Does schema markup help you get cited by AI search engines?

Last updated October 14, 2026. Written by the Third Coast AI team.

Schema markup does not, on the only controlled evidence available, raise the rate at which ChatGPT, Google AI Overviews or AI Mode cite a page, but it does help the engines identify who you are, which makes correct Organization markup with a complete sameAs list worth doing, while FAQPage and HowTo markup added in the hope of citations is work we would not bill for. Microsoft is the one engine operator that has said its models use structured data to understand pages, so schema is a real lever for Copilot and Bing and hygiene everywhere else.

This page sorts the schema types by what they are for, sets out what the evidence says and, just as important, what it does not say, and ends with the markup we put on every client site and the markup we leave off. Our overview page covers where schema sits among the other signals.

Which test should you trust?

The strongest study is Ahrefs' schema test from May 2026, which tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, matched each against a pool of 4,000 control pages that never added it, and measured citations 30 days before and after using a difference-in-differences design so that platform-wide trends washed out. Citations on Google AI Overviews moved by minus 4.6%, a small decline the authors called statistically significant, while AI Mode moved by plus 2.4% and ChatGPT by plus 2.2%, both too small to separate from noise. The authors' conclusion was that adding schema produced no major uplift in citations on any platform.

The same team had earlier found, in a correlational pass over 6 million URLs, that pages cited by AI were almost three times more likely to carry JSON-LD than pages that were not, with 53% of cited pages running schema. That earlier number is the one most vendors quote, and it is the one the controlled test was designed to check, because sites that bother with schema also tend to publish better pages and earn more links. When the confounders were removed, the effect disappeared. We think the second study is the one to act on.

Is there anything the engines themselves have said?

Google's AI features page is direct about it: there is no special schema.org structured data you need to add to appear in AI Overviews or AI Mode, no new machine-readable file, and no special optimization beyond what already helps a page in Search. Google's Organization markup page says the markup helps Google understand your administrative details and disambiguate your organization in results, which is a statement about identity rather than ranking, and it lists no required properties at all. Google's FAQPage markup page has, since September 2023, limited the FAQ rich result to well-known government and health websites, so for a business the visible benefit of that markup in Google is already gone.

Microsoft has said Copilot does use structured data for grounding, which is the one on-record statement from an engine operator that schema feeds a model, and it fits what we see when we run the same prompts in Copilot and in ChatGPT. We treat it as the reason to keep schema correct rather than as a reason to expect citations from it.

Why does sameAs matter more than the rest?

Because the engines' hardest job with a mid-sized company is deciding that the various traces of it on the web belong to one entity. A sameAs array that points at your LinkedIn page, your Google Business Profile, your Wikidata item, your Crunchbase or industry-directory entry and your review profiles gives the parser a chain to follow, and it costs ten minutes. The chain only works if the profiles agree with one another, which is why we reconcile names, addresses and descriptions before closing it; a sameAs link to a profile with a different address makes things worse. The entity page walks through that order of work.

Can schema do harm?

It can when it disagrees with the page. Markup that claims a five-star aggregate rating the page does not show, or a dateModified that moves every week while the text stays the same, is the kind of thing Google's structured data guidelines call out, and the engines' fact-check behavior will eventually notice the gap between what the markup says and what the visible text says. The rule we apply is that every schema value must be visible on the page in the same words. Ahrefs' schema test also showed the small AI Overview decline, which the authors did not explain and we would not over-read, but it is a reminder that adding markup is not free of risk.

Which markup should a $5M to $100M company put on its site this quarter?

1. Organization schema on the home page with name, URL, logo, address, founder and a sameAs list of every real profile, checked against the profiles first.
2. Service schema on each service page that matches the visible text and names the area served.
3. Article schema on every article with author, datePublished and a dateModified that only changes when the content does.
4. LocalBusiness where there is a storefront, with hours and address identical to Google Business Profile.
5. A validation pass in Google's Rich Results Test and the Schema Markup Validator, then a check that nothing in the markup is absent from the page.
6. Nothing else until the pages themselves state and prove the decision factors, because that is where the citations come from.

The order reflects where the evidence is. Identity markup has a clear mechanism and an operator's endorsement, content markup has neither, and the page structure work described in our content structure guide has the strongest published effects of anything on this list.

Frequently asked questions