Share of answer is the operating metric for this practice.

How we count it
Request an audit

Insight · 13 min

What schema does and does not do for AI search

Structured data does not get a page into AI Overviews or AI Mode, and it is not a shortcut to answer engine or generative engine optimization. Google says no special schema.org markup is required, and the only technical requirement is a page that is indexed and eligible to show with a snippet. Schema is still worth doing: it earns the rich results that remain, it tells machines which entity a page is about, and it only works when it restates what the page already says.

In short

  • Google’s guidance for its AI features, as of November 2026: no special markup, no AI text files, no additional technical requirements.
  • Structured data makes a page eligible for rich results. Google says it does not make a page rank better.
  • Google has retired a long list of rich results since 2023. FAQ rich results stopped appearing on May 7, 2026, and HowTo was removed in September 2023.
  • FAQ schema now earns nothing visible in Google, and leaving valid markup in place is harmless. Keep it only where it restates real, visible questions. The studies linking it to ChatGPT citations conflict and are correlational.
  • Microsoft says schema helps search engines and AI systems understand content. OpenAI and Perplexity publish crawler rules, not schema guidance.
  • The bigger lever is entity consistency: one name, one set of profiles, one set of facts, repeated across the site and the web.

What is structured data?

Structured data is a machine-readable description of a page, written in a shared vocabulary, so a search engine does not have to infer what the page is about. Google defines it as “a standardized format for providing information about a page and classifying the page content.”

The vocabulary is schema.org, a community project founded by Google, Microsoft, Yahoo, and Yandex. It defines types such as Organization, Article, Product, and LocalBusiness, and properties such as name, author, and datePublished. Schema.org can be written as JSON-LD, Microdata, or RDFa. Google recommends JSON-LD where the site’s setup allows it, because it sits in one <script> block instead of being woven through the HTML.

Schema.org reports that more than 45 million web domains use the vocabulary. It is ordinary infrastructure, not a specialist trick.

Does schema help you appear in AI Overviews or AI Mode?

No markup is required to appear in Google’s AI features, and Google says so in two places. The “AI features and your website” page states there is “no special schema.org structured data that you need to add,” and that “there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” The page has to be indexed and eligible for a snippet. That is the whole technical bar.

In May 2026 Google added a guide on optimizing for generative AI features. Its mythbusting section lists “overfocusing on structured data” as something not to do: “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” The same guide says to keep using structured data as part of normal SEO, because it supports rich result eligibility. In June 2026 Google added that llms.txt files are ignored by Google Search. The llms.txt question gets its own piece.

Two things follow. Schema is not an AI Overviews switch, and nobody can sell you one. And the AI features page does list structured data in its best practices, with a condition: make sure it matches the visible text on the page. The passage that gets lifted comes from the HTML. Schema can confirm what that passage is about. It cannot substitute for it. That is the extraction work described in SEO, AEO, and GEO are one system, and how AI Overviews choose a source covers what does decide the citation.

Does structured data affect rankings?

Structured data affects how a result can look, not where it ranks. Google’s general structured data guidelines say a structured data manual action removes a page’s eligibility for rich results and “doesn’t affect how the page ranks in Google web search.” Google staff have repeated the point publicly.

Eligibility is also not a guarantee. The guidelines say structured data “enables a feature to be present, it does not guarantee that it will be present.” Valid markup on a page Google does not trust will often produce nothing visible.

What Google does say is that it uses structured data it finds “to understand the content of the page.” Understanding is not ranking. It is still useful: a page that is unambiguous about its subject, author, and date is easier to retrieve for the right query and harder to confuse with a namesake.

Which rich results has Google retired?

