How AI reads Notion

notion.so Jun 10, 2026 6 min read Emerging

Short

43/ 100

AEO Level 3Emerging

The short answer

Notion scores 43/100 on our AEO scan — a sharply divided result. Content structure hits a perfect 100, so AI can parse the homepage clearly. But agent interfaces and identity authentication both score zero, leaving AI assistants unable to discover or use the very API that Notion's developer ecosystem is built around.

What AI sees

An AI agent landing on Notion's homepage finds clean, parseable prose — and almost nothing else it can act on.

The homepage delivers well-structured HTML with an unambiguous content hierarchy, which is why content_structure reaches a perfect 100. An AI can identify what Notion is, extract its value propositions, and summarize the product with high confidence. The trouble starts the moment that agent tries to go further. There are no explicit rules in robots.txt for AI crawlers like GPTBot or ClaudeBot, no Link response headers pointing to useful resources, and no API catalog document despite Notion shipping a widely-used public REST API. A productivity platform whose entire value proposition is structured information leaves its own agent-facing information layer almost completely blank.

Notion's homepage as rendered for an AI agent: rich semantic content, zero programmatic agent hooks

Notion's robots.txt: no user-agent declarations for GPTBot, ClaudeBot, or PerplexityBot

Where it loses points

Agent interfaces and identity authentication both score zero — striking for a platform whose power users live inside integrations and automations.

How to fix it

Three targeted changes, all achievable without touching Notion's product code, would unlock the agentic layer this platform is already architecturally ready for.

Publish an MCP Server Card

Goal

Expose Notion as a first-class tool source for AI agents via the Model Context Protocol.

Issue

No MCP Server Card exists at /.well-known/mcp/server-card.json, so agents cannot discover or connect to Notion's capabilities.

Fix

Serve /.well-known/mcp/server-card.json with a serverInfo block, a transport endpoint pointing to Notion's API gateway, and a capabilities list covering page reading, database querying, and block writing. This single file lets any MCP-compatible agent treat Notion as a native tool without custom configuration.

Declare AI crawler rules in robots.txt

Goal

Give AI crawlers explicit permission signals so Notion's public content is indexed accurately and on Notion's own terms.

Issue

Notion's robots.txt contains no user-agent rules for GPTBot, ClaudeBot, PerplexityBot, or any other major AI crawler.

Fix

Add explicit Allow or Disallow stanzas for each major AI crawler user-agent, followed by a Content-Signal directive declaring preferences for ai-train, search, and ai-input. This closes the ambiguity gap and ensures citation engines treat Notion's public help docs and marketing pages as intended.

Serve an API Catalog at /.well-known/api-catalog

Goal

Make Notion's public REST API automatically discoverable by agents following RFC 9727.

Issue

Despite a mature, developer-facing REST API, there is no api-catalog document at the standard well-known path.

Fix

Publish /.well-known/api-catalog as application/linkset+json with entries for each major API surface — pages, databases, blocks, users, and search. Agents following RFC 9727 will find and integrate Notion's API without relying on manually maintained connector libraries as an intermediary.

Common questions

Why does Notion score 43/100 if its content structure is perfect?

AEO scoring weighs five categories, not just one. Notion's content_structure hits 100, but agent_interfaces and identity_auth both score zero — meaning AI agents can read the homepage but cannot discover the API, authenticate actions, or connect as a tool. Two zeroes against a perfect score and two mid-range results average out to 43.

What does a zero agent_interfaces score mean for Notion users in practice?

It means AI assistants have no standardized path to discover or invoke Notion's capabilities. A user asking their agent to add a row to a Notion database must rely on pre-built, manually maintained connectors rather than autodiscovery. An MCP Server Card and an API Catalog would change that immediately, making Notion natively accessible to any standards-compliant agent.

Does the missing robots.txt AI crawler guidance put Notion's content at risk?

Not at direct risk, but at the mercy of each crawler's default behavior. Without explicit user-agent stanzas, AI crawlers decide for themselves what to index, train on, or cite. For a platform hosting public help documentation, changelogs, and marketing pages, that ambiguity means Notion loses control over how its own content appears inside AI-generated answers — a real brand and accuracy concern.