Does schema markup help with AI search? Mostly no, and that’s a tested answer now rather than a hunch. Ahrefs added JSON-LD to 1,885 pages, measured their citations against matched controls, and found no major uplift in citations on any platform.
Google’s own optimisation guide says structured data isn’t required for generative AI search, and there’s no special markup to add for it. Schema still earns its place, because it makes your facts machine-readable and keeps a page eligible for rich results in classic Search.
What does move AI citations sits on the visible page, and our studies put numbers on it: answer chunks a model can lift, freshness with a date people can actually see, and real refresh work. If AI search visibility is your job, that’s where the effort goes.
What’s the difference between structured data and schema markup?
Structured data is the general idea: facts on a page labelled so a machine can read them without guessing. Schema markup is how you usually produce it, a shared vocabulary from schema.org written into the page.
The schema.org docs describe it as “a collection of shared vocabularies webmasters can use to mark up their pages in ways that can be understood by the major search engines: Google, Microsoft, Yandex and Yahoo!”. You write that vocabulary into a format, usually JSON-LD, a script block that lists the facts in one place, or Microdata or RDFa.
Let’s say your page shows “Updated 14 March 2026” at the top. Structured data is the idea of putting that fact somewhere a crawler can parse it. The schema.org dateModified property, written in JSON-LD, is the markup.
People also search for “structured schema”, which is a pretty loose phrase for the same question, and there’s no separate thing behind it.
Does schema markup help with AI search?
The tested answer is no major lift. The cleanest experiment on this so far, run by Ahrefs, added schema to real pages and measured what happened to their citations.
They took 1,885 pages that added JSON-LD between August 2025 and March 2026, matched each one with three similar pages that never added schema, and counted citations for 30 days before and after the change. These weren’t obscure pages: each already had 100 or more AI Overview citations as of February 2025. The result, in their words: “Adding schema produced no major uplift in citations on any platform.”
A few details matter. AI Overviews citations fell 4.6%, which was small but statistically significant, roughly a 1 in 2,500 chance of seeing it by luck.
AI Mode rose 2.4% and ChatGPT rose 2.2%, both statistically indistinguishable from zero. Their practical read: “If you’re already doing the rest of the SEO work well, JSON-LD isn’t going to be the unlock.”
The study has honest limits. It only covered pages AI already cited, so if you’re starting from pages the engines haven’t discovered yet, schema might still matter there. The schema types were pooled, and Microdata, RDFa and JavaScript-injected schema weren’t tested.
Ahrefs’ earlier look at 6 million URLs found AI-cited pages were nearly three times more likely to carry JSON-LD, but they put that down to better-maintained, higher-authority sites, and a correlation isn’t a cause.
What Google and OpenAI’s docs say
Google doesn’t require structured data for AI Overviews or AI Mode, and OpenAI’s documentation says nothing about markup either way.
Google’s generative AI optimisation guide says it straight: “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” Their page on AI features and your website repeats it: “You don’t need to create new machine readable files, AI text files, or markup to appear in these features,” and “There are no additional requirements to appear in AI Overviews or AI Mode.”
Eligibility for those features is about the basics: your page being indexed, and eligible to show a snippet. Structured data does appear in the recommended practices on that page, in one form: making sure it matches the visible text on the page.
OpenAI’s bots documentation describes GPTBot, which crawls content that might be used for training, and OAI-SearchBot, which surfaces websites in ChatGPT’s search results. If you block OAI-SearchBot, your site doesn’t show up in ChatGPT’s search answers. The page says nothing about schema, either way.
That’s worth stating plainly: no provider documents a schema-specific signal for AI answers.
Where schema does help
Structured data is good for attribution and parsing, and that’s still real work in AI search. It puts your name, dates and entity facts in a form a machine can lift without guessing, which helps any system that wants to know who published something and when. It also helps a page’s eligibility for rich results in classic Google Search, which is what Google’s structured data documentation is framed around.
But none of that is a measured citation lift. The test above looked for one and didn’t find it, and the providers’ docs don’t claim one either.
