AI Summary
Seven schema types drive measurable AI citation gains in 2026: Organization, Person, Article, FAQPage, HowTo, Product, and Dataset. Every other type in Schema.org’s 800-plus vocabulary is secondary until these are solid.
Retrieval systems inside AI engines parse structured data before they parse prose. When your markup is absent or malformed, the engine falls back to probabilistic inference from raw text. When it is present and correct, the engine can extract entity relationships, verify authorship, and anchor facts to a source. The difference shows up directly in citation rates.
Our March 2026 study of 42,971 citations across six platforms found that the most-cited domains share one structural property: machine-readable entity signals at the page and site level. Schema markup is the primary mechanism for those signals. Before choosing a type, choose a format: the JSON-LD vs Microdata decision matters and is covered separately. This post focuses on which types to deploy and why.
The seven schema types that move the needle
These types are listed in implementation priority order, not alphabetical order.
1. Organization (with sameAs)
Organization schema is the foundation. It tells retrieval systems exactly who you are and connects your site to your entity in the knowledge graph. The sameAs property is where the leverage lives: link to your Wikipedia article, Wikidata entry, LinkedIn company page, Crunchbase, and GitHub. Every corroborating reference strengthens entity disambiguation. Without it, AI engines cannot reliably resolve which “Acme Corp” you are when multiple entities share a name.
Include name, url, logo, description, foundingDate, and at minimum four sameAs links. Place it in your site-wide <head>, not on individual posts.
2. Person (for authors and founders)
Person schema is the author-level equivalent of Organization schema. AI engines assess authorship trustworthiness before deciding whether to cite a claim. A Person entity with sameAs links to LinkedIn, a personal site, and professional profiles gives retrieval systems enough signal to attribute expertise. The knowsAbout property is underused and high-value: list your actual topical specializations as strings.
We cover the full implementation pattern in our Person schema and E-E-A-T guide. The short version: never pass author as a bare name string inside Article schema. Always link to a fully-defined Person entity via its @id.
3. Article (with author entity and dates)
Article schema tells AI engines when the content was published, when it was last updated, and who wrote it. All three signals affect citation selection. Our May 2026 study of 153,425 citations found that content recency is a strong predictor of citation selection: 74.9% of cited sentences appear in the first half of the document, and freshness signals compound that effect.
Always include datePublished, dateModified, headline, author as a Person @id reference, and publisher as an Organization @id reference. The Article vs FAQPage vs HowTo decision tree helps when a page could qualify for multiple types.
4. FAQPage
FAQPage is the highest-leverage schema type for AI citation extraction. Question-answer pairs are exactly the format retrieval systems prefer when synthesizing a response to a user query. Each answer in your FAQPage markup is a pre-formatted citation block: the engine can lift it directly and attribute it to your page.
Effective FAQ schema answers are 40-70 words: complete enough to stand alone, short enough to fit inside a synthesized AI response. Keep them declarative and fact-dense. Never add FAQ schema for questions that exist only in the markup and not on the visible page. Our detailed breakdown of when to use FAQPage versus HowTo versus Article is at FAQPage vs HowTo schema decision.
5. HowTo
HowTo schema signals step-by-step instructional content. When a user asks an AI engine “how do I…” or “what are the steps to…”, HowTo-marked pages have a structural advantage because the engine can extract steps directly from the HowToStep array rather than inferring sequence from prose. This is especially relevant for technical documentation, setup guides, and process articles.
Each HowToStep should have a short name (the step label) and a text value (the instruction). Keep step text under 80 words. Do not mark up content as HowTo if it is not genuinely procedural.
6. Product (and Review / AggregateRating)
Product schema is mandatory for ecommerce and SaaS pages where AI engines handle shopping and comparison queries. When a user asks “which tool should I use for X”, engines look for structured product data: name, description, offers with price, and aggregateRating with a real review count. Missing this markup means the engine must infer product details from prose, which produces lower-confidence citations.
We cover the full implementation for ecommerce in our ecommerce product schema guide, including how AI shopping agents parse product structured data differently from traditional rich results. Review schema from third-party platforms like G2 and Trustpilot integrates with Product schema for compounding citation effects as detailed in our review schema guide.
7. Dataset
Dataset schema is the highest-leverage type most sites are not using. AI engines treat pages with Dataset markup as primary data sources, not secondary commentary. If you publish original research, survey results, or proprietary data, Dataset schema signals that this page is the canonical source for those numbers. That is the strongest possible citation signal short of being a peer-reviewed journal.
