Does Schema Markup Help You Get Cited in AI Search?

• The headline stat is correlational and badly confounded — schema is a marker of a well-built site, not obviously a cause.
• No published evidence establishes that adding markup causes an AI citation.
• What structured data does do is remove inference: authorship, dates, entity identity, relationships.
• FAQ rich results are gone for nearly everyone since 2023. Guides promising them are stale.
• Ship it — for machine-readability and rich results — and do not pay anyone who sells it as a citation lever.

What the statistic actually shows

Take the finding at face value: most pages cited in AI answers carry structured data. Now ask what else is true of those pages.

They tend to be on sites with editors. With named authors. With maintained publishing systems that emit correct markup because a CMS does it automatically. With technical hygiene, working canonicals, sensible internal linking, and content someone is paid to keep accurate.

Structured data is a marker of that whole cluster. It is one of the most reliable available proxies for "this site is professionally maintained" — which is precisely why the correlation is strong, and precisely why it cannot be read as causal. A study that found most cited pages used HTTPS would be equally true and equally uninformative.

To establish causation you would need the obvious experiment: comparable pages, identical in every other respect, differing only in markup, measured for citation rate at scale. That study does not exist publicly. What circulates instead is:

• Correlational audits of cited pages, which cannot separate schema from everything schema travels with.
• Vendor multiplier claims — "sites with complete schema see 2.5× the citation rate" — which are the same correlation with a ratio attached.
• Tiny controlled demonstrations, such as a widely shared three-page experiment where only the page with JSON-LD appeared in an AI Overview. Three pages is an anecdote. It is a reasonable thing to try and not a basis for a claim.

There is also a claim in circulation that internal Google documentation confirms structured data as a ranked input to AI Mode source selection. We could find no primary source for it. Treat unsourced assertions about Google's internals as what they are, particularly when they arrive attached to something being sold. The same pattern played out with llms.txt, where a proposal with no adoption by any major AI vendor was marketed for a year as an SEO requirement.

What structured data genuinely does

Strip out the citation promise and a solid case remains.

It removes inference. Without markup, a system reading your page must guess who wrote it, when it was published, whether the date near the top is a publish date or an event date, and whether the "Sarah Chen" here is the "Sarah Chen" on your author page. With markup, it does not guess. It reads. Every guess removed is a chance to be wrong about you removed.

It drives rich results in Google Search. This is documented, stable, and unrelated to AI. Breadcrumbs, article metadata, and the rest have well-understood effects on how a result is displayed.

It builds an entity, not a pile of pages. This is the underrated part. Using a stable @id for your author so the Person node on the author page and the author node on every article are the same node means a crawler sees one person with a body of work, rather than a series of unrelated people who share a name. Do the same for the publisher Organization across the site. Entity consolidation is the mechanism by which author-level credibility accumulates at all.

It is cheap and permanent. A CMS emits it once, correctly, forever. The cost argument is not close.

The parts that are out of date

Two things in most schema guides are simply wrong now.

FAQ rich results are effectively gone. In 2023 Google restricted FAQ rich results to well-known authoritative government and health sites. If you are not one of those, marking up FAQs will not restore the accordion in the SERP. It remains reasonable for machine-readability — a question and its answer as an explicit pair is genuinely useful to a retrieval system — but the reason to do it changed, and any guide still promising SERP real estate has not been updated in three years.

HowTo rich results were deprecated too. Same story.

The general lesson: schema advice ages badly because Google retires rich result types faster than guides get rewritten. Check the current documentation rather than a blog post, including this one.

What to actually implement

For a publication, the useful set is short:

• Article — with a real author, datePublished, and an honest dateModified. Honest matters: a dateModified that updates on every deploy is a lie about freshness that systems will eventually learn to discount.
• Person for the author, with a stable @id shared between the article's author node and the author page. If those two @ids differ, you have created two people and neither accrues anything.
• Organization for the publisher, referenced by @id from every article rather than repeated inline.
• BreadcrumbList matching the visible breadcrumb trail.
• WebSite on the homepage, tying the graph together.
• FAQPage only where the questions and answers are genuinely on the page, visible to a human, word for word.

That last condition is not a style preference. Markup that does not match visible content is a structured-data policy violation and can earn a manual action. The failure mode is not "no benefit" — it is a penalty.

There is a subtler failure worth naming too. Schema that is technically valid but inaccurate — a stale dateModified, an author who did not write the piece, an Organization that overstates what you are — teaches every consuming system something false about you. Saying nothing is safer than saying something wrong, because wrong information is what gets propagated and repeated.

So: does it help you get cited?

The defensible answer, in one paragraph.

Structured data makes your page easier for a machine to parse correctly, and there is no scenario in which being harder to parse correctly helps you. That is a real argument, and it is enough to justify shipping it. What there is not is evidence that adding markup to an otherwise unchanged page causes AI systems to cite it — and the studies most often used to claim that cannot support it, because they measure sites that differ in a hundred other ways at the same time.

Ship schema because it is cheap, permanent, and removes ambiguity. Do not ship it expecting citations, do not pay a premium for it as a "GEO" service, and be suspicious of anyone quoting a multiplier. What the evidence does point at for citation — specific, self-contained, verifiable passages, and crawlers that are actually allowed in — is covered in how to get cited in AI search and AI crawlers and robots.txt. The retrieval mechanism that makes self-contained passages matter is explained in AI Mode vs AI Overviews.

Sources: Google Search Central — Intro to structured data markup · Google Search Central — Structured data general guidelines · Google Search Central — AI features and your website · Google Search Central Blog — Changes to FAQ and HowTo rich results