The structured data fields that most affect AI citation frequency

The conventional advice about structured data and AI visibility goes something like this: implement FAQPage schema, add Article schema with author attribution, make sure your Organization schema is complete — and your chances of appearing in AI-generated answers will improve measurably.

In June 2026, Ahrefs tested that claim. They tracked 1,885 web pages that added JSON-LD schema between August 2025 and March 2026, matched them against 4,000 control pages, and measured citation changes across Google AI Overviews, Google AI Mode, and ChatGPT over thirty days before and after implementation.

The result: no meaningful citation uplift on any platform. The only statistically significant movement was a 4.6% decline in Google AI Overview citations — a small number, but the only one that cleared the significance threshold.

This finding matters. Not because structured data is useless, but because a great deal of the advice circulating about schema markup and AI citation is built on correlation studies and vendor case studies that do not hold up under controlled experimental conditions. The Ahrefs study is one of the cleanest measurements available, and it produced a flat line.

Understanding what that flat line means — and what it does not mean — is the starting point for a structured data strategy that actually connects to AI citation outcomes.

What the Ahrefs data actually shows

The Ahrefs methodology matters here. This was not a correlation study — it was a matched difference-in-differences analysis, which controls for the pre-existing trajectory of both treated and control pages. That design removes a major confound in most schema studies: the fact that sites adding schema are often also improving their content at the same time, which makes it impossible to attribute citation changes to the schema itself.

Under controlled conditions, the schema alone did not move the needle. The authors were direct about the interpretation: adding schema markup is not a proven AI citation lever. ChatGPT, Perplexity, and Google's AI systems are not selecting citations because a page has FAQPage or HowTo JSON-LD attached to it. They are selecting citations because the content answers the question better than available alternatives, is clearly sourced, and demonstrates genuine expertise.

On the same week the Ahrefs study published, Google officially retired FAQ rich results entirely. As of May 7, 2026, FAQPage schema produces zero SERP lift. Search Console's FAQ filter retires in June 2026; API support ends August 2026. A schema type that an entire category of GEO advice was built around is now inert in Google Search — and has no proven positive effect on AI citations either.

These two events — the Ahrefs study and the FAQ retirement — are not unrelated. They reflect the same underlying shift: schema as a standalone visibility tactic is losing signal value as AI systems mature.

What structured data does do

The Ahrefs finding is not an argument for ignoring structured data. It is an argument for understanding exactly what structured data accomplishes — and what it does not.

Schema markup serves three functions that remain legitimate and worth pursuing:

• Entity disambiguation. Organization schema with stable @id and verified sameAs properties tells AI systems which brand you are — reliably, unambiguously, across every page of your site. This matters most for branded queries and for Knowledge Graph presence. It is not a citation trigger, but it is foundational infrastructure for entity recognition.
• Content-type signaling. Article schema with datePublished and dateModified fields helps AI systems assess freshness — a real factor in retrieval decisions, particularly on fast-moving topics. A page without explicit publication and modification dates leaves this signal to inference, which is less reliable than declaration.
• Retrieval-augmented accuracy. A March 2026 arXiv study by Volpini et al. found that JSON-LD marked content produced a 29.6% improvement in RAG retrieval accuracy compared to plain HTML in standard pipelines. This is distinct from whether a page gets cited — it concerns whether the content extracted from a page is correctly attributed and accurately reproduced when it is cited. Better structured data means the AI gets your content right, not just that it finds you.

The third function is worth sitting with. A brand with strong domain authority and citable content that lacks complete Organization schema risks being described inaccurately in AI-generated answers — not because the AI ignores it, but because the AI has to synthesize an entity description from whatever signals it can find, and those signals may be inconsistent. Schema does not make you more likely to be cited. It makes the citation more accurate when you are.

The four fields that do the most work

Most schema implementations either go too far — adding every possible property in hope that more is better — or too narrow, adding only the types that produce visible SERP features. Neither approach is optimized for AI visibility.

