Schema markup is structured data you add to a page, written in JSON-LD, that tells search engines and AI models exactly what the page is about: who published it, what type of content it is, and how its facts relate to known entities. It does not force an AI engine to cite you, but it removes the guesswork that makes an engine skip you. In 2026 the schema that matters most for AI citations is the schema that resolves your identity and structures your answers, so an engine can lift a fact, attribute it to you, and trust it. This guide covers the nine types that do that work, the JSON-LD to use, and the honest limits of what schema can and cannot do.

Quick Answer

The nine schema types that most help AI engines cite you in 2026 are Organization, Article, FAQPage, HowTo, Product, Review with AggregateRating, Dataset, BreadcrumbList, and Person, all in JSON-LD. Schema does not directly buy a citation (a 2026 Ahrefs study found no direct uplift), but it resolves your identity and structures your answers so engines can attribute and trust what they lift. The highest -leverage move is Organization schema with a stable identifier and a complete sameAs list tying you to Wikipedia and Wikidata.

The 9 schema types that win AI citations, at a glance

Every type name below links to its official definition on schema.org, the open vocabulary that Google, Microsoft, Yahoo and Yandex jointly maintain. Implement each as JSON-LD in the page head or body.

Sources for the guidance in this article include the official Google structured data documentation and the schema.org vocabulary, with the honest limits drawn from a 2026 Ahrefs analysis of schema and AI citations.

1. Organization schema: the identity anchor

Organization is the most important type for AI citations, because it answers the question every engine asks before it trusts you: who is this. Publish one consistent legal name, a canonical URL, a logo, contact details, and a complete sameAs array that links to your Wikipedia page, your Wikidata entry, and your verified profiles on LinkedIn, Crunchbase and X. Without this, an AI system has to guess your identity, and that ambiguity lowers the confidence with which it will cite you. Give the block a stable @id and reuse that same @id across every page so the engine resolves all your content to one entity. This single type does more for citation confidence than any other on the list, because it turns a pile of pages into a recognised brand the engine can name. Add a foundingDate, a numberOfEmployees range, and a knowsAbout list of your core topics to make the entity richer still.

2. Article schema: correct attribution

Article (or BlogPosting for blog content) tells an engine that a page is editorial content, who wrote it, when it was published, and when it was last updated. That matters because AI engines attribute what they lift, and an article with clear authorship and fresh dates is easier to attribute correctly than an unmarked page. Include the author as a linked Person, an image, a headline that matches the visible H1, and an accurate dateModified. Freshness signals carry weight in AI answers, so keep dateModified honest and current when you genuinely update the piece. Article schema will not make weak content rank, but it makes strong content far more likely to be named as the source rather than paraphrased anonymously. Reference the author as a linked Person and the publisher as your Organization @id, so the whole attribution chain resolves cleanly.

3. FAQPage schema: ready-to-lift answers

FAQPage marks up a set of questions and their direct answers. Google ended the FAQ rich result in 2026 and later removed the documentation, so this type no longer earns you extra space in classic search. It still matters for AI, though, because engines use FAQPage markup to identify clean question-and-answer content they can lift straight into a conversational answer. The trick is to write each answer so it stands completely on its own, one clear question, one self-contained answer of two to four sentences, no reliance on the surrounding page. Do not stuff the section with keyword variations of the same question; a handful of genuinely distinct questions your buyers actually ask outperforms a wall of near-duplicates. Place the FAQPage block on the same URL as the visible questions, and keep answers to two to four sentences so an engine can quote one without trimming.

4. HowTo schema: step extraction

HowTo describes an ordered sequence of steps to complete a task. It maps almost perfectly to the step-by-step format AI engines use when answering how-to questions, which makes a well-marked HowTo page a natural source for those answers. Give each step a name and a short description, and keep the sequence in the markup identical to the visible steps on the page. HowTo suits genuine process content, setup guides, configuration walkthroughs, repair sequences, and it is wasted on pages that are not actually instructional. Used honestly on real procedures, it is one of the cleanest ways to become the cited answer for a task query.

5. Product schema: recommendations and comparisons

Product carries the facts an engine needs to include you in product research and comparison answers: name, description, brand, price, availability, and ratings. As shopping research shifts into ChatGPT and Gemini, complete and accurate Product markup is what lets those engines represent your product correctly rather than skipping it or quoting a stale third-party figure. Keep price and availability current, link the brand to your Organization entity, and pair it with Review and AggregateRating where you have genuine ratings. For ecommerce and software brands, this is the type that decides whether an AI shopping answer mentions you at all. Keep the offers block accurate with price, priceCurrency and availability, since a stale or missing price is a common reason an engine skips a product.

