how-to 11 min read May 03, 2026

Schema Markup for AI Search: The 8 Types That Drive Citations

By Peter Foy

The 8 schema types that drive AI citations, with copy-paste JSON-LD, citation-lift data, and the Search Atlas counter-study most guides ignore.

TL;DR

Eight schema types drive most AI citations: FAQPage, HowTo, Article, Organization, Person, Product, Review, and ItemList. Stackmatix reports a 2.5x citation lift for schema-enabled pages. Search Atlas's counter-study found no correlation. The truth sits between: schema helps Google AI Overviews and entity grounding more than raw LLM retrieval. Combine FAQPage + Article + HowTo via @graph for the highest measurable lift.

FAQPage schema correlates with a 41% citation rate vs 15% without (Stackmatix, 2026)
Combine multiple schema types in a single @graph block, not separate scripts
Organization + sameAs to Wikidata is the single highest-leverage move for ChatGPT brand answers
Search Atlas found zero correlation between schema density and LLM citations -- semantics still beat structure
Google deprecated HowTo rich results in Feb 2026, but the schema still feeds AI engines

Schema markup for AI search is the JSON-LD structured data that lets ChatGPT, Perplexity, and Google AI Overviews extract facts from your page without parsing prose. Eight types drive the bulk of measurable citation lift: FAQPage, HowTo, Article, Organization, Person, Product, Review, and ItemList. Stackmatix's 2026 analysis reports a 2.5x citation lift for schema-enabled pages, while Search Atlas's LLM Citation Analysis found no correlation. Both findings are useful. This guide gives you the side-by-side data, ready-to-paste JSON-LD, and a reality check on where schema does not move the needle.

Which schema types correlate with AI citations?

Eight schema types account for the bulk of measurable AI citation lift: FAQPage, HowTo, Article, Organization, Person, Product, Review, and ItemList. The strongest correlation comes from FAQPage on question-format queries and Organization for brand entity recognition.

The data is contested. Stackmatix's 2026 Structured Data AI Search guide reports schema-enabled pages are 2.5x more likely to appear in AI answers, while Search Engine Land's controlled experiment found a well-implemented schema page was the only one of three near-identical pages to appear in an AI Overview.

The counter-evidence: Search Atlas's December 2024 LLM Citation Analysis found visibility distributions nearly identical across OpenAI, Gemini, and Perplexity regardless of schema coverage.

Here is the side-by-side breakdown.

Schema Type	Primary Use	Reported Citation Lift	Strongest On
FAQPage	Q&A pairs	2.7x (41% vs 15%)	Google AIO, Perplexity
HowTo	Step-by-step	2.5x in combo	ChatGPT, Perplexity
Article	Posts, guides	Foundational	All engines
Organization	Brand entity	Entity grounding	ChatGPT brand answers
Person	Author EEAT	Indirect lift	All engines
Product	Ecommerce	3.1x in AIO	Google AIO
Review	Ratings	Boosts Product	Google AIO, Perplexity
ItemList	Listicles	Pairs with 74.2% format lift	ChatGPT, Perplexity

AI Citation Lift by Schema Type (vs No Schema)

FAQPage (question-format queries)

67%

FAQPage (general)

41%

Article + FAQPage + HowTo + Org combo

2.7%

Generic structured data

3.1%

Organization (entity grounding)

2.5%

Schema-enabled vs no schema (Search/Atlas)

Source: Stackmatix 2026, Search Engine Land controlled experiment, Search Atlas LLM Citation Analysis

Does FAQPage schema actually increase Google AI Overview citations?

Yes, on question-format queries. Pages with FAQPage schema show a 41% citation rate versus 15% without it -- roughly 2.7x more visibility -- according to Stackmatix's structured data analysis. ZipTie.dev's testing measured a 67% citation rate on relevant question-format prompts.

Why it works: FAQPage schema gives AI engines pre-parsed question-answer pairs. The retrieval model does not have to chunk your prose, infer the question, and extract a quote. It reads the structure directly.

The minimum viable FAQPage JSON-LD:

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "What is schema markup for AI search?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "Schema markup for AI search is JSON-LD structured data that helps engines like ChatGPT, Perplexity, and Google AI Overviews extract facts from your page without parsing prose."
    }
  }]
}

Keep answers under 60 words each. Match the question text to real prompts (use AlsoAsked or People Also Ask). Validate with Google's Rich Results Test.

FAQPage Schema: Citation Rate With vs Without

With FAQPage schema

41%

Without FAQPage schema

15%

Source: Stackmatix Structured Data AI Search Guide, 2026

What is the right way to combine Article + FAQPage + HowTo on one page?

Use a single JSON-LD script with @graph -- not three separate script tags. AI engines parse @graph cleanly and Google's Rich Results Test validates the combination as long as each schema's required properties are present.

