Does Schema Markup Get You Cited by AI? What the Evidence Says

Schema markup has become one of the most common recommendations in AI search advice. Add FAQ schema, add Organization schema, add Product schema, and answer engines will understand and cite you. It is an appealing idea because schema is concrete, measurable and easy to sell. The problem is that the best evidence available does not support it.

This article reviews what the research actually found, explains why schema and AI citations are less connected than they seem, and sets out where schema is still worth the effort.

What did the largest study find?

The most rigorous test so far is a 2026 Ahrefs study. It followed 1,885 pages that added JSON-LD structured data and compared them with roughly 4,000 similar pages that did not, then measured how AI citations changed afterwards.

The results were close to zero:

The decline in AI Overviews looks alarming, but the authors note that the pages adding schema were already losing citations before they added it. The study also only looked at pages that were already being cited heavily. So the safest reading is narrow and important: for pages AI already cites, adding schema did not increase citations.

The study does not tell us whether schema helps a page that AI systems cannot currently interpret at all. That question remains open.

Why do some studies show a link?

Other analyses have found that cited pages are more likely to have schema than uncited pages. Ahrefs itself found cited pages were almost three times more likely to carry JSON-LD. A study by Growth Marshal reported that pages with detailed, attribute-rich schema had higher citation rates, while pages with thin, generic schema did worse than pages with none.

These findings are not contradictions. They are correlations. Sites that implement detailed schema tend to be well-built, well-maintained and well-resourced. Those same qualities make a site more likely to be cited. The schema travels with quality rather than causing the citation.

Controlled studies that change one thing and measure the effect are the right way to answer a causal question, and the controlled evidence points to no meaningful effect.

Do AI systems even read schema?

It depends on the system and the stage. In Ahrefs' tests, when AI systems fetched a page directly to answer a question, they extracted the visible HTML and ignored JSON-LD, hidden microdata and hidden RDFa. If the information only exists in schema and not on the visible page, a live-retrieval engine is unlikely to use it.

Google is a different case. Google's search systems have read structured data for years, and a Google engineer has reportedly said that structured data can be used as context when Google's AI features fan out into background searches. That suggests schema may matter upstream, during indexing and understanding, rather than at the moment an answer is written.

The practical conclusion is the same either way: anything you want an AI engine to use must be on the visible page. Schema can describe it; it cannot replace it.

Where schema is still worth it

Schema remains a sound investment for reasons that have nothing to do with AI citations.

Rich results in Google Search

Product, review, recipe, event, video and other schema types can earn enhanced listings in Google Search. These still drive clicks on queries where traditional results are shown.

Entity clarity

Organization, Person, LocalBusiness and sameAs properties help search engines connect a site to its brand, people and profiles across the web. That supports the broader entity work described in entity SEO for LLMs.

Data consistency

Structured product and pricing data keeps feeds, listings and on-page information aligned. When facts agree across a site and its listings, every system, AI or not, has an easier job.

Merchant and local surfaces

Shopping listings, maps and other Google surfaces rely on structured data feeds. For e-commerce and local clients, schema is part of basic visibility.

What to tell clients

Clients have often heard that schema is the key to AI search. A clear, honest message builds more trust than overselling.

1. Schema is good practice and we will keep it accurate. It helps rich results and helps search engines understand the business.
2. It is not a lever for AI citations on its own. The largest controlled study found no meaningful lift.
3. What earns citations is the visible content. Clear answers, specific facts, original information and a strong reputation across the web.

If a client is paying for "AI schema" work from another provider, this is a useful conversation to have early.

A schema checklist that makes sense in 2026

• Implement the types that earn rich results for the client's vertical. Product, Review, LocalBusiness, Event, Recipe, Article and so on.
• Fill properties fully and accurately. Thin, generic schema adds little. Detailed, correct schema is the version worth doing.
• Mirror visible content. Never put information in schema that is not on the page.
• Validate after every template change. Use Google's Rich Results Test and Schema Markup Validator.
• Connect entities. Organization with logo, sameAs links to official profiles, and consistent names and addresses.
• Do not add FAQ schema purely for AI. Write real FAQ sections because readers need them; mark them up if it fits, but do not expect citations from the markup.

Where to put the effort instead

If schema is not the lever, what is? The evidence across our research points to a handful of factors that matter on every engine:

• Crawler access. Make sure search and user-triggered AI crawlers are allowed. See AI crawler user agents.
• Visible, server-rendered content. Most AI crawlers do not run JavaScript. See can AI crawlers read JavaScript?.
• Answer-first writing. Passages that answer a question on their own. See answer-first content.
• Corroboration. The brand described consistently on trusted third-party sites. See how AI search engines choose sources.

Which schema types matter for which clients

Schema is still worth doing, but the useful types depend on the client. This table is a practical starting point.

Check Google's current documentation before rolling out a type, because Google has reduced or removed rich results for some types over time.

How to test schema for a client yourself

If a client is convinced schema will move AI citations, run a small test instead of arguing.

1. Pick two matched groups of pages. Similar templates, traffic and current citation rates. Twenty pages each is enough for a directional read.
2. Record a baseline. Run a fixed prompt set relevant to those pages on each engine and note citations for four weeks.
3. Add complete, accurate schema to one group only. Change nothing else on either group.
4. Keep measuring for six to eight weeks. Compare the change in citations between the groups, not the raw numbers.

If the schema group pulls ahead, you have evidence for that client. If not, you have saved them from paying for work that does not help. Either way, the conversation is grounded in their own data.

How Rankwyn helps

Rankwyn's audits check that structured data is valid and matches the visible page, and its AEO Citation Radar measures actual citations, so you can see whether any change, schema or otherwise, moves the result rather than assuming it will.

Key takeaways

• The largest controlled study found no meaningful increase in AI citations after pages added schema, and a small decline in AI Overview citations for pages that were already declining.
• Studies that find cited pages carry more schema are measuring correlation: well-built sites both cite-worthy and well-marked-up.
• Live-retrieval engines used visible HTML and ignored JSON-LD in testing, so anything an engine should use must be on the visible page.
• Schema is still worth doing well for rich results, entity clarity and consistent product and local data, so keep it complete and accurate.
• If a client wants proof either way, run a matched-group test on their own pages and measure citations directly.

The bottom line

Schema markup is worth doing well, but the best evidence says it will not get a page cited by AI on its own. Keep it accurate and complete for rich results and entity clarity, keep everything important on the visible page, and measure citations directly instead of relying on markup as a proxy. When a vendor promises citations from markup alone, ask for a controlled test on comparable pages before paying for it. For the wider strategy, see our AEO and GEO guide.