Structured Data for AI Search: What Schema Does, What It Doesn't, and What to Keep

The AI schema plugin, and what it actually adds

Open the settings of any popular SEO plugin and you will find a new tab with AI in the name. Turn it on and it adds a block of JSON-LD to every page: FAQPage on the service pages, a dozen properties on the home page, perhaps a speakable section. The dashboard turns green. Nothing on the page changes for a human reader, which is the first clue about how much changes for a machine.

Structured data is a description of a page written in a vocabulary from schema.org, usually as a JSON-LD script in the HTML. It tells a search engine, in labeled fields, that this page is about a plumbing company at this address with these hours, or a product with this price and stock level. It is useful and old, and it has picked up a new mythology since AI answers arrived. This page separates the two, engine by engine, and links to the GEO section front page for the rest of the mechanics.

Google says no special markup; Microsoft says schema helps

Google's guide to AI features is blunt: structured data is not required for generative AI search, and there is no special schema.org markup to add. Google still recommends structured data as part of SEO, because it makes pages eligible for rich results, and its checklist for AI features asks owners to make sure structured data matches the visible text on the page. Its general guidelines add that markup must not describe content readers cannot see, and that correct markup enables a feature without guaranteeing it.

Microsoft takes a warmer view. Its October 2025 guidance on getting content into AI answers recommends schema markup in JSON-LD, saying it can label content as a product, review, FAQ or event so its AI systems can interpret it with confidence. Copilot runs on Bing's index, so that advice applies to Microsoft Copilot directly.

Neither OpenAI, Perplexity nor Anthropic has published guidance on structured data for their assistants. Several of their crawlers read raw HTML without running JavaScript, so markup that is printed in the HTML is at least visible to them, while markup injected by a script may not be. Whether they use it is unknown; the JavaScript and AI crawlers page covers the rendering half.

May 7, 2026: the day the FAQ rich result went dark

For years FAQPage markup was the cheapest win in SEO: mark up a few questions and Google might show them as expandable answers under your result. In September 2023 Google limited FAQ rich results to well-known, authoritative government and health websites. In the same month it removed its How-to documentation because How-to rich results were no longer shown. Then, starting May 7, 2026, Google stopped showing FAQ rich results in Search entirely and later removed the documentation.

FAQPage is still valid schema.org vocabulary, and the markup still describes a page accurately. It just no longer earns anything visible in Google. What remains valuable is the content itself: real questions in the words customers use, answered in the first sentence, visible on the page where any engine can quote them. That is a writing job, covered in content AI can quote, not a markup job.

Organization markup: a business card for the home page

Organization structured data goes on the home page or a single page that describes the business, such as About. Google says it helps it understand an organization's administrative details and tell it apart from others with similar names, and that some properties can influence the logo shown in results and in the knowledge panel. There are no required properties. You add the ones that apply.

• name and legalName: the trading name customers know, and the registered name if different.
• url and logo: the canonical home page and a logo image at least 112 by 112 pixels.
• address and telephone: the same address and phone number used on every listing, with the country code.
• sameAs: links to the business's profiles on other sites, which ties scattered listings to one entity.
• iso6523Code: an international identifier such as a DUNS number or LEI, useful for banks and larger firms.
• description and foundingDate: a plain sentence about what the business does, and when it started.

For GEO the useful idea is disambiguation. Engines have to decide whether the roofing company named in a forum thread, on a directory and on a website is one business or three. Organization markup with consistent sameAs links is one more vote for one. It is a supporting act, not the lead; the lead is making every listing agree, covered in entity optimization.

LocalBusiness markup: hours, addresses and the most specific subtype

LocalBusiness structured data tells Google about a location: opening hours, departments and more, which can feed the business knowledge panel in Search and Maps. Only two properties are required, name and a full postal address. Google recommends telephone, url, openingHoursSpecification, geo coordinates to at least five decimal places, a priceRange under 100 characters, and department for businesses with separate parts. Restaurants can add menu and servesCuisine.

Google also asks for the most specific subtype possible. schema.org has dozens: Plumber, Electrician, HVACBusiness, RoofingContractor, Dentist, Attorney, AutoRepair and more. A specific type tells an engine what you are before it reads a word of your copy.

Two cautions. First, the markup supports the Google Business Profile; it does not replace it. Google's AI guidance names Business Profile, not markup, as what helps local services appear in AI responses, and Gemini reads place details from Google Maps. Second, review stars. Google's rules only allow LocalBusiness and Organization review markup on sites that review other businesses, so stars marking up a business's own testimonials are not eligible for review snippets. If your plugin does it, turn it off. For listings work, see how AI chooses businesses and Google AI Mode.

Product markup, and the feed that matters more

Product structured data comes in two flavors. Product snippets are for pages where a product is described but cannot be bought, such as a review or a dealer page. Merchant listings are for pages where it can be bought, and can show price, availability, ratings and shipping in results. Either way, the markup must match the price and stock the shopper sees.

For AI shopping answers, though, the feed usually outranks the markup. Google's AI optimization guide says Merchant Center feeds and Business Profiles help products and services appear in AI responses, and Google says its Shopping Graph holds over 60 billion listings. Microsoft says Microsoft Merchant Center feeds inform organic Copilot shopping results. OpenAI's commerce docs say product feeds in ChatGPT are open only to approved partners who apply. Keep page markup and feeds consistent, because a mismatch between them is a reason for an engine to trust neither.

A field guide to the markup worth keeping

Our view, as practice rather than any engine's rule: keep accurate Organization, LocalBusiness and Product markup, print it in the server-rendered HTML, and keep it in step with the visible page and your listings. Skip anything sold as AI-only. Note that a structured data manual action, per Google, costs rich result eligibility rather than rankings, but misleading markup is a bad trade at any price.

Checking your own markup in fifteen minutes

1. 1

   Look at the raw HTML

   Open view-source on your home page and search for application/ld+json. If nothing appears there but a browser extension shows markup, it is being added by script, and crawlers that skip JavaScript will miss it.

2. 2

   Run Google's Rich Results Test

   Paste the URL into the Rich Results Test to see which types Google detects and whether any errors block eligibility.

3. 3

   Run the Schema Markup Validator

   Validate against the full schema.org vocabulary at validator.schema.org, which catches types and properties Google's tool ignores.

4. 4

   Compare markup to the page and the listings

   Hours, phone, address and prices in the JSON-LD should match the visible page, the Business Profile, Apple Business and Bing Places exactly.

5. 5

   Remove what does not belong

   Self-serving review stars, markup for content that is not on the page, and duplicated blocks from two plugins fighting each other.

6. 6

   Recheck after every redesign

   Theme and plugin changes rewrite markup silently. Search Console's enhancement reports flag errors as they appear.

On the sites we build, markup lives in code: Next.js renders it into the HTML with the page, and Iris, the agent in Athena CMS that handles titles, descriptions and structured data, keeps it current as content changes. A well-configured WordPress site can do this too; the WordPress vs Next.js guide is fair about both. For voice assistants, voice search schema covers the overlap.

Want a second pair of eyes? A free GEO audit checks the markup alongside crawlability, listings and the answers themselves, and comes back with a fixed monthly quote. Run the free AI visibility check first to see what the engines say today, and read GEO pricing for what a program costs.