Study · aiListing

What can AI assistants read on 54 Spanish companies’ websites?

More and more purchases, bookings and property searches start with a question to ChatGPT, Gemini or Perplexity. To answer, those assistants read companies’ websites. We audited 54 websites of Spanish online stores, hotel groups and real-estate agencies to see what they find when they do.

Key findings

• The technical basics are usually fine. 70% of sites score A or B: assistants can get in, the site responds and the content is in the HTML.
• The most common gap: “No FAQPage schema”, on 85% of sites.
• The weakest area: structured data, averaging 55/100. Structured data is what lets an assistant state precisely what a company sells, at what price and with what ratings, rather than just summarising the page.

Grade distribution

By category

Average score of the 54 sites in each part of the audit.

What is most often missing

Share of sites where each finding appears.

By sector

Online stores: No FAQPage schema (84%); No rating markup on products (66%); No llms.txt published (28%).

Hotel groups: No FAQPage schema (79%); No llms.txt published (71%); Page title too long or too short (36%).

Real-estate agencies: No FAQPage schema (100%); No Organization schema (62%); No llms.txt published (62%).

Method

We selected 60 Spanish companies with their own website in three sectors where customers already ask AI assistants: online stores (mostly cosmetics, fashion and food), hotel groups, and real-estate agencies selling to foreign buyers. Before including each site we checked it was the brand’s real, working website.

Each site was analysed with aiListing’s public audit between 2026-10-28 and 2026-10-28: AI crawler access (robots.txt), Schema.org structured data on the homepage and on product pages, discovery files (sitemap, llms.txt), technical health (HTTPS, response time, content present in the HTML) and metadata. The score runs from 0 to 100 and the grade from A to F.

54 sites have a result. 2 could not be graded because they answered our auditor with a bot check or block page, or took too long; we do not count them as failures. 4 were excluded (store temporarily closed or result not reliable). Each company received its own result before publication, with the option to comment. Only aggregate figures are published here: no company is named or graded.

Limits. The audit measures what a website makes available to assistants, not whether an assistant recommends it — that also depends on reputation, reviews and what others publish about the brand. It is a snapshot; websites change.