Global AI Visibility Study 2026: 100 Business Websites Across 20 Markets

An AI assistant does not recommend a business because of one file or SEO plugin. Search and answer engines first need to discover the site, understand the entity, and verify its claims. This study audits 100 business homepages across five sectors and 20 markets with one consistent technical checklist.

Research design

On November 9, 2026, we reviewed Austin, Toronto, Manchester, Dublin, Berlin, Lyon, Barcelona, Milan, Amsterdam, Stockholm, Oslo, Warsaw, São Paulo, Monterrey, Cape Town, Dubai, Bangalore, Singapore, Melbourne, and Auckland. One official business site was selected in dental, legal, hotel, construction, and manufacturing for each market.

Directories, social profiles, public bodies, and aggregators were excluded. One eligible official business website was selected per market and sector. Exact search-result positions were not retained, so the dataset reports search_rank as null and makes no first-result claim.

What we measured

• Homepage reachability, HTTPS, and indexability.
• Title, meta description, canonical, and HTML language.
• JSON-LD, entity schema, and organization details.
• Open Graph, H1, and heading structure.
• Contact, about, privacy, service/product, and FAQ signals.
• robots.txt, XML sitemap, and explicit blocks for known AI crawlers.
• llms.txt as a separate observation, not a required scoring factor.

Each site received 21 binary checks. Fifteen sites that did not return HTTP 200 to the standard research client were excluded from quality percentages; availability is reported separately. This avoids confusing bot protection with missing page features.

Headline results

Reachable sites averaged 78.5, with a median score of 81. The largest gaps were not basic indexing tags; they were content paths that clearly explain who the business is and which questions it answers.

What the findings suggest

Structured data appeared on 76.7% of reachable sites, but only 61.6% described an organization or local business. Just 57.0% supplied at least three entity details such as name, URL, phone, address, or sameAs. Many sites publish JSON-LD without describing the entity deeply.

Only 24.4% showed an FAQ signal. FAQ schema is not a ranking switch, but concise, accurate answers to real customer questions create clear information units for users and citation systems. Useful topics include process, scope, timing, pricing logic, and limitations.

About links appeared on 53.5% and privacy links on 41.9%. Company history, people, location, contact details, and policy pages provide mutually reinforcing entity evidence. Their absence does not cause invisibility by itself, but it reduces verifiable context.

Sector comparison

This ranking is not a verdict on entire industries; each market contributes only one site. It is useful for generating a testable hypothesis that manufacturing and construction sites have less consistent entity detail, service paths, and metadata.

Practical checklist for AI visibility

• Use one consistent brand name, description, phone, address, and set of official profiles.
• Populate Organization or LocalBusiness schema with real logo, URL, contactPoint, address, and sameAs data.
• Create a page for each core service explaining scope, process, fit, and limitations.
• Answer genuine customer questions clearly and without unsupported claims.
• Show authors, update dates, methods, sources, and datasets where relevant.
• Validate robots.txt and sitemaps and check for accidental crawler blocks.
• Notify changed URLs through IndexNow while treating it as notification, not an indexing guarantee.
• Make brand claims independently verifiable through trustworthy sources.

A note on llms.txt

llms.txt can be a useful navigation layer, but it is not a universal ranking standard or recommendation guarantee. We observed it but did not include it as a required score. Accessible HTML, internal links, structured data, original evidence, and consistent entity information come first.

Limitations

The purposive sample is small: five sites per market. The audit covers homepages only. Bot protection prevented standard-client access to some sites. Binary checks do not measure factual accuracy, reputation, backlink quality, user satisfaction, or actual LLM citations. The results do not establish causality.

The evidence package is public: companies and row-level results, exact metric definitions, audit source code, and the machine-readable summary. It includes domain, market, sector, selection query, attempted and final URLs, HTTP status, and every check result.

How exactly were the metrics calculated?

The score is the equally weighted mean of 21 binary checks: round(100 × passed checks / 21). A suitable title is 15–70 normalized characters; a meta description has at least 40 characters; heading structure means at least three H1–H3 elements; entity details means at least three of name, url, address, telephone, and sameAs occur in source HTML. FAQ and service-path checks use the exact strings and URL fragments published in methodology.json and the source code.

Conclusion

The durable strategy is not writing special copy for LLMs; it is becoming a clearly defined, verifiable entity on the web. Accessibility is the foundation. Entity schema, detailed service pages, answers to real questions, sources, and consistent organization facts help systems understand the business. IndexNow can accelerate discovery; evidence, relevance, and trust determine recommendations.