Schema Markup for AI: What It Still Does

Almost every GEO checklist puts schema markup near the top. Add FAQPage. Add Organization. Add Product. Help AI "understand" your content. It is the most commonly sold recommendation in the category, and it has more direct evidence against it than any other.

That does not make schema worthless. It still earns some rich results in classic Google Search, and it is one more place where your company states its facts. That second point is the one this lesson cares about. Markup that says something different from the page is a contradiction you published yourself, in a file nobody on your team reads.

This lesson covers what Google says, what the one controlled test found and where it stops, what happened to FAQ markup, and the parity check that is worth your time.

What Google says about schema and AI

Google's guidance on its AI features is unusually direct. Its AI optimization guide (updated July 10, 2026) says: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." Its AI features documentation (December 2025) sets the bar for AI Overviews and AI Mode: a page "must be indexed and eligible to be shown in Google Search with a snippet," and "there are no additional technical requirements."

Treat these as what they are: statements by the operator, not measurements. Google has its own reasons to make them, and nobody outside can verify what its systems do. But they are the only first-party statements on the question, and the one controlled study points the same way. No AI-specific schema type exists, and no engine has published evidence that adding schema raises citations.

What the one controlled test found

Most schema studies compare sites that have schema with sites that do not. That mostly measures which sites have the budget and the developers to add it. Ahrefs ran something better: a matched-control test that followed pages before and after schema was added and compared them with similar pages that did not change.

The study has a boundary condition, and it matters for most readers of this course. Ahrefs disclosed that every page in the dataset had 100 or more AI Overview citations in February 2025, before any schema was added. It studied pages engines already cited heavily. Whether schema helps a page enter the set of sources an engine considers in the first place has not been tested by anyone. That is untested, not refuted.

If your brand is cited rarely or never, you are exactly the population the study left out. The honest position has two halves: nobody knows whether schema helps a page get in, and the evidence on pages already in says it does not get them cited more. Neither half is a reason to sell schema as an AI lever. Together they are a reason not to claim certainty in either direction.

What schema still does

Schema has two honest jobs left. Neither is AI visibility work, and you should not report either one as such.

Rich results in classic Search

Google still supports rich results for a number of structured data types, such as product snippets. The list changes, so check Google's Search Gallery for what is currently supported before you build anything. Do not trust a list in a blog post, including this one. If a type no longer produces a rich result, adding it earns nothing in Google.

One more consistent statement of who you are

Organization markup states your name, URL, logo and the profiles that belong to you (sameAs) in one machine-readable block. That helps Google tell your company apart from others with a similar name. Its value for this module is consistency, not ranking: it is one more place where your facts appear, so it has to match the About page, the facts block from How to Create Content AI Can Quote and your third-party profiles.

FAQPage markup: retired

FAQ markup was the most-sold schema tactic for AI. Google retired it in 2026:

• Google's Search changelog marked FAQ rich results as deprecated in an entry dated May 8, 2026, effective May 7.
• The FAQPage documentation was removed on June 15, 2026. The old documentation URL now redirects to the changelog.
• Search Console API support for it was removed in August 2026.

Be precise about what this means. It ends FAQ markup as a Google rich result. It does not make questions and answers on a page bad; a plain-text answer to a question buyers ask is still useful to buyers. And Google says leaving existing FAQ markup in place causes no harm, so ripping it out is as unsupported as adding more. Leave accurate markup where it is, stop adding new markup, and check that whatever remains says the same thing as the visible answers.

Parity: the check that matters

Schema parity means every fact in your JSON-LD appears, with the same value, in the visible text of the same page. A parity failure takes one of two forms:

• A fact that exists only in the markup. No engine has published evidence that it uses a fact found only inside a script tag, and most AI crawlers read your text. If the fact matters, it belongs on the page.
• A fact that disagrees with the page. This is the contradiction problem from the previous lesson, hidden in a file nobody reviews. Schema is often written once by a developer or a plugin and never touched again, while prices, addresses and leadership change around it.

Here is a real-looking example from a fictional pricing page. The visible table says "Growth: $79 per month". The page source contains this block, hardcoded into the template:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Ledgerly Growth",
  "description": "Invoicing software for B2B teams, with automatic payment reminders.",
  "brand": { "@type": "Brand", "name": "Ledgerly" },
  "offers": {
    "@type": "Offer",
    "price": "59.00",
    "priceCurrency": "USD"
  }
}
</script>

How to run a parity check

1. Open the raw source. Use view-source or a plain fetch, not the browser inspector. The inspector shows the page after JavaScript ran, which is not what most AI crawlers receive. If a tag manager injects your JSON-LD, it is missing from the raw HTML, and only crawlers that render JavaScript, such as Googlebot, will ever see it.
2. Find every block. Search for application/ld+json. Pages often carry several: one from the theme, one from an SEO plugin and one somebody added by hand.
3. List the facts. Name, description, price and currency, founding date, address, founders and leaders, and every sameAs profile link.
4. Search the visible text for each value. Mark each one present, missing or different.
5. Fix at the source. A different value gets corrected wherever the truth lives, usually your facts block. A missing fact either goes into the visible text or comes out of the markup.

Google's Rich Results Test and the Schema.org validator check that your markup parses and, for Google, whether it qualifies for a rich result. Neither checks whether it is true. 99Visibility's audit recommendations include a schema parity check on your homepage, which lists facts in your JSON-LD that it cannot find in the visible text. For other pages, the manual check above takes about ten minutes. If you need to rebuild a block, the free Schema Generator produces Organization and Product JSON-LD; fill it from your facts block, not from memory.

Organization schema: consistency, not a lever

If you keep one block, keep Organization on your homepage or About page, and treat it as a copy of your facts block rather than a separate source of truth. A few rules make it hard to get wrong:

• Generate it from the same data that renders your visible company facts, so a change to one is a change to both.
• Include only fields you can keep current. An exact employee count goes stale within a quarter; leave it out, or match the range on your About page.
• List in sameAs only profiles you control and keep up to date. A link to an abandoned profile with an old description tells anyone who follows it the wrong thing.
• Keep one Organization block per page. Two blocks from two plugins, with two descriptions, is a contradiction inside a single page.

Common mistakes

• Adding schema to fix a hallucination. If an engine states a wrong price, look for the page, or the block of markup, that states it. More markup does not outvote a contradiction.
• Adding FAQPage markup in 2026. Google retired the rich result. It earns nothing in Search, and no AI engine documents reading it.
• Treating a passing validator as done. Valid markup with the wrong price is still the wrong price.
• Checking the rendered page instead of the raw HTML. You will see markup and text that most AI crawlers never receive.
• Ripping out all schema in a panic. There is no evidence for that either. Keep what is accurate and still earns a rich result.
• Buying "AI schema optimization". Any package that sells markup as an AI citation lever is selling the claim the best available test contradicts.