Does Schema Markup Help You Get Cited by AI?

Updated 19 October 2026 · 10 min read · By Matthew Bertram

So mark up your identity with care and keep every fact you want quoted in the visible text. Two popular pieces of schema advice have also expired. Google stopped showing FAQ rich results on May 7, 2026. QAPage markup is invalid on the pages people have been moving it to.

What does schema markup do?

It labels things. Schema markup is code that uses the schema.org vocabulary to say what a page contains: this is an article, this person wrote it, this company published it, this is the date it last changed. Most sites write it as JSON-LD, a block of code in the page that visitors never see, and SEO plugins such as Rank Math add much of it automatically, as our short answer on schema markup explains.

Google's documentation says it uses structured data "to understand the content of the page" and to learn about the people and companies named in the markup. The visible payoff has been rich results. Those are the listings with extras like star ratings or prices. Even there, Google's guidelines promise nothing: "Using structured data enables a feature to be present, it does not guarantee that it will be present."

What happened when 1,885 pages added schema?

Citations didn't rise. Ahrefs found 1,885 pages that added JSON-LD between August 2025 and March 2026. For each one, the analyst picked three control pages from other domains with similar citation levels that had never added JSON-LD, then compared AI citations over the 30 days before the change and the 30 days after.

Against those controls, citations in Google's AI Overviews fell 4.6%, and Ahrefs puts the odds of a gap that size arising by chance at about 1 in 2,500. AI Mode rose 2.4% and ChatGPT 2.2%, both "statistically indistinguishable from zero". The AI Overviews drop averaged about 12 daily citations per page, in a sample where most pages drew hundreds. Ahrefs itself calls it a decline "that we can't clearly attribute to schema."

The controls are the point. Measured on their own, AI Mode citations for these pages grew 43%. The control pages gained almost as much, because in Ahrefs' words "AI Mode was exploding for everyone". Strip that out and the 43% shrinks to 2.4%. A before-and-after check on your own site, with no control pages to compare against, would have credited the markup with a 43% gain in AI Mode citations, and you'd have gone on adding schema to every template on the strength of a number that mostly measured AI Mode's own growth.

The same study shows why schema advice sounds so convincing. Ahrefs found JSON-LD on cited pages at almost three times the rate of uncited ones. Read alone, that looks like proof. Ahrefs' explanation is that sites which add markup tend to do the other work too: technical SEO, authoritative content, link building, page upkeep, ranking well in regular search.

Ahrefs lists its own limits. Every page in the study already had 100 or more AI Overview citations in February 2025, before any markup went on. The window was 30 days. Schema types were pooled, from Article to Product. Only markup in the page's HTML counted, not markup injected by JavaScript. A page that no AI system sees yet might behave differently, and Ahrefs says plainly, "Our study can't speak to that directly".

Their conclusion is narrow: "if the only reason you're adding it is to get more AI citations on pages that are already visible, our data doesn't support that bet." For pages that aren't cited yet, we'd put the effort into covering the questions your buyers ask first. Our guide to how AI Overviews choose their sources starts there.

Do AI assistants read your JSON-LD?

Not in the one controlled test we found. On October 30, 2025, searchVIU tested a page carrying eight product prices, each placed a different way. One price existed only in JSON-LD. Another sat in plain visible text. Two more were hidden from readers: one in Microdata, one in RDFa. Five systems were asked for the prices: ChatGPT, Claude, Gemini, Perplexity, Google's AI Mode.

The JSON-LD price "was not found by any of the 5 tested systems". That held even after Perplexity and AI Mode, which answer from a search index, had indexed the page. Neither hidden format registered either, and searchVIU's summary is blunt: "Hidden structured data is ignored." The visible price fared better. ChatGPT and Gemini reported it, and so did AI Mode once the page was indexed. Claude reported no prices at all, even the visible ones.

That's one page. Its write-up also gives the number of repeat questions as 10 in one place and 5 in another, so treat it as a strong signal, not a general law. searchVIU is also clear about what it didn't test. Training data is one gap. Google's AI Overviews and Bing's Copilot are another, and searchVIU says both "have access to this structured data from their search indexes." That part is its expectation, not something it measured.

Google's on-record position points the same way. Its AI features documentation says "There's also no special schema.org structured data that you need to add." Its list of good practices includes "Making sure your structured data matches the visible text on the page". The general guidelines go further: "Don't mark up content that is not visible to readers of the page." So a fact that lives only in your markup breaks Google's own rule, and the one test we have says assistants skip it anyway.

Which schema advice is out of date?

Is FAQPage schema still worth adding?

No. By September 2023, Google's documentation already limited FAQ rich results to "authoritative government and health websites". Its update log then recorded the end: the feature stopped appearing in Google Search on May 7, 2026. FAQPage markup now feeds nothing a searcher can see. It does no harm while the questions and answers it describes are visible on the page. Our own short answer pages still carry it. They carry speakable markup too, and both match the visible text. There's no reason to add FAQPage anywhere new, and no hurry to strip it from pages where it matches what readers see.

Should you switch FAQPage to QAPage?

No, and this one catches careful people. Google's QAPage documentation covers forum-style pages, and it says "Users must be able to submit answers to the question." Among the examples it calls invalid is "An FAQ page written by the site itself with no way for users to submit alternative answers". Another is "A blog post that answers a question". It caught us too. Our first plan for these guides called for QAPage on every one of them. HowTo markup is gone as well. Google removed its documentation on September 14, 2023, saying the rich result was no longer shown.

