How AI reads Notion

notion.so Jun 18, 2026 6 min read Emerging

Short

43/ 100

AEO Level 3Emerging

The short answer

Notion scores 43 out of 100 on our AEO scan — a midpoint result shaped by perfect content structure but zero agent interface support. The platform's workspace and collaboration features are well-written, but AI agents visiting today find no MCP server card, no AI crawler rules in robots.txt, and no machine-readable discovery endpoints.

What AI sees

When an AI agent visits Notion's homepage today, it lands on content-rich terrain that scores perfectly for structure but immediately runs into a protocol dead end.

Notion's homepage serves clean, semantically rich HTML with well-defined headings covering workspace organization, team wikis, and project tracking — content an LLM can summarize with high fidelity, explaining the perfect content_structure score of 100. But the agent experience deteriorates quickly at the protocol boundary. The robots.txt includes no per-agent directives for GPTBot, ClaudeBot, or PerplexityBot, leaving crawlers to infer permissions. No Link response headers appear on the homepage to point autonomous clients toward APIs or structured resources. None of the standard agent discovery endpoints — API catalog, MCP server card, agent skills index — exist. For a product built entirely around structured information, that silence is a missed opportunity.

Notion homepage as seen by an AI crawler: rich content structure scoring 100, but zero agent discovery signals in response headers

Notion robots.txt today: no GPTBot, ClaudeBot, or PerplexityBot user-agent blocks and no Content-Signal directives

Where it loses points

The sharpest drop comes from agent interfaces and identity authentication, both sitting at zero out of 100 — meaning Notion is invisible to agents that follow modern discovery protocols.

How to fix it

Three targeted changes would transform Notion from a passively-scraped web property into an active participant in the agentic web.

Declare AI crawler rules in robots.txt

Goal

Give AI crawlers explicit permission or restriction rules so they know which paths they may index and cite.

Issue

The robots.txt contains no user-agent blocks for GPTBot, ClaudeBot, PerplexityBot, or any other AI crawler.

Fix

Add dedicated User-agent blocks for each major AI crawler followed by Allow or Disallow directives. For public marketing and help content, Allow /; for private workspace paths you do not want cited, add matching Disallow lines.

Publish an MCP Server Card

Goal

Expose Notion's capabilities to AI agents that follow the Model Context Protocol discovery standard.

Issue

No MCP Server Card exists at the standard discovery path, leaving agent_interfaces at zero.

Fix

Serve a JSON document at /.well-known/mcp/server-card.json listing serverInfo, the transport endpoint, and available capabilities such as page read, database query, and block creation. This enables MCP-compatible agents to integrate directly without manual configuration.

Return Markdown for agent requests

Goal

Deliver page content as clean Markdown when an agent signals it prefers machine-readable prose over HTML.

Issue

The site returns HTML regardless of the Accept header, so agents requesting text/markdown receive a layout-heavy document instead of pure content.

Fix

Detect the Accept: text/markdown request header on public pages and return a Markdown rendering with Content-Type: text/markdown. Keep HTML as the default for browsers. This lets AI agents ingest Notion's help docs and marketing content without DOM parsing.

Common questions

What does Notion's AEO score of 43 out of 100 actually mean?

It means Notion handles classic web fundamentals well — content is structured and readable — but the platform has not yet adopted the protocol-level signals that AI agents use for autonomous discovery. Scores above 70 typically indicate a site actively optimizes for answer engines; at 43, Notion sits in the middle tier with meaningful gaps in agent interfaces and identity authentication.

Can AI agents use Notion content today despite the low score?

Yes, but only through generic web scraping. Agents can read Notion's public marketing pages and help documentation, but they cannot discover its APIs programmatically, cannot confirm what content is authorized for training or citation, and cannot integrate via MCP without manual configuration. The information is accessible; the discoverability layer is missing.

Which fix would have the fastest impact on Notion's AEO score?

Adding AI crawler rules to robots.txt is the fastest change — it requires editing a single text file, needs no infrastructure deployment, and immediately signals to GPTBot, ClaudeBot, and PerplexityBot which Notion paths they may cite. It would lift the agent_discovery score and address the Content-Signal gap simultaneously.