Structured Data Is Now AI Infrastructure

I want to revisit something I wrote in mid-2024 about GEO (Generative Engine Optimisation) because the landscape has shifted enough that the earlier piece needs an update, and because one specific technical layer has become considerably more important than it was then.

The GEO argument in mid-2024 was essentially: AI systems are becoming a significant source of product discovery, your products need to be visible and accurately represented in AI-mediated discovery contexts, and the signals that determine AI visibility are different from traditional SEO signals. That argument was correct but somewhat abstract. The mechanisms are now clearer, the evidence is in, and the stakes are higher.

What Microsoft and Google Confirmed in 2025

Fabrice Canel, Principal Product Manager at Microsoft Bing, said at SMX Munich in March 2025 that "Schema Markup helps Microsoft's LLMs understand content." That is not a vendor claim or a forward-looking statement. That is the platform itself describing how it processes the web.

Google's position is consistent. In March 2025, Google publicly stated that structured data is critical for modern search features because it is efficient, precise, and easy for machines to process. By May, both companies reinforced the message: structured data is used directly in their Generative AI features, not just as a signal for rich snippets.

Gemini-powered AI Mode uses schema markup to verify claims, establish entity relationships, and assess source credibility when synthesising answers. Your Product schema is not just helping serve rich results. It is being used to decide whether your product information is reliable enough to cite in an AI-generated response.

This is the framing that changes the priority calculus. Before 2025, schema was worth doing because it might earn you star ratings in search results. Now it determines whether AI systems trust and surface your products at all.

The Product Schema Basics

For ecommerce Product schema, the elements that matter most for AI understanding are fairly consistent with what SEO practitioners have recommended for years. What has changed is the reason they matter.

Product name and description: the schema name and description fields are the primary signals an AI system uses to understand what the product is. These should be factually accurate, complete, and consistent with the page content. Consistency is the new requirement. AI systems that encounter a schema description contradicting the rendered page text will distrust both.

Offers: price, availability, currency, and condition. The Offer object within Product schema tells AI systems whether the product can be purchased, at what price, and in what state. This is the information an AI assistant needs to confidently recommend a product. Outdated pricing or availability in schema is actively harmful. Google's own guidance is explicit: structured data must match what is visible on the page. Schema that says one price while the page says another sends a conflicting signal that AI systems cannot resolve cleanly.

AggregateRating: review count and average rating. AI systems weight reviews as a trust signal. A product with 847 reviews at 4.3 stars is recommended more confidently than an identical product with no review data in the schema. This is measurable.

Brand, manufacturer, and mpn (manufacturer part number): entity data that helps AI systems understand the relationship between your product listing and the product in their knowledge graph. Particularly important for comparison contexts, where an AI is trying to determine whether two product listings are for the same product.

The Machine Readability Gap

Adobe's analysis of the US retail sector found that product pages average a machine-readability score of 66%, meaning roughly a third of product page content is currently invisible to AI systems. Homepages average 75%. These are US figures, but there is no reason to assume UK retailers are performing materially better given similar platform and PIM inheritance.

The gap exists because of three compounding problems. Much product content is generated from PIM exports built for keyword SEO, not structured data. The workflow that creates a product listing does not include schema generation as a standard step. Product data is also dynamic: prices change, availability changes, specifications get updated, and schema not kept in sync with the underlying data degrades quickly. "Dynamic schema" that reads from the same source as the rendered page is best practice; static schema hand-written once and never updated is nearly useless after a few months. And many ecommerce platforms render schema via JavaScript rather than inline HTML, generating it client-side after page load. AI crawlers vary significantly in how well they handle JavaScript-rendered content, making client-side schema coverage unreliable.

The context here is worth noting. AI traffic to US retail sites grew 393% year-on-year in Q1 2026, and AI-referred traffic now converts 42% better than non-AI channels. The traffic is arriving. The infrastructure to capture it is not.

The Practical Guidance

If you are running a mid-market UK ecommerce operation and trying to prioritise structured data work, here is the order I would suggest.

Audit what you actually have. Run your top-selling product pages through Google's Rich Results Test and look at what schema is present, what is valid, and what is returning errors. Errors in schema are worse than absent schema for AI systems; they signal unreliable data rather than no data.

Fix what is wrong before adding what is missing. A Product schema with incorrect pricing or invalid @type values is actively confusing. Sort the existing mess before expanding it.

Prioritise server-side schema over client-side. If your platform generates Product schema in JavaScript, have a conversation with your engineering team about server-rendering it in the HTML instead. The coverage improvement for AI crawlers is meaningful and does not require a full platform migration.

Expand to FAQPage schema for product FAQ content. Google deprecated FAQ rich results in 2023, which confused the investment case for a while. But FAQPage schema retains value for AI systems' semantic understanding: question-and-answer content about products, properly marked up, is exactly what AI assistants draw on when answering detailed product questions. The rich result is gone; the semantic signal remains.

None of this is glamorous. It is technical hygiene. But it is the kind of hygiene that zero-click search and AI discovery trends are turning into a structural competitive advantage. The gap between retailers who have clean, dynamic, server-rendered schema and those who do not is measurable. It is growing.

The llms.txt conversation has moved this territory further still. Structured data and machine-readable content are converging into the same underlying question: can AI systems understand and trust your product data? Schema.org has been trying to answer that question for over a decade. It turns out the AI era is the moment it actually matters.