Schema Markup for AI Search: Which Types Actually Get You Cited by LLMs

The promise is seductive: add the right JSON-LD to your pages and watch ChatGPT, Perplexity, and Google's AI Overviews start citing your content. Vendor blogs claim schema markup can triple your AI citations. Agencies push it as the single most important GEO lever. But when you line up the actual evidence-controlled experiments, cross-platform studies, confirmed platform statements-the picture is far more nuanced than the marketing suggests.

A December 2024 study from Search/Atlas found no correlation between schema markup coverage and citation rates. Sites with comprehensive schema didn't consistently outperform sites with minimal or no schema markup. Meanwhile, a 2026 empirical study of 730 AI citations across ChatGPT and Gemini found that generic schema (Article, Organization, BreadcrumbList) provides zero measurable citation advantage-only attribute-rich schema (Product and Review types with populated pricing, ratings, and specifications) showed a significant effect, cited at 61.7% versus 41.6% for generic implementations.

That gap-between what the industry repeats and what the data shows-is where this article lives. If you're going to invest engineering time in structured data for AI search, you need to know which schema types actually move the needle, which are dead weight, and which can actively hurt you.

How AI Systems Actually Process Schema Markup

Before choosing schema types, you need to understand the mechanism. Most content about schema and AI search skips this step, and that creates dangerous assumptions.

SearchVIU conducted comprehensive tests in October 2025 and found something critical: current AI chatbots do NOT use JSON-LD schema markup during direct retrieval. Instead, they exclusively extract visible HTML content. When researchers placed product pricing exclusively inside JSON-LD with no visible HTML counterpart, zero out of five systems extracted it.

So how does schema actually help? AI models do not semantically parse JSON-LD-they treat it as text during retrieval. The real mechanism is indirect: schema enriches search engine indexes (particularly Bing's), which then feed AI response generation.

Think of it this way. Google's Knowledge Graph and Bing's index both ingest your structured data. When ChatGPT uses Bing's search backend or Google AI Overviews pulls from its own index, the enriched understanding of your entities and content relationships influences which pages get selected and cited. Google uses structured data "to understand the content of the page," and schema feeds the Knowledge Graph, which informs AI response selection-an indirect but real benefit.

What the Platforms Have Actually Confirmed

Confirmed statements from platform representatives matter more than speculative blog posts. Here's what we know:

• Microsoft/Bing:

Microsoft's Fabrice Canel confirmed at SMX Munich in March 2025 that "Schema markup helps Microsoft's LLMs understand content." This is an official statement from a Principal Product Manager at one of the major AI platforms. Since ChatGPT and Copilot both use Bing's index, this is directly relevant. - Google: In April 2025, the Google Search team said that structured data gives an advantage in search results. However, Google has explicitly stated: "There's no special schema.org structured data that you need to add" for AI features.

• OpenAI/Perplexity/Anthropic:

OpenAI, Anthropic, Perplexity, and other platforms besides Microsoft or Google haven't published their indexing methods.

The takeaway: schema matters most where platforms have confirmed it-Bing's ecosystem and Google's Knowledge Graph. For other platforms, the benefit is indirect and unverified.

The Schema Types That Actually Drive AI Citations

Not all schema is equal. The difference between schema that helps and schema that hurts comes down to attribute richness-not schema type. But within that frame, certain types consistently show up in the evidence.

FAQPage: The Highest-ROI Type for GEO

Every dataset points in the same direction. FAQPage schema pages are 3.2× more likely to appear in Google AI Overviews-the highest citation multiplier of any schema type tested. The reason is structural: AI systems are fundamentally answering questions. When your FAQ schema explicitly labels question-answer pairs, you eliminate ambiguity in extraction.

If you already rank in Google's top 10 for a keyword, adding FAQ schema increases your probability of appearing in AI Overviews for that query by approximately 40%-giving you dual visibility: traditional blue link AND AI-generated citation.

Implementation specifics matter. Structure questions as H3 headings in your visible content, matching the "name" property in your FAQ schema exactly. This consistency helps AI platforms verify the relationship between markup and content. Keep answers between 50–150 words. Pack them with specific data points rather than vague statements. If your FAQ schema contains questions and answers that don't appear in the visible page content, Google can penalize you and AI systems won't trust the markup. Schema must reflect what users actually see.

Note: FAQ rich results remain restricted to authoritative government and health websites. But that restriction applies to rich results, not to AI understanding. FAQ schema still serves its GEO purpose even without the visual SERP feature.

