Does Schema Markup Get You Cited by AI? What the Data Actually Shows

Almost every guide to AI search optimization gives the same advice near the top: add JSON-LD, mark up your FAQs, and the AI engines will cite you. The reasoning sounds sensible, because structured data is machine-readable and AI systems are machines. In 2026 Ahrefs tested that assumption at scale, and the result is awkward for anyone who sells schema as a way to get cited.

This post brings the data together. It separates what structured data is proven to do from what it only correlates with, using a controlled study of 1,885 pages, a retrieval experiment across five AI systems, and the citation research that people often quote. For the broader picture, start with the guide on SEO vs AEO .

The experiment: schema barely moved citations

In 2026, Ahrefs ran a controlled test. They tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched them against roughly 4,000 control pages that did not, and measured the change in citations across Google AI Overviews, Google AI Mode, and ChatGPT. If schema were a citation lever, the pages that added it should have pulled ahead. They did not.

Citations in AI Overviews dipped, and the small positive moves on AI Mode and ChatGPT fell within the range of random variation. Adding schema produced no meaningful lift on any platform.

Why correlation looked like causation

If schema does not cause citations, why do case studies show cited sites covered in structured data? The likely reason is that cited sites tend to be well built, and well-built sites tend to have schema. The markup travels with qualities such as authority, depth, freshness, and clean structure. Three figures are quoted often, and all three are correlations.

These figures describe what cited pages look like. They are not instructions for earning citations. The 8x FAQPage gap from Insightland shows that cited sites carry FAQPage schema far more often. It does not show that adding FAQPage schema makes a page 8x more likely to be cited. The Wellows +73% and 96% figures were measured in Google AI Overviews specifically, and E-E-A-T is a Google quality concept. Neither shows that a single markup signal causes a citation.

What AI crawlers actually read

Most guides skip the mechanism. When an AI assistant fetches your page in real time to answer a question, it may not read your JSON-LD at all. A 2025 searchVIU experiment tested ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode, and during direct retrieval every one of them extracted only the visible HTML. JSON-LD, hidden Microdata, and hidden RDFa were all ignored.

In that test, structured data did not reach the model through the fetched page. Its path is indirect: schema feeds the search index (rich results in Google and Bing), and those indexes feed AI search products. This pathway is real but second-hand, which fits the small effect of adding schema to a page that is already indexed.

Is schema useless? No.

Structured data still does three concrete jobs, although directly causing AI citations is not one of them.

What actually drives AI citations

If schema is a minor, indirect factor, your effort belongs with the signals that citation research keeps finding. These are mostly correlational too, but they describe cited content more consistently than markup does.

• Authority and brand. Cited domains skew toward established, frequently referenced sites. This is the strongest pattern across the studies.
• Comprehensive content. Pages that answer the main question and naturally cover related angles are cited more often than narrow, single-keyword pages.
• Question-led structure. In Kevin Indig / Growth Memo's analysis of 3M ChatGPT responses, among citations tied to a question, 78.4% came from a heading, and 44.2% of citations came from the first 30% of the page. AI tends to treat an H2 as a prompt and the text beneath it as the answer.
• Freshness. Recently updated content is associated with substantially more citations, though the link is a strong correlation and not a guaranteed multiplier.
• Being in the index. Allow the right crawlers (OAI-SearchBot and friends) so you can be retrieved at all. This is a hard requirement.

Where to invest your effort

This priority order follows the evidence:

1. Make sure the right AI crawlers can reach you. Without access there is no citation.
2. Write comprehensive, well-structured content with question-led headings and the answer up top.
3. Keep important pages fresh, with clear dateModified signals.
4. Build authority the slow way: become the source that other sources reference.
5. Add structured data, but for the right reasons: rich results and agent parseability, not a citation multiplier.

The bottom line

Schema markup is a baseline for machine-readability and a sensible investment for rich results and agent parseability. It is not a switch that makes AI cite you. In the controlled data, adding it to a page did not move citations. Keep your structured data clean, then spend the rest of your effort on what correlates with being cited: authority, comprehensiveness, structure, and freshness. Our scanner scores all of these.

To see how individual AI systems choose sources, read the companion posts on how ChatGPT chooses which websites to cite , how Google AI Overviews selects sources , and what AI agent readiness means .

Sources

• Ahrefs: We Tracked 1,885 Pages Adding Schema (controlled study of schema's effect on AI citations)
• searchVIU: What AI assistants actually parse (retrieval test across five AI systems)
• Kevin Indig / Growth Memo: The science of how AI picks its sources (3M responses, 30M citations, content-structure analysis)
• Insightland: Structured Data and AI Search (FAQPage schema visibility correlation)
• Wellows: AI Overview Ranking Factors (schema and E-E-A-T correlations in Google AI Overviews)