Schema Markup for AI: What the Evidence Shows and What It's For

Schema markup for AI is not a requirement, and no published test shows it earns AI citations. Google says structured data "isn't required for generative AI search," and the largest controlled test we found, 1,885 pages tracked by Ahrefs, saw no meaningful lift on any platform it measured; the one small test that found a lift cannot separate schema from keyword or indexing.

That does not make schema useless. It still feeds the Google rich results that exist, and one schema check has a plausible path into AI answers: every fact in your JSON-LD must also appear in your visible text. This guide covers what Google says, what the test measured and what it left open, which schema types still do something as of October 2026, and how to run that check.

Does schema markup help AI search? The short answer by platform

Here is what each operator documents about structured data, next to what has actually been measured. "Not measured" means nobody has published a controlled test. It does not mean the answer is no.

We read the OpenAI, Perplexity and Anthropic pages in October 2026. None of them gives structured data any role in how a page is found, retrieved or cited.

What Google says about structured data and AI, verbatim

Google's guide to optimizing for generative AI features (last updated 2026-07-10) files structured data under things people overfocus on:

"Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add. However, it's a good idea to continue using it as part of your overall SEO strategy, as it helps with being eligible for rich results on Google Search."

Google's AI features documentation 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," with no additional technical requirements. The same page lists one schema rule among ordinary SEO fundamentals: "Making sure your structured data matches the visible text on the page." Hold on to that line, because it is the only schema rule with a plausible AI consequence.

Google is an interested party, and these are statements, not measurements.

They are the most specific first-party statements on the subject: there is no AI-specific schema type, and markup is not a requirement.

Microsoft runs Bing and Copilot and leans the other way. Its Microsoft Advertising blog post of October 8, 2025 says schema "helps search engines and AI systems understand your content" and advises "Use schema, clear headings, and modular layouts." It publishes no test or data to back that up.

The one controlled test: Ahrefs, 1,885 pages

Ahrefs published the largest matched-control study of schema and AI citations we found on May 11, 2026, under the headline "AI Citations Barely Moved." It tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, matched them against about 4,000 control pages, and compared citations in the 30 days before and after each change.

Ahrefs is careful about the one significant result: the drop is small, and they say it cannot be cleanly attributed to schema. The fair reading is "no lift," not "schema hurts you."

What the test did not answer

Ahrefs disclosed its own boundary: "Every page in the dataset had 100+ AI Overview citations in February 2025, before any schema was added." So the study answers whether schema helps an already heavily cited page get cited more. Whether schema helps an uncited page enter the set of sources an AI considers is untested, not refuted.

Ahrefs states more limits. It pooled every schema type together (Article, FAQ, Product, HowTo, Organization and others) and tested JSON-LD only, not Microdata or RDFa. It measured only the 30 days after each change, so a slower effect is untested, and it looked only at schema in the page HTML, not schema added by JavaScript.

Pages that add schema often change other things at the same time, too.

If you are a mid-market brand with zero AI citations, nobody has run your experiment yet, and that gap is not evidence in your favor either.

Other evidence you will see cited

The most-shared counterexample is a Search Engine Land experiment from September 2025: three single-page sites, one with well-built schema, one with deliberately broken schema, one with none. Only the well-built page appeared in an AI Overview.

Read the setup before the result. Each page targeted a different keyword, so they never competed for the same answer, and the no-schema page was never indexed, so it could not have appeared anywhere. One page per arm cannot separate schema from keyword, indexing and timing; it is a reason to run a bigger test, not a finding.

The wider literature has the same shape: the July 2026 critical survey of 45 GEO studies concludes that "already-retrieved content can causally alter its citation or use, but no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability." Tests that hand a page to a model and ask questions about it measure the first thing. Getting found is the second thing.