Product and Review Schema: Where Attribute Richness Pays Off

This is where the Growth Marshal study delivers its sharpest insight. Attribute-rich Product and Review schema with populated pricing, ratings, and specifications was cited at 61.7% versus 41.6% for generic implementations. That 20-percentage-point gap represents real commercial value.

The advantage is most pronounced for lower-authority domains (DR ≤ 60): attribute-rich schema achieves a 54.2% citation rate versus 31.8% for generic-a meaningful lift. Among high-authority domains (DR > 75), the gap narrows considerably because domain authority dominates.

For e-commerce and SaaS sites, this has clear strategic implications. Fill every relevant property: price, priceCurrency, aggregateRating, reviewCount, brand, sku, availability. Don't stop at the required fields. AI engines interpret incomplete schema as a mismatch between what you claim and what you deliver. A Product schema with only "name" filled in is a false signal.

Article and BlogPosting: The Authority Backbone

Article and BlogPosting schema clarify key content attributes like publication date, author, and topic. These signals reinforce your content's authority and freshness, which can influence how LLMs surface your content in responses.

The dateModified field deserves special attention. AI systems have a strong recency bias, with 95% of ChatGPT citations coming from content updated within 10 months. A current dateModified value signals that your content is maintained and trustworthy.

Essential properties beyond the basics: author (linked to a Person schema via @id), publisher (linked to Organization), about (declaring main topics as entities), and mainEntityOfPage. The interconnections matter-an Article that references its author and publisher creates a verifiable entity chain.

Organization and Person: The Entity Foundation

Organization schema anchors your brand identity. It tells LLMs who you are, where you're located, and how to describe you. Without it, AI systems have to piece together your identity from scattered mentions across the web.

Person schema defines structured details about an author or contributor, helping LLMs connect your content to real individuals and understand authority. This strengthens E-E-A-T signals.

The sameAs property is underrated. Linking your Organization and Person schemas to Wikipedia, Wikidata, LinkedIn, and other authoritative platforms creates entity disambiguation that AI systems use for validation. Sites with comprehensive schema implementation report higher citation rates when schema includes strong entity validation through sameAs properties linking to Wikipedia, Wikidata, and other authoritative sources.

The Generic Schema Trap: When Structured Data Hurts You

Here's the finding that should reshape how agencies sell schema implementation: The most common schema mistake is implementing it at all without populating attributes-generic schema produces an 18-percentage-point citation penalty compared with having no schema.

That's not a marginal difference. The Growth Marshal study confirmed this at scale: generic, minimally populated schema actually underperforms having no schema at all (41.6% vs 59.8%). The CMS-default schema your WordPress theme ships with? If you haven't enriched it, it's working against you. Why does this happen? The LLM feedback loop reinforces the untested schema consensus, driving continued implementation of generic schema that the data does not support. AI systems appear to interpret sparse, templated schema as a signal that the site hasn't invested in quality signals-essentially the opposite of what you intended.

Five Implementation Errors That Undermine AI Visibility

Beyond the generic schema problem, five implementation errors consistently undermine AI visibility: schema with only required fields, broad types instead of specific subtypes, and no sameAs, about, or knowsAbout properties.

Additional critical mistakes:

• Mismatched markup and visible content:

If schema markup inaccurately reflects the visible content on a page, it can mislead users and violate Google's guidelines. Common mistakes include applying markup to hidden content, using irrelevant schema types, or overloading pages with excessive markup.

• Stale dateModified values:

A page with Article schema but a stale dateModified value signals to AI systems that the content is outdated. Update this field every time you revise the content-and actually revise the content when you update the date.

• Auto-generated schema without review:

Automated tools often produce valid but incomplete markup. They'll generate an Article with a headline and author but skip dateModified, description, and publisher.

Google's Schema Deprecations: What They Mean for AI Strategy

In June 2025, Google deprecated seven structured data types: CourseInfo, ClaimReview, EstimatedSalary, LearningVideo, SpecialAnnouncement, VehicleListing, and Book Actions. PracticeProblem was deprecated from January 2026.

FAQ rich results remain restricted to authoritative government and health websites. HowTo rich results have been deprecated for both desktop and mobile.

Some SEOs interpreted this as Google moving away from structured data. Google is not killing schema-it's refining which types matter most. John Mueller confirmed that structured data "comes and goes," but core elements remain critical.

Core types-Product, Organization, Article, Review, BreadcrumbList-remain fully supported and relevant for AI. The deprecations targeted low-adoption, feature-specific markup. The message: focus schema investment on entity-defining types, not on chasing specific SERP features.

