numastays.com — AI Site Grade

Analysis

Numa's entire site is invisible to every major AI crawler — Vercel's security checkpoint returns HTTP 429 to GPTBot, ClaudeBot, Google-Extended, PerplexityBot, and all others tested, while the robots.txt naively declares Allow: / for * with zero AI-bot-specific rules.

Crawler Access

Every AI crawler tested against https://numastays.com — GPTBot, ClaudeBot, Google-Extended, OAI-SearchBot, PerplexityBot, ChatGPT-User, anthropic-ai, Applebot-Extended, Bytespider — receives a Vercel Security Checkpoint page (HTTP 429) with ~31KB of JS shell and zero visible content. The browser baseline returns 200 with ~290KB of real HTML. The robots.txt is 118 bytes with a single User-agent: * Allow: / rule and no mention of any AI bot. The llms.txt returns a 404 Next.js error page. The site runs on Vercel behind Cloudflare DNS (NS: chip.ns.cloudflare.com, maria.ns.cloudflare.com). The 429 is not a robots.txt block — it is a Vercel WAF/rate-limit layer that silently blocks non-browser UAs, meaning no AI model can ingest any page content through live crawling.

Cold-Knowledge Gap

A frontier LLM queried cold about "Numa Stays" recalls the brand as a Berlin-founded (2019) tech-driven hospitality challenger with over €50M in venture funding, design-led apartments, contactless check-in, and mixed reviews on Trustpilot/Google citing inconsistent cleanliness and support issues. The actual site — which the model has never crawled — presents Numa as a polished, 36-city, 8,000+ unit operator with 4.7 aggregate rating across 15,972 reviews on the Berlin page, a sustainability blog series, a corporate travel arm, and a newly launched "Native by Numa" premium brand. The model's prior knowledge is stale and negatively skewed toward early operational complaints; it knows nothing about the scale (1M+ nights served, 100% revenue growth in 2024 per press releases), the ESG positioning, the Native Places acquisition, or the current 36-city footprint.

Schema Posture

The homepage and /our-story page contain zero JSON-LD schema. City-level pages (e.g., Berlin) include BreadcrumbList, FAQPage, and Product with aggregate rating (4.7, 15,972 reviews) and price range. Individual property pages (e.g., Numa Roca Barcelona) add LodgingBusiness with full address, geo-coordinates, amenities, check-in/out times, and AggregateOffer. However, the homepage — the most crawled URL — has no Organization, WebSite, or Hotel schema, and the /our-story page (which redirects to pages.numastays.com/about-us) also has zero schema. The blog has no BlogPosting or Article schema despite 17 pages of posts.

External Signals

The press center at press.numastays.com is rich with 2024-2025 announcements: 1M+ nights served, 100% revenue growth, the Native Places acquisition (800 UK units), a "Native by Numa" premium brand launch (Oct 2025), and new GMs across Europe. The corporate site at corporate.numastays.com targets B2B travel with 24-city coverage. DNS TXT records confirm anthropic-domain-verification and apple-domain-verification are set, suggesting the brand has engaged with AI platforms for verification — yet the actual site blocks those same crawlers at the edge. No external review aggregators (Trustpilot, Reddit) surfaced in search results, meaning the model's negative prior about "mixed reviews" is based on pre-2024 data that the site's own 4.7-star aggregate contradicts.

Findings

1. All AI crawlers blocked by Vercel security checkpoint High

   Every AI crawler tested (GPTBot, ClaudeBot, Google-Extended, etc.) receives an HTTP 429 from a Vercel security checkpoint, returning a ~31KB JS shell with no visible content. The robots.txt has no AI-bot-specific rules, and llms.txt returns a 404.

   What to change: Remove or relax the Vercel WAF rate-limit for known AI crawler user agents, or add explicit Allow rules in robots.txt for those bots.

2. llms.txt file missing Medium

   The llms.txt file at numastays.com/llms.txt returns a 404 Next.js error page, preventing AI models from discovering a curated set of pages for training.

   What to change: Create an llms.txt file listing key pages (homepage, locations, blog, press) to guide AI crawlers.

3. Cold knowledge of Numa is stale and negatively skewed High

   A frontier LLM queried about 'Numa Stays' recalls early operational complaints and mixed reviews, but knows nothing about the current scale (36 cities, 8,000+ units), 4.7 aggregate rating, ESG positioning, or recent acquisitions. This gap is caused by the crawler block preventing ingestion of current content.

   What to change: Unblock AI crawlers to allow models to ingest current site content, and consider publishing a press release or updating external profiles to reflect current ratings.

4. Homepage and about page lack JSON-LD schema High

   The homepage and /our-story page contain zero JSON-LD schema. The homepage is the most crawled URL and should have Organization, WebSite, and Hotel schema to help AI models understand the brand.

   What to change: Add Organization, WebSite, and Hotel JSON-LD schema to the homepage, and Organization schema to the about page.

5. Blog posts missing Article schema Medium

   The blog page and individual blog posts lack BlogPosting or Article schema, reducing their visibility in AI-generated summaries and search results.

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

6. No external review aggregators surfaced in search Medium

   Searches for Numa reviews on Trustpilot and Reddit returned no results, meaning the model's negative prior about 'mixed reviews' is based on pre-2024 data that the site's own 4.7-star aggregate contradicts. This undermines trust signals.

   What to change: Encourage guests to leave reviews on major platforms (Trustpilot, Google) and link to them from the site.

7. Robots.txt lacks AI-bot-specific rules Low

   The robots.txt file has only a single Allow: / rule for User-agent: * and does not mention any AI crawler by name, missing an opportunity to explicitly permit or guide them.

   What to change: Add explicit Allow rules for known AI crawlers (e.g., GPTBot, ClaudeBot) to complement the Vercel fix.

8. Vercel WAF blocks all non-browser user agents High

   The Vercel security layer returns HTTP 429 to any non-browser user agent, effectively blocking all AI crawlers. This is a site-wide issue affecting every page.

   What to change: Configure Vercel WAF to allow known AI crawler user agents, or disable the security checkpoint for those bots.

What's working

• City-level pages include BreadcrumbList, FAQPage, and Product schema — City pages like Berlin include structured data for breadcrumbs, FAQs, and product with aggregate rating (4.7, 15,972 reviews), helping AI models understand offerings.
• Individual property pages include LodgingBusiness schema — Property pages like Numa Roca Barcelona include LodgingBusiness schema with address, geo-coordinates, amenities, and AggregateOffer, providing rich data for AI.
• Press center contains rich 2024-2025 announcements — The press center at press.numastays.com includes announcements about 1M+ nights served, 100% revenue growth, and acquisitions, providing authoritative content for AI models if crawled.
• Corporate site targets B2B travel with 24-city coverage — The corporate subdomain provides dedicated content for business travel, expanding the brand's reach and providing structured information for AI models.
• DNS TXT records include anthropic and apple domain verification — The DNS records show anthropic-domain-verification and apple-domain-verification, indicating engagement with AI platforms, though the crawlers are still blocked.
• FAQ page provides structured Q&A content — The FAQ page at pages.numastays.com/faq contains 523 words of Q&A content that could be leveraged for FAQ schema to improve AI visibility.