Google has been shrinking the set of rich results since 2023, and several types that sites still mark up no longer display anything in Search. As of November 2026:

Rich resultChangeDate
FAQLimited to well-known government and health sitesAugust 2023
HowToRemoved from mobile, then from desktopAugust to September 2023
BreadcrumbsNo longer shown on mobile results; still supported on desktopJanuary 2025
Book Actions, Course Info, Claim Review, Estimated Salary, Learning Video, Special Announcement, Vehicle ListingPhased out of SearchAnnounced June 2025
Practice ProblemDocumentation removedJanuary 2026
FAQNo longer appears in Google Search for any siteMay 7, 2026

Google removed the FAQ rich result documentation on June 15, 2026. When it phased out the seven types in 2025, Google said the change “won’t affect how pages are ranked” and that use of those types outside Google Search is not affected.

The practical reading: a type that no longer produces a rich result is not harmful, but it is no longer a reason to do the work. FAQ is the type most sites still carry, so it gets its own section.

Does FAQ schema still matter in 2026?

Not in Google, and only a little anywhere else. FAQPage markup produces no rich result in Google Search for any site, and Google says AI Overviews and AI Mode need no special markup. It is still a valid schema.org type, it does no harm, and Microsoft says schema helps Bing’s AI interpret content. It is worth keeping only where it restates real questions a reader can see. The visible question and answer do the work. The markup only labels them.

What happened to FAQ rich results?

Google cut them back in two steps, three years apart.

DateChangeSource
August 2023Limited to “well-known, authoritative government and health websites”Google Search Central Blog
May 7, 2026Stop appearing in Google Search for all sitesGoogle documentation changelog
June 2026FAQ search appearance, rich result report, and Rich Results Test support droppedGoogle’s deprecation notice, as quoted by PPC Land
June 15, 2026FAQ rich result documentation removedGoogle documentation changelog
August 2026FAQ support removed from the Search Console APIGoogle’s deprecation notice, as quoted by PPC Land

The page that carried Google’s deprecation notice has since been removed, so the June and August tooling dates rest on a trade publication’s copy of it. Treat them as Google’s stated plan. Two consequences follow for reporting. A drop in “FAQ” appearances in an old dashboard is the feature ending, not a site problem. And FAQ markup is now checked with the Schema Markup Validator, not the Rich Results Test.

Removing the markup is optional. When Google restricted FAQ results in 2023, it said “structured data that’s not being used does not cause problems for Search.” Nothing in the 2026 changes says otherwise.

Does FAQ schema help ChatGPT, Perplexity, or Copilot cite you?

No engine has said it does. Microsoft’s October 2025 post says schema “can label your content as a product, review, FAQ, or event” so machines “can interpret with confidence,” which is a statement about understanding, not citation. OpenAI and Perplexity have published nothing on schema either way.

The two studies most often quoted on FAQ schema and ChatGPT point in opposite directions.

StudyMethodFAQ findingLimits
SE Ranking, November 2025216,524 pages on 129,000 domains; model of ChatGPT citation factorsPages with FAQ schema averaged 3.6 citations, against 4.2 withoutCollection dates not stated; correlational
AirOps, July 2025Over 12,000 URLs from 900 queries; pages cited in ChatGPT compared with uncited page-one resultsFAQPage or QAPage on 10.5% of cited pages, against 5.4% of uncited onesNo significance tests; publisher sells AI content tools; correlational

They measure different things, and neither controls for the obvious confounder: sites that maintain careful markup tend to maintain careful pages. SE Ranking itself calls the FAQ result “surprisingly weak.” Any vendor promise of a fixed citation lift from FAQ schema traces back to studies like these or to nothing. How to choose a GEO agency lists other promises to test. Test it on your own query set with the method in measuring AI search visibility.

What the first-party guidance does favor is the visible content. Microsoft’s post says assistants “can often lift these pairs word for word into AI-generated responses.” A generated answer quotes the HTML, not a <script> block. A two-line answer under a question heading can be what gets cited. Markup around a vague answer cannot make it one.

Should you keep or remove FAQ schema?

Remove it where it is wrong, keep it where it is right, and don’t spend a sprint on it either way.