Why schema is still sold as an AI lever

Four reasons, none of them evidence:

• It is easy to generate and easy to check. A tool can emit JSON-LD in a second and detect it in a second. That makes it a perfect audit line item, which is a fact about audit tools, not about AI.
• It borrows from a real mechanism. Schema did earn rich results in classic Google Search for years. The industry carried that intuition over to AI answers without checking whether the pipeline was the same.
• The incentive is structural. Most GEO frameworks are published by companies that sell the fix the framework recommends. That includes us, which is why we treat our own on-page advice with the harshest scrutiny.
• The story is intuitive. "Without schema, AI has to guess" reads well. Microsoft says schema helps AI systems "understand your content," but neither it nor any other AI vendor has published a test, so the causal chain has no measured first link.

What schema is still good for in 2026

Schema earns its place in Google Search, not in AI answers. As of October 2026, Google's structured data gallery lists 25 supported features, and FAQ and HowTo are no longer among them.

Sources for the two retirements: Google's Search documentation changelog (FAQ) and its 2023 HowTo and FAQ announcement, which also says "structured data that's not being used does not cause problems for Search." So existing FAQ markup can stay; adding new markup buys nothing visible.

Organization schema, done honestly

The honest job of Organization markup is disambiguation: if your brand name is also a common word or another company's name, this is where you tell Google which one you are. Google lists no required properties; it recommends starting with name, url and logo, plus sameAs links to your real profiles. Every field must match your visible About page.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Example Analytics",
  "url": "https://www.example.com",
  "logo": "https://www.example.com/logo.png",
  "foundingDate": "2019",
  "sameAs": [
    "https://www.linkedin.com/company/example-analytics",
    "https://github.com/example-analytics"
  ]
}
</script>

Product schema, with the price that matches the page

Product schema still powers product snippets and merchant listings, which is a real reason to have it. The AI-relevant part is that the price in the Offer must be the exact price a crawler reads in your page text.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Example Analytics Team Plan",
  "offers": {
    "@type": "Offer",
    "price": "119.00",
    "priceCurrency": "USD",
    "availability": "https://schema.org/InStock"
  }
}
</script>

The one AI-relevant schema check: parity

Google's structured data policies have always required markup to reflect the page: "Don't mark up content that is not visible to readers of the page," and "Your structured data must be a true representation of the page content." That rule is also the only schema check with a plausible AI mechanism, because it catches two failures that can plausibly reach AI answers.

1. A fact that lives only in the schema. Google says it does not need the markup for AI features, and Microsoft's claim that schema helps AI systems comes with no data. Treat a price, rating or founding year that exists only in JSON-LD as missing for any crawler that reads only visible text.
2. A fact that disagrees with the page. That is a published contradiction. One system reads the number in your text, another reads the number in your markup, and you do not get to choose which one an answer repeats.

There is a third, quieter failure. If your visible price is injected by JavaScript after the page loads, the raw HTML may carry the schema price and no visible price at all. Vercel's December 2024 crawler study found ChatGPT's and Claude's crawlers fetch JavaScript files but "don't execute them," so for those crawlers the page text and the markup can say different things.

A worked example

A pricing page carries this markup:

{
  "@type": "Product",
  "name": "Team Plan",
  "aggregateRating": { "@type": "AggregateRating", "ratingValue": "4.8", "reviewCount": "120" },
  "offers": { "@type": "Offer", "price": "119.00", "priceCurrency": "USD", "priceValidUntil": "2026-12-31" }
}

Check each fact against the raw HTML with JavaScript off:

Two of the three facts exist only in the markup, and the fix is not more schema: put the facts on the page or delete them from the markup.

Now imagine the table read "$99/mo" while the schema said 119.00. That is the kind of contradiction behind "ChatGPT gets our pricing wrong," and it is covered in our guide to pricing page GEO.

How to run a schema parity check

You need a browser and nothing else. Do it once per page template, not once per page.