6. Review and AggregateRating: the verdict signal

Review and AggregateRating express opinions and scores in a machine-readable way. AI engines lean on structured verdicts when they assemble best-of and comparison answers, so genuine ratings marked up correctly can put you into those lists. The hard rule is authenticity: only mark up reviews and ratings that really exist and that you can stand behind, because fabricated or self-serving review markup is both a policy violation and something engines increasingly cross-check. Tie ratings to the specific product or service they describe, and never apply a site-wide rating to unrelated pages. Done honestly, this type helps you show up exactly where buyers are comparing options. Attach each review to the specific item with an itemReviewed reference, and never reuse one rating across unrelated pages.

7. Dataset schema: original data as citation bait

Dataset describes a structured collection of data and how it is organised. It matters for AI citations because original data is the single most citable thing you can publish: when you run a study and report numbers nobody else has, you become the primary source every other writer and every AI answer points to. Marking that data as a Dataset, with a clear description, measurement method, and licence, makes it easier for engines to recognise, understand and attribute. If you invest in original research, and for AI visibility you should, Dataset schema is how you make sure the credit and the citation come back to you rather than to whoever quotes your numbers next. Include the measurement method, the sample size, and a clear licence in the Dataset block so an engine can describe your data accurately when it cites it.

8. BreadcrumbList: topical context

BreadcrumbList tells an engine where a page sits within your site: which section it belongs to and what its parent topics are. That hierarchy is a quiet but real signal of topical context, helping an engine understand that a page about, say, technical SEO belongs to a broader, authoritative SEO section rather than sitting in isolation. It also reinforces the internal structure that both crawlers and readers follow. Breadcrumbs will not win a citation on their own, but as part of a coherent, well-structured site they help engines place your content in the right topical neighbourhood, which is part of how they decide who is authoritative enough to cite. Match the breadcrumb trail to your real URL structure and on-page navigation, so the hierarchy an engine reads is the one users actually see.

9. Person schema: author-level trust

Person marks up the real human behind the content: their name, role, credentials, and their own sameAs links to author profiles and, where they exist, a Wikipedia or Wikidata entry. As AI engines weigh expertise and first-hand experience more heavily, a clearly identified, genuinely qualified author is a trust signal that supports being cited. Link each article’s author to a real Person entity, and give recurring authors a proper author page that the Person schema points to. This is how you move from an anonymous brand publishing content to a set of named experts an engine can recognise, which is exactly the kind of source AI answers prefer to name. Give each author a jobTitle, an alumniOf or affiliation, and knowsAbout topics, so the expertise is machine-readable rather than implied.

The entity layer: sameAs, Wikidata and a stable @id

The thread running through the most valuable types is entity resolution: helping an engine map your pages to a single, known thing in the world. The sameAs property is the workhorse here. Use it on your Organization and Person blocks to link to your authoritative profiles, and above all to your Wikipedia and Wikidata entries. Linking your content to a Wikidata identifier, and referencing the entities you discuss through the about and mentions properties, is widely regarded as the strongest 2026 signal for being retrieved and cited in generative AI. Combine that with a stable @id reused across pages, and you give engines a clean, consistent identity to attach every citation to. This entity layer, not any single rich-result trick, is what compounds into durable AI visibility. In practice the highest-value single action for most brands is to create or claim a Wikidata entry and reference its Q-identifier from the Organization sameAs, because that one link plugs you directly into the graph these engines already trust.

What schema can and cannot do (the honest version)

It would be dishonest to promise that adding schema wins citations by itself. A widely-discussed 2026 Ahrefs study found that adding markup produced no direct uplift in citations across the AI platforms it tested. That result is not a reason to skip schema; it is a reason to understand what schema is for. Schema does not make an engine want to cite you. It makes it possible for an engine to understand and trust you once your content is genuinely worth citing. Think of it as removing friction rather than adding force. The brands that win combine both: content that is genuinely the best answer, and structured data that lets an engine resolve, attribute and trust it without guessing. Schema on thin content changes nothing; schema on excellent, well-sourced content is what turns a good page into a confidently cited source. That is the principle behind our approach to answer engine optimization, where structured data supports the content rather than standing in for it.

