Does Schema Markup Get You Cited by AI? What the Evidence Says

If someone is trying to sell you a "schema package" on the promise that it will get your business cited by ChatGPT or shown in Google's AI Overviews, be sceptical. The question sounds technical, but it reduces to one honest test. Does adding structured data actually cause more AI citations, or does it just happen to show up on sites that were going to be cited anyway? Both are plausible. Only one is true, and the difference is the difference between money well spent and money set on fire. Here is where the evidence lands, whatever the sales deck would prefer.

What schema markup actually is

Schema markup is structured data you add to a page, usually as JSON-LD, that labels your facts for machines: who you are, where you are, what you sell, your opening hours, your prices, your reviews. None of it changes what a human sees on the page. It sits underneath, telling software what each piece of content means so the software does not have to infer it.

That is the whole point of it, comprehension. A crawler chewing through raw HTML can misread an address, mix up two businesses with similar names, or fail to notice that a block of text is really a list of questions and answers. Schema closes those gaps by naming everything explicitly, and that is genuinely useful work. The argument is not about whether schema helps a machine understand a page. It is about whether that understanding earns you citations in AI answers, and that is exactly where the two camps part company.

The case for: platform engineers say their systems use it

The case that schema helps is a serious one, and it starts with the people who build these systems. In March 2025 Microsoft's Fabrice Canel stood up at SMX Munich and confirmed that Bing and Copilot use schema markup to help their large language models understand content (Search Engine Land, 2025). Google's engineers have said much the same thing for years: structured data makes a page easier for their systems to understand, even if they stop short of calling it a ranking factor.

John Mueller has made that distinction over and over. Structured data helps Google grasp what a page is about, which can make it easier to surface that page where it fits, but on its own it delivers "no generic ranking boost" (Search Engine Roundtable, 2024). Then there are the correlation studies, and they look persuasive. One analysis of roughly 6 million URLs found that AI-cited pages were almost three times more likely to carry JSON-LD than pages that went uncited (Search Engine Land, 2026). Put the engineer quotes and the numbers side by side and it reads like an open-and-shut case. It is not, and the reason why is the whole game.

The case against: the controlled test found almost nothing

Correlation is not causation, and for years that objection was all anyone had, because nobody had run the experiment. In 2026 Ahrefs finally did. They tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched them against roughly 4,000 control pages, and measured what happened to AI citations. What happened was almost nothing (Ahrefs, 2026).

Run through a matched difference-in-differences analysis, ChatGPT citations ticked up around 2.2% and Google AI Mode around 2.4%, both well inside the range of random noise you get across thousands of URLs, while AI Overviews actually fell by a statistically significant 4.6% (Ahrefs, 2026). Adding structured data did not buy citations. And there is a mechanical reason it could not. In a separate hands-on experiment, searchVIU checked whether five major systems, ChatGPT, Claude, Perplexity, Gemini and Google's AI Mode, actually read JSON-LD when fetching a page live. Not one did during direct retrieval. Every system pulled only the visible HTML a human would see and ignored the markup entirely (searchVIU, 2025).

Why the correlation studies mislead

The distance between "AI-cited pages often have schema" and "schema causes AI citations" is a textbook confounding problem. Think about who actually bothers to implement structured data. The same teams tend to publish stronger content, earn more links, keep their pages tidy and get the technical basics right. Schema turns out to be a symptom of a well-run site, not the thing pulling in the citation. It rides along with quality rather than creating it.

That is precisely why one controlled test outweighs a stack of correlation studies. Hold the site constant, change only the schema, and the effect evaporates. Ahrefs was also candid about the ceiling on its own finding, and it is worth saying plainly: every page in the dataset was already being cited heavily before the schema went on, so what the test really proves is that schema will not push an already-visible page any higher. A page AI systems cannot see at all is a different case, and there structured data may still earn its keep by helping with crawling and parsing (Ahrefs, 2026). That is comprehension doing its job. It is not a citation lever.