1. Open the raw HTML. Press Ctrl+U (Cmd+Option+U on Mac) to view source. This is the page before JavaScript runs, which is what ChatGPT's and Claude's crawlers read.
2. Find every JSON-LD block. Search the source for application/ld+json. Themes and SEO plugins often each add one, so expect more than one block.
3. List every fact. Names, prices, ratings, review counts, dates, addresses, people. Ignore structural fields like @type.
4. Search for each value as visible text. If "119" appears only inside the script block, it is schema-only.
5. Fix at the source. Put schema-only facts on the page or remove them. For mismatches, make the markup and the page read from the same CMS field so they cannot drift.
6. Validate the syntax. Run the page through Google's Rich Results Test and the Schema Markup Validator. One missing comma breaks a whole block.

If you prefer the command line, this pulls the schema price and the visible price side by side (replace the URL with yours):

curl -s https://www.example.com/pricing | grep -o '"price": *"[^"]*"'
curl -s https://www.example.com/pricing | grep -oE '[$][0-9]+/mo'

Our free schema markup generator builds JSON-LD for the Google rich results that still exist and checks each fact in it against your visible text. For a slower walk through schema types, see the schema markup guide in our academy.

Common schema mistakes

1. Stale data. Schema with last year's pricing is worse than no schema, because it publishes a second, wrong version of a fact.
2. Facts that live only in the markup. Ratings and review counts are the usual culprits, and Google's guidelines require them to be visible.
3. Competing blocks. A theme and an SEO plugin each emit an Organization block with a different name or logo. Pick one source.
4. Adding FAQPage or HowTo in 2026. Both rich results are gone. If you want questions answered on the page, write them as text; our post on FAQ page SEO covers that.
5. Adding schema to move an AI number. The best available test found no lift. Add it for the rich results that exist, and for nothing else.

What to do instead for AI visibility

The levers with a documented mechanism are about access, and they are subtractive: remove what blocks you. Each has a vendor statement behind it.

• Allow the retrieval crawlers. OpenAI says sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers" (OpenAI crawler documentation). Check robots.txt per bot, not with one blanket rule.
• Check your CDN, not only robots.txt. OpenAI's ChatGPT search help page makes eligibility a conjunction: allow OAI-SearchBot "and confirm that the website host or content delivery network allows traffic" from its published IP addresses.
• Remove accidental snippet blocks. Google's robots meta tag documentation says nosnippet and max-snippet apply to AI Overviews and AI Mode too.
• Make your facts agree everywhere. Pricing page, About page, footer, schema and third-party profiles should all state the same number.

Why these four come first, stage by stage, is in our post on how LLMs prioritize information about brands.

Frequently asked questions

Does schema markup help AI search?

Not measurably. Google says structured data isn't required for its generative AI features, and Ahrefs' matched-control test of 1,885 pages found no lift on AI Overviews, AI Mode or ChatGPT. Whether schema helps a page with zero citations get its first one is untested, not refuted.

Do ChatGPT and Perplexity read JSON-LD?

Neither documents it. The markup sits in the HTML their crawlers fetch, but OpenAI's and Perplexity's crawler and search documentation never mention structured data. Put any fact you want AI to repeat in your visible text.

Should I remove FAQ schema now that FAQ rich results are gone?

There is no need. Google stopped showing FAQ rich results on May 7, 2026, and its 2023 guidance says structured data that is not being used "does not cause problems for Search." Keep the questions and answers as visible text and stop adding new FAQPage markup.

Is structured data for AI different from schema for SEO?

No. There is no AI-specific schema type, per Google, and no other AI vendor documents one. "Structured data for AI" is the same schema.org markup, and its only proven job is Google rich results and disambiguation.

What schema should I add if I care about AI visibility?

Organization on the homepage and Product with Offer on product or plan pages, because both still do something in Google Search. Then run the parity check so every fact in the markup also appears on the page.

To see whether AI crawlers can reach the pages your schema describes, run a free 99Visibility audit, which checks robots.txt and firewall access for 18 AI and search crawlers.