How each major AI engine uses structured data

The engines do not treat schema identically, so it helps to know who leans on what. Google AI Overviews and Gemini draw on the full structured-data stack that classic Google Search already understands, which means the Article, FAQPage, HowTo and Product markup you built for SEO carries straight into AI answers, and a clean Knowledge Graph presence built on Organization and sameAs is a direct advantage. ChatGPT search leans on the Bing index and values Article and FAQPage for conversational answers, plus Organization to attribute a claim to the right brand. Perplexity, which cites openly, uses FAQPage, Organization and Product markup to help decide which sources to footnote. Microsoft Copilot, running on the same Bing index as ChatGPT search, rewards the same Bing-side structured data. The practical lesson is that one well-built schema layer serves every engine at once, so you are not maintaining a different setup per platform, you are maintaining one clean, consistent identity that all of them can read.

The JSON-LD blocks you actually need

Two blocks do most of the work. First, a site-wide Organization block that establishes your identity and ties you to the public knowledge graph:

{   "@context": "https://schema.org",   "@type": "Organization",   "@id": "https://example.com/#org",   "name": "Example Company",   "url": "https://example.com/",   "logo": "https://example.com/logo.png",   "sameAs": [     "https://en.wikipedia.org/wiki/Example_Company",     "https://www.wikidata.org/wiki/Q000000",     "https://www.linkedin.com/company/example"   ] }

Second, a FAQPage block on any page with genuine question-and-answer content, so an engine can lift the answer cleanly:

{   "@context": "https://schema.org",   "@type": "FAQPage",   "mainEntity": [{     "@type": "Question",     "name": "What is answer engine optimization?",     "acceptedAnswer": {       "@type": "Answer",       "text": "Answer engine optimization is the practice of structuring content so AI answer engines can extract, trust and cite it."     }   }] }

Keep the @id in the Organization block identical on every page, and make sure each FAQ answer matches the visible text on the page word for word. Reference this Organization entity from your Article and Person blocks using the same @id, so the whole site resolves to one recognised brand rather than a set of unconnected pages.

Schema mistakes that get you ignored

A few avoidable errors quietly cancel out the benefit of structured data. The most common is markup that does not match the visible page, marking up a rating, an FAQ, or a price that a human cannot actually see, which engines treat as a trust violation. The second is duplication: two conflicting Organization blocks, or a site-wide rating stamped onto unrelated pages, which muddies the very identity you are trying to establish. The third is invalid JSON-LD, a missing bracket or a mistyped property name, which can cause the whole block to be ignored silently. The fourth is over-marking, wrapping every paragraph in HowTo or QAPage to game extraction, which reads as manipulation. The rule that prevents all four is simple: mark up only what genuinely exists on the page, keep one clean entity, and validate every template before it ships.

Beyond the nine: supporting types worth knowing

A handful of secondary types round out a strong setup. WebSite with a SearchAction helps engines understand your site-level search. LocalBusiness, a specialisation of Organization, carries the name, address, hours and geo data that matter for near-me and local AI answers, so multi-location brands should layer it on top of Organization. VideoObject describes video content, worth marking as AI answers increasingly surface video. Event covers time-bound happenings, and Course, JobPosting and Recipe serve their obvious niches. None of these replaces the core nine, but where they genuinely apply they extend the same principle: give the engine a precise, machine-readable description of exactly what a page offers, so it never has to guess.

Where schema fits in the bigger AI-visibility picture

Structured data is one layer of three, and it is the layer most teams over-index on because it feels technical and controllable. The other two matter more. The content layer decides whether you deserve to be cited at all: is your page genuinely the clearest, most complete, most trustworthy answer to the question. The entity layer decides whether an engine can figure out who you are and connect your pages into one recognised brand. Schema serves both, it structures the content so it is extractable and it carries the identity signals that build the entity, but it cannot manufacture either. The right sequence is content first, entity second, schema as the connective tissue that lets engines read both. Teams that invert that order, bolting markup onto thin content, get the Ahrefs result: no uplift. Teams that build all three together compound. That ordering is the backbone of how we run AEO and AI-search programmes, and it is why structured data is a supporting act, not the headline.

A schema priority order for a small team

If your time is limited, do these in order and ignore the rest until they are done. First, one clean Organization block sitewide with a stable @id and a full sameAs list, the highest-leverage hour you will spend. Second, a real author page for each recurring writer, marked up with Person and linked from every article. Third, Article schema on every blog and resource, with honest dateModified values. Fourth, FAQPage on the pages that genuinely answer discrete questions, with self-contained answers. Fifth, HowTo, Product, Review and Dataset only where they truly apply. That order front-loads the identity and attribution signals that move the needle for AI and leaves the niche types for later. A small team that gets the first three right will out-perform a large one that sprinkles every type everywhere with no coherent entity underneath.