The @graph pattern lets you cross-reference entities by @id, which signals to LLMs that the Article, the HowTo inside it, and the FAQ at the bottom belong to one canonical document with one author and one publication date.

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Article",
      "@id": "https://growthengineer.ai/post#article",
      "headline": "Schema Markup for AI Search",
      "author": {"@id": "https://growthengineer.ai/#peter"},
      "datePublished": "2026-05-03",
      "dateModified": "2026-05-03"
    },
    {
      "@type": "HowTo",
      "@id": "https://growthengineer.ai/post#howto",
      "name": "How to combine schema types",
      "step": [{"@type": "HowToStep", "text": "Wrap all schema in one @graph array."}]
    },
    {
      "@type": "FAQPage",
      "@id": "https://growthengineer.ai/post#faq",
      "mainEntity": []
    },
    {
      "@type": "Person",
      "@id": "https://growthengineer.ai/#peter",
      "name": "Peter Foy"
    }
  ]
}

Three rules:

One @id per entity, reused across the graph. This is what lets retrieval models connect the Article to its author.
Do not duplicate properties. If the Article has datePublished, the HowTo does not need it.
Validate as a unit. Schema.org's validator accepts @graph natively; some older parsers do not.

Why did the Search Atlas study find no schema-citation correlation?

Search Atlas's LLM Citation Analysis compared schema coverage to citation frequency across OpenAI, Gemini, and Perplexity and found visibility distributions "nearly identical" regardless of schema density. The interpretation: LLMs retrieve on semantic relevance and embedding similarity, not on structured data.

The study has caveats Search Atlas calls out themselves:

It measured correlation, not causation. Pages with comprehensive schema also tend to be from larger, older domains. The schema effect may be confounded with domain authority.
It looked at raw LLM retrieval, not Google AI Overviews specifically. AIO is the platform where schema effects are strongest. Excluding it removes the strongest signal.
It analyzed schema density, not schema quality. A page with poorly implemented schema scores the same as one with surgical FAQPage matching real queries.

The takeaway is not "schema is dead." It is: schema helps Google's pipeline (Index, Knowledge Graph, AI Overviews, AI Mode) more than it helps raw LLM retrieval at OpenAI or Anthropic. If your traffic comes from ChatGPT browse-mode or Perplexity, schema still helps via citation extraction. If you're hoping to influence what GPT-5 "knows" pre-training, structured data is not the lever.

How do you implement Article schema for AI search?

Article schema signals author, publication date, and freshness -- the three properties AI engines weight most for content trust. It is the foundational schema every blog post, news article, and guide needs.

Google's Article documentation lists author, datePublished, and dateModified as the high-priority fields. Reference your Person and Organization blocks via @id rather than inlining them.

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Schema Markup for AI Search",
  "image": ["https://growthengineer.ai/og.png"],
  "datePublished": "2026-05-03T08:00:00-05:00",
  "dateModified": "2026-05-03T08:00:00-05:00",
  "author": {
    "@type": "Person",
    "name": "Peter Foy",
    "url": "https://growthengineer.ai/about"
  },
  "publisher": {
    "@type": "Organization",
    "name": "Growth Engineer",
    "logo": {"@type": "ImageObject", "url": "https://growthengineer.ai/logo.png"}
  }
}

Update dateModified whenever you refresh the article. 50% of AI citations come from content less than 13 weeks old, so dateModified is your freshness lever.

What Organization schema does a B2B SaaS homepage need for ChatGPT?

Organization schema with a complete sameAs array is the single highest-leverage move for ChatGPT brand answers. ChatGPT's entity grounding pulls from Wikidata, Crunchbase, LinkedIn, and your own structured data. sameAs is the property that tells AI engines those profiles are the same entity as your homepage.

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "@id": "https://growthengineer.ai/#org",
  "name": "Growth Engineer",
  "url": "https://growthengineer.ai",
  "logo": "https://growthengineer.ai/logo.png",
  "description": "AEO and GEO software for B2B teams.",
  "foundingDate": "2025",
  "founder": {"@id": "https://growthengineer.ai/#peter"},
  "sameAs": [
    "https://www.linkedin.com/company/growthengineer",
    "https://www.crunchbase.com/organization/growthengineer",
    "https://www.wikidata.org/wiki/Q-PLACEHOLDER",
    "https://github.com/growthengineer",
    "https://x.com/growthengineer"
  ]
}

Checkpoints:

Get a Wikidata entry. The single highest-impact off-site move. ChatGPT explicitly grounds entities via Wikidata.
Pair with WebSite schema including a SearchAction for the homepage.
Reuse the @id in every Article, FAQ, and Person block on your site so the entity graph is one connected mesh, not 200 disconnected fragments.

How do you use Person, Product, Review, and ItemList schema?