What you findDecisionWhy
Real, visible FAQ generated from the same source as the HTMLKeepAccurate restatement, readable by Bing and other consumers, no cost
Questions or answers in the markup that are not on the pageRemove now“Don’t mark up content that is not visible to readers of the page”
Sales copy reshaped into questions on every service pageRemove the markup, then rewrite or cut the blockIt earns nothing in Google, and readers skip it
The same FAQ block repeated across many templatesKeep it on one canonical pageDuplicate answers compete with each other
User-submitted questions with several answersUse QAPage, not FAQPageSchema.org defines FAQPage for questions the site itself answers
A page with no FAQDon’t add one for the markupAny AI benefit comes from visible answers, not the label

To audit a site, crawl it and list every URL that emits FAQPage or QAPage, since the Search Console FAQ report is gone. Compare each one’s markup with the rendered page, group duplicate question sets, and trace whether the markup comes from the CMS, a plugin, or the theme. A theme that wraps every accordion in FAQ markup, pricing tables included, is the common culprit. Fix it at the source, validate, and log the date of the change so any later movement in citations can be read honestly.

This site uses FAQPage on one page. The six questions on the engagement page come from sales conversations, and the markup is generated from the same data as the visible text, so it cannot say anything the page does not. We kept it after May 2026 because it is accurate by construction and costs nothing to maintain. We have not tested whether removing it would change how any engine cites the page. The insight essays carry Article markup and question-phrased headings, not FAQPage. A reference guide is not a list of frequently asked questions.

Do ChatGPT, Perplexity, and Copilot use schema?

Microsoft is the only one of the three that has said publicly that schema helps its AI answers, and even that is a general statement, not a mechanism. In an October 2025 post on optimizing content for AI search answers, Krishna Madhavan of Microsoft Bing wrote that schema “helps search engines and AI systems understand your content,” usually added as JSON-LD. In March 2025, Search Engine Roundtable reported that Bing’s Fabrice Canel confirmed at SMX Munich that schema markup helps Microsoft’s LLMs understand content. That report relays a LinkedIn summary of the talk, not a transcript.

OpenAI’s documentation for ChatGPT search covers OAI-SearchBot: allow it in robots.txt, or your pages will not be shown in ChatGPT search answers. It does not mention structured data. Perplexity’s crawler documentation is the same: user agents, robots.txt, and IP ranges, with no reference to schema.

So the verified position is narrow:

  • Bing and Copilot: schema is stated to help understanding.
  • ChatGPT search and Perplexity: no published statement either way.
  • Every engine: the crawler has to be allowed to fetch the page first. Robots.txt for AI crawlers covers that decision.

Vendor studies that claim schema raises citation rates by a fixed percentage are not first-party evidence. Treat them as hypotheses to test against your own query set.

Which schema types are worth adding to a B2B or service site?

Most B2B and service sites need five or six types, done carefully, not twenty done loosely.

TypeWhereWhy
OrganizationHome page or about pageName, logo, URL, contact, and sameAs profiles that identify the company
WebSiteHome pageDeclares the site name and ties it to the organization
BreadcrumbListEvery page below the home pageStates the hierarchy; still shown on desktop
Article or BlogPostingEditorial pagesauthor, datePublished, dateModified, headline
Product with OfferPages that sell a priced itemMerchant listings and product snippets
LocalBusinessEach physical locationAddress, hours, phone, per location

Google’s Organization documentation recommends placing the markup on the home page or a single page that describes the organization, and says it does not need to be on every page. Its Article documentation has no required properties, and recommends author with a url or sameAs that uniquely identifies the person, plus ISO 8601 dates with a timezone.

