Mostly no — or more precisely, nobody has shown that it does. Schema markup is structured data added to a page so machines can read facts from it without guessing: this is a price, this is an address, this is a review score. It is useful, it is good practice, and it is one of the most commonly sold "AI SEO" deliverables. The evidence that it makes ChatGPT, Perplexity or Google's AI Overviews more likely to cite you is thin to the point of absence.
This post lays out what the evidence actually says, why the claim is so widespread anyway, and what schema is genuinely for.
The short version
The academic paper most often cited as proof that schema helps AI citation tested nine methods. Schema was not one of them.
One independent study looked for a correlation between schema and LLM citations and found none.
Google's John Mueller, asked directly in January 2026, answered "yes, no, and it depends" — with a carve-out only for prices, shipping and availability.
Schema is still worth having, for a different and well-evidenced reason: it makes your facts machine-readable and your pages eligible for Google's rich results.
What is schema markup?
Schema markup — structured data, usually written as JSON-LD — is a block of code on a page that labels its facts in a vocabulary machines share. A human reads "Open 9 to 5, Monday to Friday" and understands it. A machine reading the same sentence has to parse it. Schema says openingHours: Mo-Fr 09:00-17:00 and the ambiguity is gone.
Google has used this for years to power rich results: star ratings under a listing, FAQ dropdowns, event dates, product prices. That use is real and well documented. The question is whether the same markup does anything for the AI systems that now compose answers.
What does the research actually say?
Three pieces of evidence, in order of how often they are misquoted.
The Princeton and IIT Delhi GEO paper. Researchers (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande) built a benchmark of 10,000 queries and tested nine ways of changing a page to make a generative engine more likely to feature it. The methods were: authoritative tone, adding statistics, keyword stuffing, citing sources, adding quotations, simplifying language, improving fluency, using unique words, and adding technical terms. Structured data is not on the list. The paper does not test schema, mention schema, or draw any conclusion about schema. It is nonetheless the single most common citation offered as proof that schema improves AI visibility. If a proposal cites it for that purpose, the author has not read it — we cover what the paper did find in our post on answer engine optimization.
The Search Atlas study. An independent analysis looked for a correlation between the presence of schema markup and being cited by large language models. It found none.
Google's own position. John Mueller, speaking in a personal capacity in January 2026, summarised whether structured data helps with AI systems as "yes, no, and it depends." The one area he carved out as plausibly useful was commerce facts — pricing, shipping and availability — where a machine reading a product page benefits from unambiguous fields. That is a real but narrow endorsement, and it is about accuracy, not visibility.
The only first-party confirmation that schema feeds an AI answer system at all came from Bing (Fabrice Canel, March 2025), and it was specific to Bing's own pipeline.
That is the sum of it. No peer-reviewed evidence, one null result, one "it depends", one narrow confirmation from a single engine.
Why do so many agencies sell it anyway?
Three reasons, none of them dishonest in origin.
It is concrete. Schema is a deliverable you can point to: here is the code, here is the validator showing it passes. Most AI-visibility work — earning mentions, rewriting copy in the buyer's vocabulary, building review flow — is slower and harder to show in a screenshot.
It used to be the frontier. Rich results were a genuine ranking-adjacent win for a decade. The habit of selling schema as the technical edge predates AI answers, and the pitch was updated faster than the evidence.
The misattribution is self-reinforcing. Once a few widely-read articles cited the GEO paper for schema, the claim entered the aggregator posts — "47 AI SEO statistics for 2026" — and from there into agency decks. Nobody in the chain went back to the paper.
The result is a deliverable that is sold on the weakest evidence in the category, often as the centrepiece of an "AI SEO" package.
How do AI systems actually read your page?
This is the part that should reset the question.
Vercel's analysis of roughly a billion crawler requests found that none of the major AI crawlers execute JavaScript. They read the HTML your server sends and nothing more. Schema injected by a script — which is how a great many sites add it — is invisible to them. If your structured data only exists after JavaScript runs, the AI crawlers never saw it, whatever it might have done.
And when an AI agent visits a page on a user's behalf, it often does not render it as a browser would at all. Reporting from GA4 property data found roughly 46% of ChatGPT agent visits render in reading mode — the text, with no CSS, no JavaScript and no schema.
So the systems in question frequently cannot see the markup, and when they can, nobody has shown it changes what they say. That is two separate reasons to be sceptical of the pitch, and the first one is a technical fact rather than a research gap.
So what is schema actually for?
Three things, all real.
Rich results on Google. Ratings, prices, FAQs, events, breadcrumbs. This is the documented, long-standing benefit and it has not gone away.
Unambiguous facts. For a business, the fields that matter are name, address, phone, opening hours, services and prices. Marked up once, correctly, they stop a machine — any machine — from having to infer them. Mueller's carve-out for commerce facts is this point.
Hygiene. A site whose structured data agrees with its visible text, its Business Profile and its directory listings is a site that is easy to corroborate. Corroboration is what the brand-mention research suggests AI systems are actually doing.
Notice what is missing from that list: "gets you cited by ChatGPT." Schema is a sound part of technical SEO. It is not an AI-visibility lever, and any proposal that prices it as one is pricing a hope.
What should you buy instead?
If the goal is to appear in AI answers, the work with evidence behind it is elsewhere.
Make the page readable without JavaScript. Server-rendered content, so the crawlers that build these answers can see it at all. This is the single most under-sold technical fix in the category.
Answer the question in the first sentences, with sources and numbers. The GEO paper's best-performing methods were citing sources, adding quotations and writing fluently — roughly a 27–28% lift on its visibility metric. Keyword stuffing made things worse.
Get mentioned. Ahrefs' 75,000-brand study found branded web mentions correlate about three times more strongly with AI Overview visibility than backlinks do.
Keep schema for what it does. Add it for rich results and for your core business facts, in the server-rendered HTML, kept in sync with the visible page. Then stop paying for it as an AI strategy.
Been sold "schema for LLMs" and not sure what you got?
Our written audit checks what your structured data actually does, whether the AI crawlers can even see it, and what is genuinely keeping you out of AI answers — the six areas are in our post on what a real SEO audit tells you. It's free, and it's a document, not a sales call.
This article is written by
Founder & Lead Strategist at CognitiveAISEO. Marvin Blanco has spent 20 years in digital marketing, helping businesses grow through search. He founded CognitiveAISEO to make sure small and mid-sized brands aren't left behind as AI reshapes how people find information.
Follow on