These four close the loop on author trust, ecommerce, social proof, and listicle structure. Each gets a copy-paste snippet below.

Person schema validates author EEAT and connects bios to their published work. Use on author pages and link via Article author.

{
  "@context": "https://schema.org",
  "@type": "Person",
  "@id": "https://growthengineer.ai/#peter",
  "name": "Peter Foy",
  "jobTitle": "Head of Growth",
  "worksFor": {"@id": "https://growthengineer.ai/#org"},
  "knowsAbout": ["AEO", "GEO", "AI search"],
  "sameAs": ["https://www.linkedin.com/in/peterfoy"]
}

Product schema drove a 3.1x citation lift in AI Overviews per ALM Corp's ecommerce analysis. Required fields: name, image, offers (price + availability), brand, GTIN.

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Growth Engineer Pro",
  "image": "https://growthengineer.ai/product.png",
  "brand": {"@id": "https://growthengineer.ai/#org"},
  "offers": {
    "@type": "Offer",
    "price": "99.00",
    "priceCurrency": "USD",
    "availability": "https://schema.org/InStock"
  }
}

Review plugs into Product or any reviewable entity. Use aggregateRating for star averages.

{
  "@type": "Review",
  "reviewRating": {"@type": "Rating", "ratingValue": "5"},
  "author": {"@type": "Person", "name": "Jane Smith"},
  "reviewBody": "Cut our AEO research time by 80%."
}

ItemList wraps listicles. Listicles account for 74.2% of all AI citations, so this is the highest-leverage schema for content marketers.

{
  "@context": "https://schema.org",
  "@type": "ItemList",
  "itemListElement": [
    {"@type": "ListItem", "position": 1, "name": "FAQPage", "url": "#faqpage"},
    {"@type": "ListItem", "position": 2, "name": "HowTo", "url": "#howto"},
    {"@type": "ListItem", "position": 3, "name": "Article", "url": "#article"}
  ]
}

Is HowTo schema still worth implementing after Google deprecated it?

Yes. Google deprecated HowTo rich results in February 2026, but the schema still feeds AI engines. ChatGPT and Perplexity continue extracting step-by-step content from HowTo markup at high rates because procedural answers map directly to user prompts like "how do I configure X."

The deprecation removed the visual rich-snippet treatment in classic SERPs. It did not remove the parsing pathway into Google's Index, Knowledge Graph, or AI Overviews. Stackmatix's analysis lists HowTo as one of the four schema types that delivers a 2.5 to 2.7x AI Overview citation lift in combination with Article, FAQPage, and Organization.

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "How to validate JSON-LD",
  "totalTime": "PT5M",
  "step": [
    {"@type": "HowToStep", "position": 1, "name": "Paste URL", "text": "Paste your page URL into the Rich Results Test."},
    {"@type": "HowToStep", "position": 2, "name": "Inspect output", "text": "Confirm all schema types render without errors."},
    {"@type": "HowToStep", "position": 3, "name": "Cross-validate", "text": "Run the same URL through Schema.org's validator for spec compliance."}
  ]
}

Use HowTo only when the page genuinely teaches a process. Do not retrofit it onto opinion pieces or comparisons -- AI engines flag mismatches and discount the page.

When does schema NOT move the needle?

Schema is a parsing layer, not a magic citation machine. It does not move the needle in five specific scenarios.

Thin or duplicative content. If the page has nothing original, schema cannot rescue it. AI engines extract facts; they do not extract opinions from pages that have none.
Pre-training citations from frontier models. Schema does not influence what GPT-5, Claude, or Gemini "know" from training. That depends on inclusion in Common Crawl, Reddit, Wikipedia, and the model's training mix.
Brand-new domains with no entity grounding. Search Atlas's findings hit hardest here: without external sameAs anchors (Wikidata, Crunchbase, LinkedIn), schema alone does not establish entity identity.
Mismatched schema. Tagging a comparison post as HowTo, or wrapping testimonials as FAQPage, gets penalized by Google's spam filter and discounted by AI engines that cross-check structure to content semantics.
Schema without semantic alignment. Search Engine Land's coverage notes schema improves machine parsing but does not change underlying relevance. If your content does not actually answer the query, FAQPage tags will not save you.

The honest framing: schema is a 2x to 3x multiplier on content that already deserves to be cited. It is not a substitute for primary research, original data, or expert authorship.

How do you validate and track schema for AI citations?

Run every page through three validators before publish, then track citation rates monthly. Validation catches syntax errors. Tracking catches semantic drift -- when your schema validates but the content no longer matches what AI engines are pulling.

Pre-publish validation stack:

Google Rich Results Test -- strict on Google-specific eligibility.
Schema.org Markup Validator -- strict on spec compliance, catches deprecated properties.
Bing URL Inspection -- coverage check for Bing Copilot.