A compact example for an article page, using one @graph so each entity is defined once and referenced by @id:

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Organization",
      "@id": "https://www.example.com/#org",
      "name": "Example Labs",
      "url": "https://www.example.com/",
      "logo": "https://www.example.com/logo.png",
      "sameAs": [
        "https://www.linkedin.com/company/example-labs",
        "https://github.com/example-labs"
      ]
    },
    {
      "@type": "WebSite",
      "@id": "https://www.example.com/#website",
      "name": "Example Labs",
      "url": "https://www.example.com/",
      "publisher": { "@id": "https://www.example.com/#org" }
    },
    {
      "@type": "BreadcrumbList",
      "itemListElement": [
        { "@type": "ListItem", "position": 1, "name": "Insights", "item": "https://www.example.com/insights" },
        { "@type": "ListItem", "position": 2, "name": "Structured data for AI search" }
      ]
    },
    {
      "@type": "Article",
      "headline": "What schema does and does not do for AI search",
      "datePublished": "2026-11-07T09:00:00+00:00",
      "dateModified": "2026-11-07T09:00:00+00:00",
      "author": {
        "@type": "Person",
        "name": "Jane Doe",
        "url": "https://www.example.com/team/jane-doe"
      },
      "publisher": { "@id": "https://www.example.com/#org" },
      "isPartOf": { "@id": "https://www.example.com/#website" }
    }
  ]
}

Every value in that block should appear on the page or on the page it points to. The author named in the markup is the author named in the byline. The dates are the dates printed under the headline.

Why does entity consistency matter more than markup?

An engine that is unsure which company you are will not cite you with confidence, and markup cannot settle a question the rest of the web contradicts. Google’s Organization documentation says some of its properties help Google tell an organization apart from others with similar names. That only works if the name, address, and profiles agree everywhere they appear, including the third-party pages generative engines cite and the stale ones that cause AI to get your brand wrong.

Consistency means:

  • One legal or trading name, used the same way in the title tag, the footer, the Organization markup, and every third-party profile.
  • One canonical @id per entity, reused across pages instead of a new organization node on every template.
  • sameAs pointing to profiles you control and keep current, not to every directory that ever listed you.
  • One set of facts for founders, locations, and services, with retired ones removed from both the site and the profiles.

This is the same work that decides what gets cited: a claim that is attributable, stable, and corroborated. sameAs is a pointer to corroboration. The corroboration has to exist.

How do you validate structured data?

Use two tools, because they answer two different questions.

  • The Rich Results Test checks whether markup is eligible for the rich results Google still supports, and shows what Googlebot renders.
  • The Schema Markup Validator checks syntax and conformance with schema.org, regardless of Google features. It replaced Google’s Structured Data Testing Tool, which moved to schema.org in 2021.

After launch, the URL Inspection tool in Search Console shows what Google found on the indexed version, and the rich result reports show errors at scale. A test that passes in isolation can still fail on the live page if a tag manager injects markup late or a template overwrites it. Injected markup is also invisible to most AI crawlers, because they do not run JavaScript. Render JSON-LD on the server.

What are the most common structured data mistakes?

The common mistakes all break the same rule: the markup says something the page does not.

  • Markup for invisible content. Google’s guidelines say “don’t mark up content that is not visible to readers of the page.” Reviews, prices, or questions that exist only in JSON-LD are a manual action risk.
  • FAQ spam. Sales copy reshaped into questions, marked up on every service page. It earned little after 2023 and earns no rich result in Google now.
  • Panic-removing valid FAQ markup. Google says unused structured data causes no problems. Remove what is wrong, not what is merely unrewarded.
  • Conflicting @ids. Two organization nodes with different names or logos, one from the theme and one from a plugin, describing the same company two ways.
  • Stale dates. A dateModified that changes on every deploy, or one that never changes after a real rewrite. Both make the date useless as a signal.
  • The wrong type. Google’s relevance guideline gives the example of instructions labeled as recipes. A service page is not a Product unless it sells a priced item.
  • Uncrawlable images. Google requires image URLs in structured data to be crawlable and indexable.

Structured data is a restatement. Get the page right, keep the entity consistent, then describe both accurately. In that order it helps. Out of that order it decorates a page that was never going to be the answer.

Sources

Bring us the domain.

We will tell you where you are absent, which page is in the way, and what has to be true before a citation is possible.

Request an audit