2.9Advanced10 min

Structured Data as AI Comprehension Layer: Confirmed and Measurable

Lucas Blochberger·7 April 2026·Updated 11 June 2026

Definition

Structured data has evolved from a rich-results tool into an AI comprehension layer in 2025-2026. Google, Microsoft and ChatGPT confirm the use for generative features. Microsoft's NLWeb initiative makes every schema-annotated website automatically an MCP server for AI agents.

Key Takeaways

✓Google confirmed Structured Data for AI-Features (March/May 2025)
✓Microsoft confirmed Schema usage for Copilot at SMX Munich
✓ChatGPT Shopping (50M daily shopping queries) prioritizes Product Schema
✓NLWeb makes schema-annotated sites into MCP Servers for AI-Agents
✓Search Engine Land Experiment: Only the page with Schema appeared in AI Overview
✓Google has discontinued 7 Schema types since June 2025
✓Counterpoint: No peer-reviewed studies on Schema-AI impact

Structured Data has transcended the Rich Results context and become the primary machine-readable layer through which AI systems parse website meaning.

The threefold confirmation

At SMX Munich in March 2025, Bing's Fabrice Canel explained: "Schema Markup helps Microsofts LLMs understand content." Google confirmed in April/May 2025 that Structured Data offers benefits for AI search features. ChatGPT Shopping (launched November 24, 2025, 50 million daily shopping queries) prioritizes server-side rendered Product Schema in JSON-LD.

NLWeb: Schema as AI infrastructure

Microsoft's NLWeb was announced on May 19, 2025 by R.V. Guha — the inventor of RSS, RDF and Schema.org, now Microsoft CVP and Technical Fellow. NLWeb crawls websites, extracts Schema.org JSON-LD and converts it into queryable data. Each NLWeb instance automatically becomes an MCP server, making website content discoverable for AI agents across the MCP ecosystem.

Discontinued schema types

Google has discontinued seven Structured Data types since June 2025. FAQ Rich Results were restricted to government and health pages (August 2023). HowTo was completely discontinued in September 2023. Important: The discontinuation affects Rich Result display, not semantic use by AI systems.

The counterpoint

A December 2024 study (Search/Atlas) found no correlation between Schema coverage and AI citation rates. There are no peer-reviewed studies on Schema-AI impact. Schema functions as an AI understanding accelerator in combination with good content, not as a standalone signal.

Data & Statistics

JSON-LD war 2024 auf 41 Prozent der Seiten vorhanden (gegenueber 34 Prozent 2022) und ist das am schnellsten wachsende der drei Formate (JSON-LD, RDFa, Microdata)

Web Almanac 2024 (HTTP Archive), Kapitel Structured Data (2024)

JSON-LD-Typverteilung (mobil, 2024): WebSite 12,73 Prozent, Organization 7,16 Prozent, BreadcrumbList 5,66 Prozent, LocalBusiness 3,97 Prozent, Product nur 0,77 Prozent

Web Almanac 2024 (HTTP Archive), Kapitel Structured Data (JSON-LD section, mobile pages) (2024)

Fabrice Canel (Microsoft Bing) bestaetigte auf der SMX Muenchen im Maerz 2025, dass Schema-Markup den LLMs von Microsoft (Copilot) hilft, Inhalte zu verstehen

Search Engine Land (Barry Schwartz), zitiert Fabrice Canel, Principal Product Manager, Microsoft Bing, SMX Muenchen (2025)

Das Google-Search-Team erklaerte im April 2025, dass strukturierte Daten einen Vorteil in den Suchergebnissen verschaffen

Search Engine Land, Schema markup and AI search (with no hype) (2025)

Jede NLWeb-Instanz ist zugleich ein Model-Context-Protocol-Server (MCP); NLWeb baut auf vorhandenen Schema.org- und RSS-Daten plus LLMs auf und wurde von R.V. Guha (Schoepfer von RSS, RDF, Schema.org) entwickelt; vorgestellt am 19. Mai 2025

Microsoft Source, Introducing NLWeb: Bringing conversational interfaces directly to the web (2025)

Generative Engine Optimization (GEO) kann die Sichtbarkeit in generativen Antworten um bis zu 40 Prozent steigern; Quellen, Zitate und Statistiken erhoehen die Sichtbarkeit um ueber 40 Prozent; auf Perplexity.ai bis zu 37 Prozent; Benchmark aus 10.000 Anfragen und 9 Methoden

GEO: Generative Engine Optimization, Aggarwal et al. (Princeton et al.), KDD 24, arXiv:2311.09735v3 (2024)