Post-publish AI citation tracking:

Profound and Otterly -- monitor brand mention rates inside ChatGPT, Perplexity, Gemini, and AI Overviews.
Manual prompt testing -- weekly, ask the engine the question your page answers; record citation status.
Server logs for GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot -- presence indicates retrieval-pool inclusion.

Expect 2 to 4 weeks for AI Overviews to reflect schema changes after recrawl, and 3 to 5 business days for Perplexity. ChatGPT live-browse reflects schema almost immediately; pre-training citations only update at major model releases.

Schema Type	Primary Use	Reported Citation Lift	AI Engine Preference	Implementation Difficulty
FAQPage	Question-answer pairs on any page	2.7x (41% vs 15%)	Google AI Overviews, Perplexity, ChatGPT	Low (under 30 min)
HowTo	Step-by-step procedural content	2.5x (combo with Article)	ChatGPT, Perplexity for procedural queries	Low
Article	Blog posts, news, guides	Foundational (signals author + date)	All engines (recency + EEAT)	Low
Organization	Brand entity, homepage, footer	Entity grounding (no fixed lift)	ChatGPT brand answers, Knowledge Graph	Medium (needs sameAs)
Person	Author bios, About pages	EEAT signal (indirect lift)	All engines for expertise	Low
Product	Ecommerce product pages	3.1x in AI Overviews	Google Shopping, Perplexity Shop	Medium (GTIN, offers)
Review	User and editorial reviews	Boosts Product citations	Google AIO, Perplexity	Low
ItemList	Listicles, ranked content	Pairs with 74.2% listicle citation rate	ChatGPT, Perplexity for 'best of' queries	Low

Frequently asked questions

Does schema markup actually help AI search citations in 2026?

Most published studies say yes, but with caveats. Stackmatix reports a 2.5x citation lift and Search Engine Land's controlled experiment found schema-enabled pages were the only ones surfaced in AI Overviews. However, Search Atlas's December 2024 LLM Citation Analysis found no correlation across OpenAI, Gemini, and Perplexity. The honest answer: schema helps Google AI Overviews and entity grounding, but it's not a magic switch for raw LLM citation.

Which schema types matter most for ChatGPT and Perplexity?

FAQPage, Organization, Article, and Person schema have the highest reported impact. FAQPage gives ChatGPT and Perplexity clean question-answer pairs to extract. Organization plus sameAs links to Wikidata grounds your brand as a verifiable entity. Article with author and dateModified signals freshness. Person schema validates expertise for EEAT-aware retrieval.

Is HowTo schema still worth implementing after Google deprecated it?

Yes. Google deprecated HowTo rich results in February 2026, but the schema still parses cleanly into AI engines. ChatGPT and Perplexity continue extracting step-by-step content from HowTo markup at high rates because procedural answers map directly to user prompts. Use it when the page genuinely teaches a process.

Can you stack FAQPage, Article, and HowTo schema on the same page?

Yes, and you should. Use a single JSON-LD <script> block with an @graph array containing all three schema objects, each with a unique @id. AI engines parse the graph cleanly and Google's Rich Results Test will validate the combination as long as each schema's required properties are present.

Why did Search Atlas find no correlation between schema and AI citations?

Search Atlas analyzed schema coverage versus LLM citation frequency across OpenAI, Gemini, and Perplexity and found visibility distributions were nearly identical regardless of schema density. Their interpretation: LLMs retrieve based on semantic relevance and embedding similarity, not on parsed structured data. The study notes its own limitation: it measured correlation, not causation, with confounding variables unmeasured.

What schema does a B2B SaaS homepage need to appear in ChatGPT brand answers?

Organization schema with name, url, logo, description, and a complete sameAs array linking to LinkedIn, Crunchbase, Wikidata, GitHub, and your X/Twitter profile. Add WebSite schema with a SearchAction. If you have a founder or CEO with public expertise, include Person schema linked via founder. The sameAs array is the single highest-leverage property for ChatGPT entity recognition.

Which JSON-LD validator should I use before publishing?

Use Google's Rich Results Test for rich-snippet eligibility, Schema.org's Schema Markup Validator for spec compliance, and Bing's URL Inspection for Bing Copilot coverage. Run all three. Rich Results Test is the most strict on Google-specific requirements; Schema.org's validator catches deprecated properties Rich Results may silently accept.

How long until schema changes show up in AI citations?

AI Overviews typically reflect schema changes within 2 to 4 weeks after Google recrawls the page. ChatGPT and Perplexity index more aggressively with fresh content surfacing in 3 to 5 business days for retrievable queries. ChatGPT's training-data citations only update at major model refreshes, but live browsing reflects schema almost immediately.

After the validation section

Generate AEO-ready JSON-LD for any page