Empirank

Original Study · 17 Aug 2026

Schema Markup Wasn’t Linked to More AI Mentions in Our 3,000 Business Study

We tested 3,000 US local businesses. After the prespecified business and query controls were considered, the schema gap shrank to 0.47 percentage points and was too uncertain to use in a forecast.

Study sample
3,000 businesses
Measured outcome
Business mentions
AI platform
OpenAI

What this means for marketers

Keep schema accurate and complete for machine-readable clarity, search features, and entity consistency. Do not make it the centre of an AEO plan or promise that adding markup will increase AI mentions.

  • Use relevant business types and populate accurate identity fields instead of adding markup for its own sake.
  • Keep names, URLs, phone numbers, addresses, opening hours, and public profiles consistent with the visible page and the wider web.
  • Validate structured data and fix broken or misleading markup as part of technical SEO maintenance.
  • Prioritize the signals with clearer visibility evidence: strong Maps performance, genuine review growth, and credible third party presence.
  • Measure AI mentions with a repeated prompt panel before and after any substantial technical change.
  • Treat schema identity consistency as a question for further testing, not a proven growth tactic.

What we tested

Does adding relevant Schema.org markup make a local business more likely to appear in a search-enabled OpenAI answer? The study covered 3,000 businesses.

What the result does not prove

This study tested homepage schema against one frozen OpenAI outcome set. It did not test every page type, AI platform, or downstream benefit of structured data.

  • This was an observational linkage study, so it cannot show what would happen if the same business added or removed schema.
  • The outcome came from one search-enabled OpenAI configuration and one August 2026 collection period.
  • The sample covered US local businesses in 30 markets and 10 industries; results may differ for ecommerce, SaaS, publishers, or other countries.
  • Homepage collection succeeded for 2,479 of 3,000 business rows. Failed crawls were kept as missing rather than treated as having no schema.