Include name, description, creator (as Organization or Person entity), datePublished, license (use a Creative Commons URL if applicable), distribution, and variableMeasured where relevant. Our own studies use Dataset markup, and it drives direct citations from AI engines on research queries.
What to skip until the core seven are solid
Schema.org has types for recipes, events, movies, medical conditions, and hundreds of other domains. Unless your site is literally in those domains, implementing those types is a distraction. More damaging is implementing types incorrectly: invalid markup actively signals low technical quality to crawlers.
SpeakableSpecification is the one exception worth noting early. If your content targets voice interfaces and AI assistants, Speakable schema marks which text segments are best suited for audio delivery. We cover when and how in our Speakable schema guide. It belongs in the implementation queue after the core seven, not before.
Schema type comparison: what each one signals
| Schema type | Primary signal to AI | Page types | Priority |
|---|---|---|---|
| Organization | Brand entity + knowledge graph anchoring | Homepage, about | 1 - implement first |
| Person | Author expertise + E-E-A-T signal | Author pages, bios | 1 - implement first |
| Article | Freshness, authorship, publisher | All blog posts | 1 - implement first |
| FAQPage | Pre-formatted citation blocks | FAQ sections, guides | 2 - implement second |
| HowTo | Procedural structure, step extraction | Tutorials, walkthroughs | 2 - implement second |
| Product | Purchasable entity + pricing signal | Product, SaaS pages | 2 - ecommerce/SaaS only |
| Dataset | Primary data source authority | Research, reports | 2 - data publishers only |
| SpeakableSpecification | Voice delivery segments | Any page with audio target | 3 - voice-specific |
Implementation priorities and common errors
The most common structural error we see in client audits is implementing type-level schema without entity linking. An Article schema that lists author as "author": "Jane Smith" instead of "author": {"@type": "Person", "@id": "https://example.com/about/jane-smith"} gives AI engines nothing to verify. The engine cannot cross-reference that name string against any knowledge base. Always use @id references to connect entities to each other.
The second most common error is duplicate schema blocks. WordPress themes often inject their own schema while plugins add more. The result is conflicting datePublished values, mismatched author claims, and validation errors that suppress rich results entirely. Audit with Google’s Rich Results Test and the Schema Markup Validator before deploying at scale. Our GEO audit checklist includes a structured data audit section covering these issues.
Third is using Microdata or RDFa on sites where JSON-LD is feasible. Google explicitly recommends JSON-LD. It separates markup from HTML, making it maintainable and easier to validate. The only case where Microdata remains preferable is legacy CMS environments where injecting JSON-LD into <head> is not possible without a plugin. The full comparison is in our JSON-LD vs Microdata guide.
sameAs: the thread connecting your entities
The sameAs property deserves its own section because it does the disambiguation work that all other schema types depend on. sameAs links tell AI engines: this Organization or Person in our markup is the same entity as this Wikipedia article, this Wikidata item, this LinkedIn profile. Without those cross-references, schema markup is local information. With them, it connects to the global knowledge graph that AI engines query constantly.
Minimum sameAs for an Organization: Wikipedia (if you have an article), Wikidata, LinkedIn company page, Crunchbase, and your primary industry directory listing. Minimum sameAs for a Person: LinkedIn profile, personal domain, GitHub (for technical authors), and any professional association listing. More corroborating references mean stronger entity resolution. We track entity resolution scores as part of our GEO audit service and consistently find that sameAs coverage is the single largest gap in client schema implementations.
For a deeper treatment of entity optimization, including knowledge graph entity authority and brand entity disambiguation, those posts cover the mechanisms that schema markup feeds into.
Measuring schema impact on AI citations
Schema changes take time to propagate. Google re-crawls and re-processes structured data on its own schedule, not yours. Allow four to eight weeks before attributing citation changes to schema updates. Track changes with manual AI engine queries targeting your brand and your key topic clusters. Log citation rates before and after changes using a consistent query set.
The open-source GEO/AEO Tracker (github.com/danishashko/geo-aeo-tracker) handles this automatically: it tracks citation presence across six AI models on a scheduled basis and surfaces the delta between runs. The AEO Audit tab specifically checks for missing or malformed schema as part of its site readiness report. More hands-off tracking options, including commercial tools, are compared in our AI search tracking tools guide.
Schema markup is not a one-time implementation. AI engines update their parsing logic. Schema.org adds new types and deprecates old properties. Your content changes. Build a quarterly schema audit into your technical SEO workflow, and keep entity sameAs links current as your brand gains new directory listings and press mentions. The compounding effect of correct schema over time is what separates consistently-cited sites from those that appear occasionally and disappear again.