram illustrates the high-level interaction flow:
Database and Data Source Logic
Notion AI differentiates between Source Databases (which own a data source) and Linked Databases (which reference data sources). Data Sources serve as the underlying storage mechanism, defined by a schema of properties.
Supported Property Types
The AI can manage various property types, each requiring specific formatting for updates.
| Property Type | Description | Format Requirement |
|---|---|---|
title | The primary identifier for a page. | String (Required) |
date | Single date or range. | ISO-8601 string; start is required if end is provided. |
relation | Links to pages in other data sources. | Array of URLs. |
status | Progress tracking (Todo, In Progress, Done). | Must match exact option names. |
place | Geographic location. | Must include latitude and longitude. |
Sources: NotionAi/Prompt.txt:46-64, NotionAi/Prompt.txt:87-111
View Configurations
The AI supports creating and updating several view types to render database information:
- Table: Default spreadsheet-like view.
- Board: Kanban-style organization.
- Map: Requires at least one
placeproperty. - Calendar/Timeline: Requires at least one
dateproperty.
Sources: NotionAi/Prompt.txt:120-134
Search Strategy and Intelligence
The AI employs a specific hierarchy when retrieving information to minimize inaccuracies. It is instructed never to rely on internal memory if workspace information could provide a more accurate answer.
Search Types
- Internal Search: Queries the user's workspace and third-party connectors. Used for company-specific data.
- Web Search: Queries the public internet. Used for real-time facts (e.g., weather).
- Default Search: A superset that combines internal and web results. This is the default starting point for most queries to "cast the widest net."
Sources: NotionAi/Prompt.txt:172-200
Notion-Flavored Markdown (NFM)
To maintain platform consistency, the AI generates content using specialized Markdown tags. This includes advanced block types that are not part of standard Markdown.
Advanced Block Types
- Callouts:
<callout icon="emoji" color="Color">Content</callout> - Columns:
<columns><column>Content</column></columns> - Synced Blocks:
<synced_block url="URL">Content</synced_block> - Mentions: Support for users, pages, and dates (e.g.,
<mention-user url="URL"/>).
Sources: NotionAi/Prompt.txt:306-368
Operational Constraints and Security
The system prompt enforces strict behavioral guardrails to prevent over-performance and ensure user privacy.
- Gender Neutrality: The AI must use gender-neutral language (they/them) unless a person's gender is explicitly confirmed in the context.
- No Unsolicited Actions: The AI is forbidden from editing user pages unless explicitly asked. It should not offer to contact people or use external tools outside of defined connectors.
- Refusals: The AI must acknowledge its inability to manage permissions, billing, sharing, or account settings, directing users to the help documentation instead.
Sources: NotionAi/Prompt.txt:265-285, NotionAi/Prompt.txt:240-258
Implementation Tools (JSON)
The underlying capabilities of Notion AI are defined by a set of tool schemas. These tools allow the agent to traverse the workspace and perform modifications.
| Tool Name | Purpose | Key Parameters |
|---|---|---|
search | Semantically search workspace and web. | questions (array), scopes |
update-page | Modify content or properties of a page. | pageUrl, properties, content |
create-database | Create a new database container. | parentPageUrl, title, views |
query-data-sources | Perform structured queries on data sources. | dataSourceUrls, filter, sort |
Sources: NotionAi/tools.json, NotionAi/Prompt.txt:5-10
Notion AI Prompts prioritize grounding and precision, utilizing a robust toolset to bridge the gap between large language model capabilities and structured workspace data. By adhering to Notion-flavored Markdown and specialized search strategies, the AI ensures that its output is both accurate to the user's private data and native to the Notion interface.