The honest landing: comprehension infrastructure, not a citation lever

Schema is plumbing. It lets machines read your facts without guesswork, it feeds the classic Google features, it costs almost nothing to add, and it clears up the kind of ambiguity that can stop a cautious model naming you when it is not sure who or where you are. The one thing it does not do, on the evidence we have, is buy citations by itself.

Which means both camps were right all along, and that is the honest resolution rather than a fudge. The engineers are right that their systems lean on schema to understand content. The controlled test is right that bolting it on does not, by itself, make AI name you any more often. Anyone pitching "schema package equals AI visibility" is selling well past what the data supports. For the fuller picture of what genuinely moves the needle, read our UK guide to AI search optimisation.

When schema still earns its keep

None of this means skip schema. It means understand what you are buying. Schema is worth doing well wherever accurate facts and a clear entity matter most, and there are places where that is worth real attention. Local businesses, products with concrete attributes and lesser-known brands trying to establish who they are all get the most out of it. Think of it as cheap comprehension insurance rather than a growth tactic, and price it accordingly.

The clearest wins line up neatly. Local businesses using LocalBusiness types, Dentist, Restaurant and the rest, with name, address and phone details that stay consistent across the web. Products with concrete attributes like price, availability and specifications. FAQ content, which Google stopped showing as a rich result on 7 May 2026 but still parses to understand pages (Search Engine Journal, 2026). And Organisation or entity markup that helps a less-established brand assert who it is. Every one of those is about removing ambiguity, not about talking a model into preferring you.

What to do instead of buying a schema package

Add sensible schema once, as part of getting your facts accurate and consistent everywhere, and then stop thinking of it as a growth channel at all. Take the money you would have handed over for an ongoing "schema for AI" retainer and point it at the things with real evidence behind them. That is where citations are actually won and lost.

In practice that comes down to a few levers. Answer-shaped content that states your facts in plain, visible prose, because that visible text is what AI systems actually read. Genuine reviews and reputation signals. Mentions and citations in the third-party sources these tools draw from. And clean crawler access, so the tools can reach you at all. If you want a structured way to see what AI says about your business today and which fixes are worth paying for, our AI visibility audit guide walks through the method.

Questions people ask

Does schema markup help you get cited by ChatGPT?

Not on its own, based on the only controlled test available. Ahrefs added schema to 1,885 pages and saw ChatGPT citations move about 2.2%, within statistical noise (Ahrefs, 2026). A separate test found ChatGPT reads only visible HTML during live retrieval, ignoring JSON-LD entirely (searchVIU, 2025). Schema helps comprehension, but it does not buy ChatGPT citations by itself.

Is structured data still worth adding for AI search?

Yes, but for the right reasons. Schema is cheap to add and removes ambiguity about your facts, which matters most for local businesses, products with concrete attributes and lesser-known brands. Platform engineers confirm their systems use it to understand content (Search Engine Land, 2025). Just do not expect it to increase citations on its own, and do not pay a premium retainer for it.

Should I still use FAQ schema now that Google stopped showing it?

You can leave it in place. Google stopped displaying FAQ rich results on 7 May 2026, but says it still uses FAQ structured data to understand pages, and other retrieval systems may parse it too (Search Engine Journal, 2026). It will not produce a visible rich result any more, so treat it as comprehension markup rather than a way to win extra space in search.

Sources

• Ahrefs (2026), We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved.
• Search Engine Journal (2026), Schema Markup Didn't Move AI Citations In Ahrefs Test
• Search Engine Land (2025), Microsoft Bing/Copilot use schema for its LLMs
• Search Engine Roundtable (2024), John Mueller On If Schema Helps With LLMs & Google
• Search Engine Land (2026), How schema markup fits into AI search, without the hype
• searchVIU (2025), Schema Markup and AI in 2025: What ChatGPT, Claude, Perplexity & Gemini Really See
• Search Engine Journal (2026), Google Drops FAQ Rich Results From Search