The fields that do the most work for AI citation accuracy and entity recognition are concentrated in four schema types:

Schema Type	Field	What it does for AI visibility
Organization	@id, sameAs, name, url, description, knowsAbout	Establishes stable entity identity. sameAs links to verified profiles (Google Business, LinkedIn, industry directories) allow AI systems to cross-reference and confirm the brand entity. knowsAbout declares topical expertise — the most underused property in 2026, with measurable impact on branded AI Mode citations.
Article	datePublished, dateModified, author, about, headline	Freshness and authorship signals. dateModified matters more than datePublished for retrieval — AI systems weight recency, and a page last modified six months ago is treated differently from one modified last week, even if the publish date is the same. about with named entities increases topical attribution precision.
FAQPage	mainEntity, Question, Answer, acceptedAnswer	Does not improve citation frequency in controlled studies. Does produce independently-citable content chunks when content is retrieved. Each FAQ pair is a discrete extractable unit — if the page is cited, the FAQ content extracts more cleanly than equivalent prose. Implement for content accuracy, not as a citation trigger.
BreadcrumbList	itemListElement, item, name, position	Site architecture signal. Helps AI systems understand content hierarchy and topical context — particularly useful on deep product or service pages where the breadcrumb path clarifies the page's role within the site's topical structure.

Fields to deprioritize: HowTo schema is inert in Google Search since May 2026. SiteNavigationElement, WPSideBar, and other WordPress/Shopify plugin defaults add page weight without providing extractable information. QAPage is redundant with FAQPage for most implementations. Adding these does not help AI citation outcomes, and in the case of conflicting or inaccurate schema, may actively confuse retrieval.

What actually drives citation frequency

If schema markup is not the lever for AI citation frequency, what is? The Princeton/Georgia Tech GEO research — the first academic study of generative engine optimization, published in 2023 — tested a range of content and technical interventions and measured their impact on AI citation rates. The findings are specific and directional:

• Adding inline citations to primary sources improved AI citation rates by 40%. When a page cites specific named sources for its claims, AI systems treat it as higher-confidence material. The content is verifiable — the AI can cross-reference the claim against the source cited, and that verification increases citation likelihood.
• Adding specific statistics improved citation rates by 37%. Original or specifically-attributed data points are cited at significantly higher rates than general assertions. A page that says "48% of Google queries now trigger an AI Overview" is more citable than a page that says "most Google queries now trigger AI Overviews." The precision is the signal.
• Adding named expert quotations improved citation rates by 22%. First-person attribution from identifiable experts increases citation likelihood across platforms. AI systems are weighting authority indicators in content, not just content type.

These are content structure interventions, not technical markup interventions. The implication is direct: a page that cites its sources, uses specific data, and attributes claims to named experts will be cited more frequently by AI systems than an equivalent page with comprehensive schema but generic content. The schema is the packaging. The content quality and sourcing are what determine whether the package gets opened.

The practical priority order

Given the current research, a rational structured data strategy for AI visibility looks like this:

• Implement Organization schema completely. Name, URL, logo, description, sameAs (all verified profiles), knowsAbout (your actual areas of expertise), telephone, and address. Do this once, do it correctly, put it in your sitewide template. This is entity infrastructure — it does not expire and it compounds over time.
• Implement Article schema on all editorial content. Include datePublished, dateModified, author as Organization entity, about with named topic entities, and headline. Keep dateModified current — update it whenever content is substantively revised, not just reformatted.
• Add FAQPage schema where you already have visible FAQ sections. Only where the Q&A content is genuinely visible on the page — marking up hidden content is both against Google's guidelines and ineffective. Implement for extraction accuracy on pages you expect to be cited, not as a traffic tactic.
• Invest the remaining time in content structure. The Princeton data is clear: sourced claims, specific statistics, named expert attribution, and answer-first paragraph structure move citation rates more than any technical implementation tested. Write content that cites its sources. Use specific numbers. Put the answer in sentence one.

The brands that are winning in AI citation in 2026 are not winning because their schema is more comprehensive than their competitors'. They are winning because their content answers questions better, cites its sources more reliably, and is recognized by AI systems as a trustworthy entity. Structured data is part of the entity recognition infrastructure. It is not the citation mechanism.

That distinction is the one most agencies are not making — and it is where the meaningful competitive advantage lies.