Where to place your JSON-LD

Placement is simpler than it looks. Put each JSON-LD block inside a script tag with type set to application/ld+json, and it can live in either the head or the body, since engines read both. Keep the sitewide Organization block on every page, ideally injected by your theme or tag manager so it never drifts, and put page-specific blocks such as Article, FAQPage or Product on the individual pages they describe. Avoid splitting one logical entity across several conflicting blocks, and do not rely on client-side JavaScript that an engine may not execute; server-rendered JSON-LD is the safe default. On WordPress a dedicated SEO plugin will output much of this for you, but always confirm what it actually emits rather than assuming, because default settings rarely include the sameAs and @id wiring that does the real work for AI citations.

Validate every template before it ships

Invalid markup is worse than none, because an engine may ignore the whole block, so validation is not optional. Run each template through Google’s Rich Results Test and the Schema Markup Validator before it goes live, fix every error, and re-test after any redesign or template change. Check that every value in the markup matches what a human sees on the page, that there is exactly one Organization entity with a consistent @id, and that dates, prices and ratings are real and current. Build the check into your release process rather than treating it as a one-off, because a single broken template can silently strip structured data from hundreds of pages at once.

How to roll it out without breaking anything

Prioritise in order of leverage. Start with Organization and Person schema across the whole site, because identity is the foundation everything else builds on. Then add Article schema to every blog and resource, FAQPage to pages with genuine question-and-answer sections, and HowTo to real process content. Layer Product, Review and Dataset where they truly apply. Write everything in JSON-LD, keep every marked value identical to what a human sees on the page, and validate each template before it ships, because invalid or mismatched markup can do more harm than none at all. Roll out by template rather than page-by-page so the work scales, and re-check after any major redesign. Handled this way, structured data becomes a durable asset that supports both classic search and AI answers, which is how we build it into every SEO engagement and AEO programme we run. Get identity and answers structured first, and the citations follow as your content earns them, because you have removed every reason for an engine to overlook or misattribute you. Structured data is the quiet groundwork that makes everything else you publish easier for AI to trust.

Frequently asked questions

Does schema markup directly increase AI citations?

Not on its own. A 2026 Ahrefs study found that adding schema produced no direct uplift in citations. Schema works by helping engines understand, attribute and trust content that is already genuinely worth citing, so treat it as an amplifier of good content, not a shortcut.

What is the single most important schema type for AI search?

Organization schema with a stable @id and a complete sameAs array that links to your Wikipedia and Wikidata entries and your verified profiles. It resolves your identity so engines can attribute claims to you with confidence, which is the foundation for being cited.

Is FAQPage schema still worth using after Google deprecated the rich result?

Yes. Google ended the FAQ rich result in 2026, so it no longer wins extra search real estate, but AI engines still use FAQPage markup to identify clean question-and-answer content they can lift into conversational answers. Write self-contained answers and avoid near-duplicate questions.

Should I use JSON-LD or microdata?

JSON-LD. It keeps structured data separate from your HTML, is easier to maintain, and receives the strongest support from modern AI systems including ChatGPT, Perplexity and Google AI Overviews. Microdata is legacy and harder to keep consistent.

How do sameAs and Wikidata help AI citations?

The sameAs property links your Organization and Person entities to authoritative profiles, especially Wikipedia and Wikidata. Tying your content to a Wikidata identifier is regarded as one of the strongest 2026 signals for being retrieved and cited by generative AI, because it maps you to a known entity.

Can bad or fake schema hurt me?

Yes. Markup that does not match the visible page, or fabricated reviews and ratings, can trigger policy issues and erode trust, and invalid markup can be ignored entirely. Only mark up what genuinely exists on the page, and validate every template before it ships.

Do different AI engines use different schema?

They use the same schema vocabulary but lean on it differently. Google AI Overviews and Gemini use the full stack that classic Google Search understands, while ChatGPT search and Copilot rely on the Bing index and value Article, FAQPage and Organization. One well-built JSON-LD layer serves all of them, so you do not need a separate setup per platform.

How long does schema take to affect AI visibility?

There is no fixed timeline, because engines have to recrawl and reprocess your pages, and schema only helps once the underlying content is strong. Treat it as foundational work that compounds over weeks and months alongside content and entity building, not as a switch that changes results overnight.

Which schema type should a small team implement first?

Organization schema sitewide, with a stable @id and a complete sameAs array linking to your Wikipedia and Wikidata entries and verified profiles. It establishes the identity every other signal builds on, and it is the single highest-leverage hour of schema work you can do.