Does speakable schema help with AI answers?

Nothing we found says it does. Google's speakable documentation says the feature "is in beta and subject to change." Its stated use: "The Google Assistant uses speakable structured data to answer topical news queries on smart speaker devices." It "works for users in the U.S. that have Google Home devices set to English". That's a news feature for one kind of speaker, not an AI search lever.

What about llms.txt?

It isn't schema, but it gets sold the same way. Google's update log says these files "aren't needed for Google Search". Other engines are a separate question, and our guide to AI SEO gives the full verdict.

Which markup still earns its place?

Identity markup does. A machine reading your page has to work out which person and which company stand behind it, and markup is the one place you can state that outright. In practice that means a Person node for each author and an Organization node for the business, each with a stable @id, plus Article markup on articles that points to both.

Google's Article documentation says "There are no required properties", but for authors it says "we strongly recommend using the type and url (or sameAs) properties." It also asks you to "Make sure that all the authors that are presented as authors on the web page are also included in markup" and to "Add the dateModified property if you want to provide more accurate date information to Google." That's item five on the checklist in our guide to answer engine optimization: a named author, marked up so a machine can tell who's answering.

Markup won't make an assistant describe you correctly. It does stop your own pages from contradicting each other about who you are, and accuracy is its own subject in our guide to AI visibility. You can hold this page to the same rules. The date under its headline is the dateModified in its markup. It ends with a visible FAQ and carries no FAQPage markup.

A minimal @graph to adapt

This goes in one JSON-LD script block in the page. Every value below is a placeholder.

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Organization",
      "@id": "https://www.example.com/#organization",
      "name": "Example Co",
      "url": "https://www.example.com/",
      "logo": {"@type": "ImageObject", "url": "https://www.example.com/logo.png"}
    },
    {
      "@type": "Person",
      "@id": "https://www.example.com/#jane-doe",
      "name": "Jane Doe",
      "url": "https://www.example.com/about/jane-doe",
      "sameAs": ["https://www.linkedin.com/in/example-jane-doe"],
      "worksFor": {"@id": "https://www.example.com/#organization"}
    },
    {
      "@type": "WebSite",
      "@id": "https://www.example.com/#website",
      "url": "https://www.example.com/",
      "publisher": {"@id": "https://www.example.com/#organization"}
    },
    {
      "@type": "Article",
      "@id": "https://www.example.com/guides/widgets#article",
      "headline": "How Widgets Work",
      "image": "https://www.example.com/images/widgets.jpg",
      "datePublished": "2026-10-01",
      "dateModified": "2026-10-18",
      "author": {"@id": "https://www.example.com/#jane-doe"},
      "publisher": {"@id": "https://www.example.com/#organization"},
      "isPartOf": {"@id": "https://www.example.com/#website"},
      "mainEntityOfPage": "https://www.example.com/guides/widgets"
    }
  ]
}

• One graph per page. If a plugin already writes one, extend it instead of adding a second.
• An @id is a name for a node, not a link anyone visits. Give each thing one and never change it.
• The headline and both dates must match what the page shows.
• Replace every example value before it ships.

Check the result in the Schema Markup Validator and in Google's Rich Results Test. Both check the markup itself. Neither can tell you whether an AI engine will cite the page.

Which @id mistakes confuse machines?

Each of these turns one clear identity into something muddier. Open your page source and search for @id. Check each one against the table.

The cause of the first row is in the JSON-LD 1.1 specification: "In JSON-LD, a node is identified using the @id keyword". When a processor flattens a document, it "collects all properties of a node in a single map", so two nodes that share an @id become one thing carrying both sets of properties. For the last row, schema.org defines sameAs as the "URL of a reference Web page that unambiguously indicates the item's identity."

This site made the second mistake until September 1, 2026. Its pages minted a separate Organization node for EWR Digital, the agency behind the show, under a bestseopodcast.com @id. That node had its own name and logo. EWR Digital's own site already had one. We fixed it. Our pages now point to the @id on ewrdigital.com instead of competing with it.

FAQ

Do I need schema on every page?

Not every type. Identity markup belongs on every page, since every page has someone behind it. Content-type markup belongs where the type is plain: Article on an article, Product on a product page. Leave out anything added on speculation.

Can bad schema hurt you?

In two ways. Google can take manual action, and its guidelines spell out the cost: "A structured data manual action means that a page loses eligibility for appearance as a rich result; it doesn't affect how the page ranks in Google web search." The second cost is quieter. Markup that gives your company two names, or claims profiles that aren't yours, tells every machine that reads it something false about you. Google's guidelines also say "Don't impersonate any person or organization".

How do you test whether a schema change worked?

The way Ahrefs did, with controls. Choose the pages you're changing and a matched set you aren't touching, then count citations for both groups on the same questions in the month before the change and the month after it. Our guide to checking AI citations covers how many repeat runs a fair count needs.

Judge the markup by the gap between the two groups, never by the raw rise. Ahrefs itself flags 30 days as short for a slow-burn effect, so a longer window makes the stronger test.

Want a read of what your markup tells machines about your business, including any second identity a plugin added without asking? Book a discovery call with EWR Digital →

Keep reading: How Do Google's AI Overviews Decide What to Cite? · What Is AI SEO, and How Is It Different From Traditional SEO? · How Do I Know If ChatGPT or Google AI Overviews Are Citing My Site?

Seen an AI assistant quote a fact that existed only in a page's JSON-LD? Show us the page and the answer. One documented case would change what this guide says, and we'd credit whoever found it.