what3words.com — AI Site Grade

Analysis

The homepage is a JavaScript shell that returns 7 words of visible text to every AI crawler — yet the site has a sophisticated AI strategy buried in its subdomain.

Crawler Access

All AI bots — GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot, Bytespider, Applebot-Extended — receive the same 94KB Next.js HTML blob as a browser does, with no UA-based blocking. The robots.txt has a single * rule allowing / and disallowing internal Facebook test paths and thank-you pages; no AI-specific directives exist for any bot. The homepage resolves to pretty.needed.chill?redirect=true and yields only 7 words of extractable text. The /llms.txt at the root domain returns a 404 (a Next.js error page). However, the developer subdomain openapi.what3words.com hosts a thorough llms.txt and llms-full.txt (111KB) covering API docs, SDKs, and e-commerce plugins — a split that means consumer-facing AI crawlers hitting the main domain get nothing, while developer bots get rich metadata.

Cold-Knowledge Gap

The LLM knows what3words as a geocoding system dividing the world into 3m squares, used by emergency services, logistics, and Mercedes/Uber. It also recalls safety concerns from 2023-2024 about similar-sounding word combinations causing emergency response confusion. The site itself does not address this controversy anywhere on the fetched pages — the /about FAQ and /business pages discuss precision and uniqueness but never mention the homophone risk. The site also promotes "AI agents" and "voice-ready" features, yet the cold model already knows the system is voice-friendly; the site's AI messaging reinforces rather than fills a knowledge gap.

Schema Posture

Zero JSON-LD schema was found on any fetched page — not on the homepage, /about, /business, /business/ai-agents, /products/what3words-app, or /swiftcomplete. No Organization, WebSite, FAQPage, Product, or SoftwareApplication markup exists despite the site containing FAQ blocks, comparison tables, and product descriptions that would naturally map to those types. The /about page has a comparison table (what3words vs street addresses) and a FAQ section with no FAQPage schema. The /business/ai-agents page promotes API integration for LLMs but carries no schema at all.

External Signals

The site is hosted on AWS CloudFront (A records point to CloudFront IPs, via header confirms CloudFront). DNS records show Google Workspace mail, HubSpot SPF, Stripe verification, and multiple Google Search Console verifications — indicating a mature marketing stack. The sitemap contains 1,295+ URLs across multiple languages (Arabic, Chinese, Japanese, Korean, Mongolian, Vietnamese). The /news section shows recent partnerships (Pall-Ex, Simba, IVECO, Sygic, ShipEngine) and the Swiftcomplete acquisition for address validation. The /business/ai-agents page and blog posts explicitly position what3words as AI-native, noting that LLMs already recognize the three-word format from training data but cannot decode addresses without an API subscription — a deliberate moat strategy.

Findings

1. Homepage renders as a JavaScript shell for AI crawlers High

   The homepage returns only 7 words of extractable text to AI crawlers, despite being a 94KB Next.js HTML blob. This severely limits the content available to LLMs and AI search engines.

   What to change: Implement server-side rendering or static generation for the homepage to deliver meaningful text content to all crawlers.

2. No llms.txt at the root domain High

   The root domain's /llms.txt returns a 404 error, while the developer subdomain openapi.what3words.com hosts a thorough llms.txt and llms-full.txt. This split means consumer-facing AI crawlers get no guidance or metadata.

   What to change: Create an llms.txt file at the root domain that summarizes the site's content and links to key pages.

3. No JSON-LD schema on any fetched page High

   None of the fetched pages (homepage, /about, /business, /business/ai-agents, /products/what3words-app, /swiftcomplete) include any JSON-LD structured data. The /about page has FAQ and comparison content that would benefit from FAQPage and Organization schema.

   What to change: Add JSON-LD structured data for Organization, WebSite, FAQPage, Product, and SoftwareApplication across relevant pages.

4. Safety controversy about similar-sounding words not addressed on site Medium

   The LLM recalls safety concerns from 2023-2024 about similar-sounding word combinations causing emergency response confusion, but the site's /about FAQ and /business pages do not mention or address this controversy.

   What to change: Add a dedicated section or FAQ entry that explains the system's safeguards against homophone confusion.

5. Robots.txt lacks AI-specific directives Low

   The robots.txt file contains only a single wildcard rule and does not specify any directives for AI crawlers like GPTBot, ClaudeBot, or PerplexityBot. While all bots are currently allowed, the lack of explicit rules could lead to unexpected behavior if defaults change.

   What to change: Add explicit allow/disallow rules for known AI crawlers to ensure intended access.

6. AI agents page promotes LLM integration but lacks schema Medium

   The /business/ai-agents page describes API integration for LLMs and positions what3words as AI-native, but carries no structured data to help AI crawlers understand its content.

   What to change: Add SoftwareApplication or WebAPI schema to the AI agents page to describe the API.

7. Blog posts about AI visibility lack article schema Medium

   Blog posts such as 'Does your AI tool already know what3words?' and 'Why what3words is the future of AI Voice Navigation' do not include Article or BlogPosting schema, reducing their discoverability in AI-driven search.

   What to change: Add Article or BlogPosting schema to all blog posts.

8. Sitemap lacks hreflang annotations for multilingual content Low

   The sitemap contains over 1,295 URLs across multiple languages (Arabic, Chinese, Japanese, etc.) but does not include hreflang annotations, which could confuse AI crawlers about language targeting.

   What to change: Add hreflang annotations to the sitemap to indicate language and regional variants.

What's working

• Developer subdomain hosts thorough llms.txt and llms-full.txt — The openapi.what3words.com subdomain provides a comprehensive llms.txt and llms-full.txt (111KB) covering API docs, SDKs, and e-commerce plugins, giving developer-focused AI crawlers rich metadata.
• All major AI crawlers are allowed access — The robots.txt and server configuration allow all 11 tested AI bots (GPTBot, ClaudeBot, PerplexityBot, etc.) to crawl the site, with no UA-based blocking.
• Site actively positions itself as AI-native with dedicated pages and blog posts — The /business/ai-agents page and blog posts like 'Does your AI tool already know what3words?' explicitly discuss LLM integration and voice navigation, reinforcing the brand's AI relevance.
• Mature marketing and infrastructure stack — DNS records show Google Workspace, HubSpot, Stripe, and Google Search Console verification, indicating a well-managed online presence with strong external signals.
• Sitemap covers over 1,295 URLs across multiple languages — The sitemap includes a large number of URLs in languages such as Arabic, Chinese, Japanese, Korean, Mongolian, and Vietnamese, supporting international AI visibility.
• OpenAPI documentation is publicly available — The openapi.what3words.com subdomain hosts API documentation that can be consumed by AI agents and developers, supporting integration.