Schema Markup for AI: The Types That Actually Get Used

Schema markup for AI works differently than it does for classic search. Google says so directly: no special AI schema markup exists, and structured data for AI search still runs through the same Organization, Article, and Product types that already feed rich results, Bing Copilot, and AI Overviews. Here’s what still earns its keep, what got quietly retired, and where a script tag stops mattering once an engine skips JSON-LD altogether.

The schema types with real backing

1. Organization

Organization schema is the anchor. Name, url, logo, sameAs, and contact details tell Google which entity owns a site, and that record feeds straight into the Knowledge Graph. Yogoo AI’s GEO Site Audit skill checks this block first, because a site with no clear Organization markup reads as ambiguous, both to Google and to a person skimming the About page.

2. Article

Blog posts, guides, and news pieces use Article schema to spell out a headline, an author, and a publish date. That’s what makes a page eligible for the Article rich result, and it hands any engine reading the page a clean answer to who wrote it and when.

3. Product, with Offer and Review nested in

Product schema carries price, availability, and ratings. That’s what lets a listing show up in Merchant Center feeds, which Google’s own generative AI guidance names directly as a source for product information inside AI responses. Skip this one only if nothing on the site is actually for sale.

4. Review snippet and AggregateRating

A star rating pulled into a search result is still one of the highest-trust signals a page can carry. Google only shows it paired with specific types: Book, Course List, Event, Local business, Movie, Product, Recipe, or Software app, not on its own, so pair it with whichever of those the page already has.

5. Breadcrumb

Breadcrumb markup is cheap to add and easy to keep accurate. It tells any system reading the page exactly where that page sits in the site, and there’s no real reason to skip something this cheap.

6. Q&A

Q&A schema isn’t the same thing as a marketing team’s FAQ list. It’s for a page built around one question and multiple community answers, like a support forum thread, not a set of questions someone wrote to target keywords.

7. Local business

Local business schema feeds hours, address, and booking actions into the knowledge panel and local search results. It only matters if the business has a physical location or a service area, but nothing else stands in for it when it does apply.

What’s not worth the code anymore

Google stopped showing FAQ rich results in Search in May 2026. Wrapping a list of questions in schema markup doesn’t earn a page anything there anymore. Keep the questions and direct answers on the page itself, in plain headings. That structure still helps a reader, and it still gets pulled into a passage an engine can cite. The wrapper is what’s gone, not the format.

Standalone HowTo dropped out of Google’s structured data gallery back in 2023. Step-by-step instructions still belong on the page. They just don’t need a HowTo wrapper to earn a place in search anymore.

And llms.txt: Google says plainly that its search features, generative ones included, need no new machine-readable files or special markup. That doesn’t make llms.txt useless everywhere, since some other tools do read it, but it’s not a lever Google pulls, and it’s not doing the same job as schema.

JSON-LD for AI crawlers: format and placement

JSON-LD is the format Google recommends over Microdata and RDFa. It’s a single script tag, and it doesn’t have to get woven into the visible HTML the way the other two do. It sits in the head or the body as <script type="application/ld+json">, and whatever’s inside it has to match what a visitor actually sees on the page. Don’t build a page just to hold markup, and don’t describe something in schema that isn’t visible in the content.

Whether that script tag gets read at all depends on who’s asking. Google and Bing render JavaScript, so they can read JSON-LD even when it’s injected client-side. ChatGPT and Claude run their own crawlers, and those typically fetch a page without rendering JavaScript, so markup added only through a script never reaches them at crawl time. Perplexity runs its own live retrieval and leans hard on Reddit, YouTube, and LinkedIn for what it cites, so its access to a page’s schema depends on whether those upstream sources rendered it first. If a site depends on JavaScript to inject its content or its schema, checking whether ChatGPT can read the website at all matters more right now than any markup fix.

Does adding schema actually move AI citations?

The honest answer is mostly no, and the evidence is specific enough to say why. Microsoft’s Fabrice Canel said at SMX Munich in March 2025 that schema markup helps Microsoft’s LLMs understand content. The clearest test of that question comes from Ahrefs.

Ahrefs tracked 1,885 pages that added JSON-LD schema against a matched control group, across Google AI Overviews, Google AI Mode, and ChatGPT, from August 2025 to March 2026. AI Mode and ChatGPT citations moved by about 2%, too small to call a real effect. AI Overviews citations dropped 4.6%, and Ahrefs couldn’t tie that to the schema itself either — the decline showed up on control pages too, before schema was even added. Every page in that study already had over 100 AI Overview citations before the test started, so none of it says what schema does for a page no engine has noticed yet.

Schema still earns Organization and Product their place in the Knowledge Graph and in rich results. It just isn’t the lever that gets a page cited by an engine that wasn’t already paying attention.

What schema markup helps AI search engines?

Organization data carries the clearest link to how search engines, including the AI-driven ones built on the same index, currently understand a page. Article and Product back it up, and those three are what most audits, including Yogoo AI’s own, check first.

Does schema markup help AI understand content?

It helps identify what a page is and who’s behind it. It doesn’t guarantee a page gets extracted or quoted. Treat it as a comprehension aid a system can fall back on, not a shortcut to a citation.

What schema markup do I need for AI Overviews?

None of it is required. Google states this directly: there’s no special AI schema markup made for AI Overviews or AI Mode, and the same structured data and helpful content that ranks in ordinary search is what shows up there too. Keep the types that are already supported, Organization and Product especially, and treat rich-result eligibility as the actual reward.

None of this replaces measurement. A page can carry every schema type on this list and still never get named in an answer, and that’s a citation problem, not a markup problem. Yogoo AI’s GEO Site Audit skill checks whether the Organization and Product blocks are complete. The Yogoo Score covers the other half: out of the AI answers collected for a business’s market, how many name or link it.