notion.so — AI Site Grade

Analysis

I have enough. Key findings: notion.so 301s to notion.com; AI crawlers get 200 on .com (the .so "errors" were tool read-limits on the CSP header, not blocks); Bytespider and Google-Extended get 403 from Cloudflare; zero JSON-LD on homepage, /product/ai, and /pricing despite rich FAQ content; llms.txt exists but describes the OLD positioning ("connected workspace," "50 million users") while the site now sells "AI workspace / agents." Cold LLM knowledge is also stale (Cron acquisition, $10B valuation, no agents).

Let me write the audit.

Notion: strong crawl access, but a positioning split between what the site sells and what AI already believes

Notion's AI-visibility problem is not access - it is a two-year positioning lag baked into llms.txt and confirmed by cold LLM priors. The site now sells "the AI workspace" with Custom Agents; the machine-readable layer still describes a 2023-era "connected workspace" for notes and wikis.

Crawler Access

notion.so 301-redirects to notion.com, and the canonical, sitemaps, and llms.txt all live on the .com host. On notion.com, GPTBot, ClaudeBot, OAI-SearchBot, PerplexityBot, anthropic-ai, and Applebot-Extended all return HTTP 200 with full content - no UA-based blocking. The two exceptions are Bytespider and Google-Extended, both 403'd by Cloudflare ("Attention Required" challenge). robots.txt names no AI bots at all; it only blocks SEO scrapers (AhrefsBot, SemrushBot, BLEXBot, Amazonbot, dotbot). Blocking Google-Extended is a deliberate opt-out from Gemini grounding, but it is undocumented in robots.txt.

Cold-Knowledge Gap

A frontier model asked cold about Notion describes blocks, databases, Cron/Calendar, the 2021 ~$10B valuation, and "criticism over performance and offline support" - and never mentions Custom Agents, Enterprise Search, AI Meeting Notes, or "Workers." The homepage H1 is now "Where teams and agents Think together." The prior is roughly two product cycles stale, and llms.txt reinforces it: it leads with "connected workspace," "50 million people," and lists AI as one bullet among ten products.

Schema Posture

Zero JSON-LD on the homepage, /product/ai, and /pricing - no Organization, SoftwareApplication, Product, or FAQPage. This is the sharpest gap: /pricing and /product/ai contain genuine FAQ blocks ("What can Notion AI help me do?", "What are Custom Agents?") and a full plan-comparison table, all invisible to structured-data parsers. The llms.txt is the only machine-readable positioning asset, and it is out of date.

External Signals

llms.txt and the sitemap are healthy (a 90KB help sitemap with lastmod dates), but off-domain authority is fragmented across notion.so, notion.com, developers.notion.com, academy.notion.com, and notion.dev - splitting the citation surface AI engines draw from.

Findings

1. llms.txt describes an outdated connected-workspace positioning High

   The llms.txt file leads with 'connected workspace' and '50 million people' and lists AI as one bullet among ten products, while the site now sells 'the AI workspace' with Custom Agents. This machine-readable asset reinforces a two-year-old product narrative for AI engines.

   What to change: Rewrite llms.txt to lead with the current AI workspace and Custom Agents positioning, and remove or demote the legacy connected-workspace framing.

2. Zero JSON-LD on homepage, /product/ai, and /pricing High

   The homepage, /product/ai, and /pricing contain no Organization, SoftwareApplication, Product, or FAQPage structured data. Rich FAQ blocks and a full plan-comparison table on those pages are invisible to structured-data parsers.

   What to change: Add JSON-LD for Organization, SoftwareApplication, Product, and FAQPage on the homepage, /product/ai, and /pricing, marking up the existing FAQ blocks and plan-comparison table.

3. Cold LLM priors describe a two-product-cycle-old Notion High

   A frontier model asked cold about Notion describes blocks, databases, Cron/Calendar, the 2021 ~$10B valuation, and performance criticism, but never mentions Custom Agents, Enterprise Search, AI Meeting Notes, or Workers. The prior is roughly two product cycles behind the current homepage H1.

   What to change: Publish and promote current AI-product content and structured data so AI engines can update their priors on Notion's agent capabilities.

4. Bytespider and Google-Extended receive 403 from Cloudflare Medium

   Cloudflare returns a 403 'Attention Required' challenge to Bytespider and Google-Extended on notion.com, while other major AI crawlers get 200. Blocking Google-Extended is a deliberate opt-out from Gemini grounding but is undocumented in robots.txt.

   What to change: Document the Google-Extended opt-out in robots.txt and review whether Bytespider should be allowed or explicitly disallowed.

5. robots.txt names no AI bots and only blocks SEO scrapers Medium

   The robots.txt file contains six user-agent rules but names no AI crawlers, only blocking SEO scrapers such as AhrefsBot, SemrushBot, BLEXBot, Amazonbot, and dotbot. AI crawler policy is therefore implicit rather than declared.

   What to change: Add explicit user-agent rules for AI crawlers to robots.txt so access policy is declared rather than inferred.

6. Off-domain authority is fragmented across five Notion domains Medium

   Citation surface is split across notion.so, notion.com, developers.notion.com, academy.notion.com, and notion.dev, diluting the authority AI engines draw from for a single brand.

   What to change: Consolidate canonical signals and cross-link the subdomains so AI engines attribute authority to the primary notion.com surface.

7. notion.so 301-redirects to notion.com, splitting the canonical host Low

   The audited domain notion.so 301-redirects to notion.com, where the canonical, sitemaps, and llms.txt live. The redirect is functional but leaves two hosts in circulation for the same brand.

   What to change: Ensure all external references and citations point to notion.com to reduce host-level dilution.

What's working

• Major AI crawlers receive HTTP 200 with full content on notion.com — GPTBot, ClaudeBot, OAI-SearchBot, PerplexityBot, anthropic-ai, and Applebot-Extended all return HTTP 200 with full content on notion.com, with no UA-based blocking. This gives AI engines direct access to the site's current content.
• llms.txt is published and served successfully — A 6,776-byte llms.txt file is served with HTTP 200, providing AI engines a machine-readable positioning asset. It is the only such asset in place, though its content is out of date.
• Help sitemap is healthy with lastmod dates — A 90KB help sitemap is served successfully and includes lastmod dates, giving crawlers a well-maintained index of help content.
• Key pages contain rich FAQ blocks and plan-comparison content — /product/ai and /pricing contain genuine FAQ blocks such as 'What can Notion AI help me do?' and 'What are Custom Agents?', plus a full plan-comparison table. This content is substantive and ready for structured-data markup.