The Right Way to Prioritize: A Practitioner's Framework

76% of AI Overview sources come from the organic top 10 (as of mid-2025). If a page already ranks well in traditional search, adding schema to it has the highest probability of generating AI citations. That data has shifted- Ahrefs research published in March 2026 found that only approximately 38% of AI Overview citations come from pages that rank in the top 10 organic results, with the remaining split between positions 11–100 and beyond position 100. But the principle holds: start with your best-performing content.

Tier 1: Implement Immediately

Organization schema on your homepage and about page. Populate name, url, logo, sameAs (every verified profile), foundingDate, numberOfEmployees, and contactPoint. This is your entity anchor. Article/BlogPosting schema on every content page. Include headline, author (linked Person), publisher (linked Organization), datePublished, dateModified, about, mainEntityOfPage, and image. Use @id references to connect entities across pages. FAQPage schema on any page with Q&A content. Mirror questions and answers exactly between visible HTML and markup. Keep answers data-rich and self-contained.

Tier 2: Add After Tier 1 Is Fully Populated

Product/Service schema on commercial pages-only if you can populate pricing, availability, ratings, and specifications completely. Half-filled Product schema is worse than none. Person schema for every author with alumniOf, jobTitle, worksFor, sameAs, and relevant credential properties. This feeds the E-E-A-T signals that both Google and Bing use for source evaluation. Review/AggregateRating schema wherever you display genuine, first-party reviews. This is the attribute-rich schema type the Growth Marshal data identified as having the strongest citation effect.

Tier 3: Advanced Entity Linking

Connect schemas using @id and @graph structures. For AI search, the more useful pattern is to connect nodes into a coherent graph using @id: an Organization node with a stable @id, a Person node for the author who works for your organization, and an Article node authoredBy that person and publishedBy that organization, with about properties that declare the main topics.

Use the @id property to link related entities across your site. For example, connect an Article's author property to a Person schema on their bio page, and link that Person back to your Organization schema. This builds a strong knowledge graph that AI models trust.

How to Measure Whether Your Schema Is Working

Until recently, measuring AI citation impact was mostly guesswork. That's changing.

The Bing AI Performance Dashboard, launched on February 10, 2026, is a free tool inside Bing Webmaster Tools. It shows total citations, average cited pages, grounding queries (the user questions that triggered AI to cite your content), and page-level citation activity.

This tool is a breakthrough, though it has limits. It doesn't show whether users actually clicked through from those citations, and it only covers Bing and Copilot, which represents a fraction of AI search traffic-87.4% of AI referrals reportedly come from ChatGPT.

Otterly.AI's early analysis found one site was used by AI 44,469 times while being visibly cited only 169 times. That ratio-99.6% of AI usage invisible-means citation counting alone understates schema's potential impact. For a complete measurement stack:

• Bing Webmaster Tools AI Performance: Free citation tracking for Copilot ecosystem
• Google Search Console: Monitor for changes in impressions that correlate with schema updates (AI Overview data appears under "Web" search type)
• Manual prompt testing: Query ChatGPT, Perplexity, and Gemini monthly with questions your buyers would ask. Document which sources get cited.
• Schema validation:

Use Google Rich Results Test before publishing, Schema Markup Validator for comprehensive checks, and Google Search Console Enhancements for ongoing error tracking.

The Honest Assessment: Schema as Infrastructure, Not Silver Bullet

Schema markup is infrastructure, not a magic bullet. It won't necessarily get you cited more, but it's one of the few things you can control that platforms such as Bing and Google AI Overviews explicitly use.

The evidence supports a clear conclusion. Schema markup contributes to AI citations indirectly-through index enrichment, entity disambiguation, and extraction accuracy. It does not function as a ranking factor or a citation trigger on its own. Schema is an enabler for machine comprehension and feature eligibility. It's not a ranking factor; quality, authority, and freshness still drive outcomes.

Organic rank position is the dominant predictor of AI citation (OR = 0.762). Content quality, topical authority, and freshness carry more weight than any structured data implementation. Schema amplifies those signals. It doesn't replace them. Where schema earns its keep is in reducing ambiguity. When multiple pages compete for citation on the same query, the page whose entities, authorship, and data points are machine-verified through rich schema has an edge. For lower-authority domains especially, that edge is measurable and meaningful. The practitioners who treat schema as one layer of a comprehensive GEO strategy-alongside answer-first content structure, genuine expertise, and cross-platform brand authority-are the ones who will compound their visibility as AI search matures.