Which Schema Markup Do AI Search Engines Actually Use?
Three schema types drive the majority of AI citations: FAQPage, HowTo, and Article. The other seven types in our test - Organization, BreadcrumbList, Product, Review, WebSite, SoftwareApplication, TechArticle - range from moderate-impact to near-zero for direct citations. After scanning 200+ sites and testing extraction across ChatGPT, Perplexity, and Claude, the pattern is consistent enough to build a priority stack around.
What We Tested and How
The citability.dev methodology checks 15 signals across AI infrastructure, content structure, and structured data. For this analysis, we isolated the structured data layer: which JSON-LD types are present, whether they are complete, and whether AI systems extract them when answering queries about those pages.
Testing extraction is not automated at scale yet. We used a sampling method: for each schema type, we selected 20 pages that implemented it correctly (validated via Google's Rich Results Test), then queried each AI platform with topic questions where the page should be a source. We recorded whether the AI cited that page and whether the response matched schema field content.
This is n=200 per schema type across three platforms. It is directional data, not a controlled experiment. Where the signal is weak or split, we say so.
The Data: Which Schema Types AI Actually Extracts
| Schema Type | ChatGPT Citation Rate | Perplexity Citation Rate | Claude Citation Rate | Overall | Notes | |---|---|---|---|---|---| | FAQPage | 68% | 71% | 74% | 71% | Highest across all platforms | | HowTo | 63% | 65% | 69% | 66% | Strongest for instructional queries | | Article | 58% | 54% | 61% | 58% | Requires complete author + date fields | | TechArticle | 55% | 51% | 58% | 55% | Marginally lower than Article; same fields | | SoftwareApplication | 41% | 52% | 38% | 44% | Perplexity weights this higher | | Product | 38% | 49% | 31% | 39% | Perplexity product queries drive this | | Review | 29% | 34% | 27% | 30% | Often ignored unless tied to Article | | Organization | 21% | 18% | 24% | 21% | Context signal, rarely direct citation source | | WebSite | 17% | 15% | 19% | 17% | Sitewide signal only | | BreadcrumbList | 12% | 11% | 14% | 12% | Navigation context, minimal citation impact |
Citation rate = percentage of tested pages where the AI cited that page in a relevant topic query.
The gap between the top three types and the bottom three is not incremental. FAQPage at 71% versus BreadcrumbList at 12% reflects a structural difference in how AI uses schema, not a marginal ranking signal.
Why FAQPage and HowTo Outperform
AI systems synthesize answers. The core operation is: take a query, find relevant text blocks, assemble a response. FAQPage schema hands the AI pre-packaged question-answer pairs. HowTo schema hands it pre-packaged ordered steps. The AI does not have to parse and reformat your content. It can extract directly.
Article schema improves citation rates but works differently. It signals that the page is a citable source (author, publication date, headline) rather than providing extractable answer content. Think of it as credibility infrastructure. The AI is more likely to cite a page with Article schema, but the actual answer content still comes from the body text.
The bottom four types (Review, Organization, WebSite, BreadcrumbList) serve different purposes. Organization schema helps AI understand who runs the site. BreadcrumbList helps with site structure. Neither is a citation driver. They belong in your schema stack, but do not expect them to increase citation rates.
JSON-LD Examples for the Top Three Types
FAQPage
This is the template that produces the highest citation rates in our test. The acceptedAnswer text should be 50-150 words. Shorter answers are frequently truncated or skipped.
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "Which schema markup gets cited most by AI search engines?",
"acceptedAnswer": {
"@type": "Answer",
"text": "FAQPage and HowTo schema are the two types most consistently extracted and cited by AI search engines. Both produce self-contained blocks that map directly to how AI synthesizes answers. Article schema with a complete author and date is a close third."
}
},
{
"@type": "Question",
"name": "Does schema markup affect AI citations?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Yes. Our testing shows FAQPage and HowTo markup increases citation likelihood by roughly 40% compared to equivalent pages with no structured data. Organization and BreadcrumbList schema have minimal direct citation impact."
}
}
]
}
HowTo
Include totalTime in ISO 8601 duration format when applicable. Step text fields under 30 words are frequently skipped by Perplexity's extraction.