What actually moves AI citations
What increases AI citations, if not schema markup, is what’s visible on the page, and we’ve measured it in our own studies.
The first is the answer chunk. AI scores one chunk of a page at a time, which is the finding of our 3 numbers study: chunk frequency, chunk score and extractable units decide whether there’s something to quote. A generative engine optimisation benchmark, the GEO paper by Aggarwal et al., found much the same, with quotations raising visibility in AI answers by 40.9%.
The second is freshness with a visible date. Our title tag study ran 1,000 searches across five AI engines, with findings re-tested on 40 random half-splits. Google showed a page’s fresh date 87% of the time when the date was visible, and only 40% of the time when you kept it just in the code.
Pages showing a date from the last year got 186% more AI Overview citations than pages showing a date more than two years old. And this is the direct schema comparison: dateModified in your markup is the code-only route, and the code-only route is the weaker one in our data.
The third is real refresh work. Our 10x AI Visibility Refresh study refreshed 98 posts, published nothing new, and saw citations rise 10x over 91 days. The same study found the correlation between Google rank and AI citation was basically zero, Spearman rho 0.000 with p = 0.914, so classic rank didn’t explain the lift.
The fourth is the year in the title. Adding the current year to a title tag raised Claude citations by as much as 192% in the same title tag study, and only the current year counts: 2026 titles got 252% more Claude citations than titles carrying an old year. Google’s version of the gain didn’t survive re-testing, holding in 0 of 40 random re-runs.
All of this is correlational, and the study says so itself: “This shows patterns, so it can’t prove cause and effect.” Its warning lands here too: “Just changing the year on a page you haven’t touched won’t do much for you on Google.” The freshness that counts is real freshness, visible on your page.
What to implement, in what order
Which schema markup you should implement first matters less than keeping it honest, so here’s the order I use:
- Run the pages that matter through Google’s Rich Results Test. Then run them through the Schema Markup Validator, and fix what’s broken.
- Mark up only what matches visible content: Organization for the site, Article or Product for the page, with the dates. Set dateModified only alongside a date people can see.
- Don’t build schema-only facts. Anything in the markup that a person can’t see on the page is hidden data, and hidden data is the weaker route in our freshness data anyway.
- Put the time you save into extractable answers under the buyer questions you actually want to win, and a refresh cycle you actually run. Every measured lift above came from there.
How to tell whether it worked
Here’s how to measure schema changes in AI answers: you can’t rely on one search, because the same question gets different answers run to run. So a real before-and-after needs the same buyer questions asked repeatedly over weeks, a few questions you deliberately leave untouched as a comparison, and per-page evidence of whether the answer chunk clears what retrievers measure. It also helps to see the searches each assistant ran before answering, and the pages it read without citing, because those are the closest wins.
You can check a few pages by hand. Past that, Omnipresence tracks whether ChatGPT, Claude, Gemini and Perplexity cite or recommend a page after the change, shows the searches each assistant ran and the pages it read but didn’t cite, keeps comparison questions untouched so the lift is provably your work. It also scores whether the answer chunk clears what retrievers measure. It tracks those four assistants and no others, and you pay the model cost on your own key.
FAQ
What is structured schema markup?
It’s the same thing as schema markup, typed loosely: the schema.org vocabulary, usually written as JSON-LD, that turns visible page facts into structured data. There’s no separate standard behind the phrase.
Do ChatGPT and Perplexity read schema markup?
No provider documents a schema-specific signal in their answers. OpenAI’s page for its crawlers says nothing about markup either way, and Ahrefs’ measured test found ChatGPT’s citation change after adding schema was statistically indistinguishable from zero. Treat the visible text as what counts, because that’s the part anyone has measured.
Is FAQ schema worth adding for AI citations?
No measured lift supports it. The Ahrefs test pooled schema types, FAQ markup included, and found no major uplift in citations on any platform. Add it where it matches visible content, which is Google’s own recommended practice, and skip it as a citation tactic.
Set the markup that matches your page, and spend the rest of your effort on the page itself.