Crawl-to-refer-Verhaeltnis (Juli 2025): Anthropic 38.065,7 zu 1, OpenAI 1.091,4 zu 1, Google 5,4 zu 1; rund 80 Prozent des KI-Crawlings dienen dem Training (18 Prozent Suche, 2 Prozent Nutzeraktionen)

Cloudflare Blog, From crawl to click: the bots, the gap, and the road ahead (2025)

In Deutschland nutzen 67 Prozent der Menschen ab 16 Jahren zumindest gelegentlich generative KI (2025, gegenueber 40 Prozent im Sommer 2024); ChatGPT 43 Prozent, Copilot 39 Prozent, Gemini 28 Prozent

Bitkom, repraesentative Befragung (Daten fuer Deutschland) (2025)

ChatGPT erreichte 800 Millionen woechentlich aktive Nutzer (Ankuendigung 6. Oktober 2025)

TechCrunch (Sam Altman / OpenAI, Dev Day Keynote) (2025)

8,69 Millionen Internetnutzer in Oesterreich bei 95,3 Prozent Penetration (Januar 2025)

DataReportal, Digital 2025: Austria (2025)

FAQ

What are structured data and why are they important for AI?

Structured data are machine-readable annotations, usually following the Schema.org vocabulary in JSON-LD format, that communicate the meaning of content to a machine. Instead of displaying a phone number merely as text, the markup explicitly states that it is a phone number or a product price. They are important for AI because generative systems use this semantic layer to extract entities and facts more reliably. Microsoft has officially confirmed that Schema markup helps its LLMs understand content.

Does ChatGPT use structured data?

This has not been directly confirmed. For ChatGPT, there is no comparably clear official statement as with Microsoft that Schema markup is processed as a distinct signal. However, ChatGPT Search builds on the Bing index, and Microsoft has confirmed that Schema helps its LLMs. Anyone optimizing content for the Bing index is thereby indirectly addressing ChatGPT Search as well. An independent, causal Schema effect on ChatGPT is not, however, supported by peer-reviewed studies.

Which format should I use for structured data, JSON-LD or Microdata?

JSON-LD. It is the format recommended by Google and the de facto dominant one. According to the Web Almanac 2024, JSON-LD was present on 41 percent of pages, compared to 34 percent in 2022, and is the fastest-growing of the three formats. JSON-LD is bundled in a script block, separate from the visible HTML, which makes maintenance and validation easier. Microdata and RDFa weave the markup into the HTML tags and are more error-prone.

What is NLWeb and what does it mean for my website?

NLWeb is a project introduced by Microsoft in May 2025 that combines existing markup such as Schema.org and RSS with language models into a natural-language interface. Crucially, every NLWeb instance is simultaneously a Model Context Protocol server (MCP), enabling websites to make their content discoverable for AI agents. For your website, this means that clean, complete Schema markup no longer models only for search engines but also lays the foundation for agentic queries.

Is FAQ schema still worthwhile, even though Google has restricted FAQ rich results?

That depends on the goal. Google now displays FAQ rich results only for a few authoritative sources, so the guaranteed visual presentation in classic search has largely disappeared. However, the underlying markup remains a clear extraction signal for content that directly answers a question. Schema types should therefore be selected based on their value for machine understanding, not solely on the current rich-result presentation, which Google can change at any time.

How do I measure whether my structured data are working?

On three levels. First, technical validation with the Rich Results Test, which checks whether the markup is correct and contains all required fields. Second, Google Search Console, which shows which types have been recognized, whether errors exist, and how impressions and clicks are developing. Third, AI citation tracking, meaning the monitoring of mentions in systems such as ChatGPT or Perplexity. The third level is still nascent, and establishing a clean causal link to the markup is difficult.

Are structured data a GDPR problem?

Generally not. Structured data are unproblematic as long as they only make publicly available, factual information machine-readable, such as company name, address, or product data. JSON-LD does not set cookies and is not tracking, so it does not belong behind a consent barrier. Caution is only required with personal data, for instance when a natural person is marked up with name, photo, or contact details. In that case, the usual GDPR requirements for legal basis and purpose limitation apply.

How does your website perform?

Get a free, AI-powered SEO report of your website by email – technical SEO, on-page, keywords & competitors. No obligation.

Get a free SEO audit →

Previous← AI Crawler Management: Compliance Crisis and Enterprise Strategy NextSchema Markup Implementation: Priorities for Enterprise B2B →