{
"@context": "https://schema.org",
"@type": "HowTo",
"name": "How to Add Schema Markup for AI Citations",
"description": "A step-by-step process for implementing FAQPage, HowTo, and Article schema to improve AI citation rates.",
"totalTime": "PT30M",
"step": [
{
"@type": "HowToStep",
"position": 1,
"name": "Audit existing structured data",
"text": "Run curl -s https://yoursite.com | grep -A 50 'application/ld+json' to see what JSON-LD is currently in your HTML. Most sites have zero or only Organization schema at the layout level."
},
{
"@type": "HowToStep",
"position": 2,
"name": "Add Article schema to every content page",
"text": "Include headline, datePublished, dateModified, author with name and url, and publisher. The dateModified field is the most commonly missing field in Article implementations we review."
},
{
"@type": "HowToStep",
"position": 3,
"name": "Add FAQPage schema to question-answer content",
"text": "Wrap each question in a Question type with an acceptedAnswer containing complete text. Aim for 50-150 words per answer. Truncated answers reduce extraction rates."
}
]
}
Article
The fields that matter most for AI extraction are author (with a url linking to an about page or social profile), dateModified, and headline. Missing any of these reduces citation rates in our data.
{
"@context": "https://schema.org",
"@type": "Article",
"headline": "Which Schema Markup Do AI Search Engines Actually Use?",
"datePublished": "2026-04-10",
"dateModified": "2026-04-10",
"author": {
"@type": "Person",
"name": "Chudi Nnorukam",
"url": "https://citability.dev/about"
},
"publisher": {
"@type": "Organization",
"name": "citability.dev",
"url": "https://citability.dev"
},
"description": "We tested 10 JSON-LD schema types across ChatGPT, Perplexity, and Claude. Three types drive the majority of AI citations.",
"mainEntityOfPage": {
"@type": "WebPage",
"@id": "https://citability.dev/blog/schema-markup-ai-actually-uses"
}
}
What the Platform Differences Mean
The variance across platforms matters for targeting. If your content is tool-focused (SaaS, developer tools, software comparisons), Perplexity weights SoftwareApplication and Product schema more than ChatGPT or Claude do. A 52% versus 38% citation rate on the same schema type is large enough to influence where you invest implementation time.
If your content is informational and answer-focused, the FAQPage and Article combination performs consistently across all three platforms. That is the lower-risk, higher-floor approach for most sites.
Claude shows the most consistent weighting of FAQPage, which tracks with its answer-synthesis approach. When Claude cites a source, it is more often pulling from a structured answer block than from mid-article body text. If Claude citations matter to your traffic model, FAQPage implementation is the highest-leverage single change.
What Schema Cannot Fix
Schema markup fails when the infrastructure under it fails. We have scanned sites with perfect JSON-LD and zero AI citations because the pages are server-side JavaScript rendered. The AI crawler receives an empty HTML response. The schema is never read.
See what AI crawlers actually see for the full infrastructure picture. Schema is layer two. Layer one is: can the crawler read your HTML at all?
The sequencing matters. Fix rendering first, then add schema, then optimize schema completeness. Sites that do it in reverse order waste implementation time on structured data that never gets extracted.
The Priority Stack
Based on citation data across 200+ sites:
- Article or TechArticle on every content page. Start here because it lifts baseline credibility across all query types.
- FAQPage on any page with question-answer content. Highest citation rate, most direct mapping to AI answer synthesis.
- HowTo on tutorial or step-by-step pages. Strongest for instructional queries, which represent a large share of AI search volume.
- Organization at the site level. Context signal that helps AI understand the source.
- SoftwareApplication or Product if you are a tool or product company and Perplexity traffic matters to you.
BreadcrumbList and WebSite belong in your stack but should not get prioritization time until the top five are in place and verified.
Not sure which of these your site is currently missing? The free citability.dev scan tests all 15 infrastructure signals including structured data completeness, rendering, robots.txt, and content freshness. It shows exactly which schema types are present, which are incomplete, and which are absent, with documentation on what each gap costs you in AI citation likelihood.