Schema markup for AI search: what the evidence says, and what to implement anyway.

Does schema markup help you appear in Google AI Overviews and AI Mode?

Not directly, according to Google. Its documentation on AI features says there are no additional requirements to appear in AI Overviews or AI Mode, and no special schema.org structured data you need to add. Eligibility works as it does in traditional search: a page has to be indexed and eligible to show in Google Search with a snippet.

Google's guide to optimizing for generative AI search, added May 15, 2026 and last updated July 10, 2026, makes the same point and adds a reason. These AI-powered search features are rooted in Google's core ranking and quality systems. They use retrieval-augmented generation, which grounds AI-generated summaries in pages from the Search index, plus "query fan-out," where the system also searches for related questions beyond the user's query. Google's own framing is that optimizing for generative search experiences is still SEO, so traditional SEO practice carries over to AI-driven search. The guide still calls structured data a good idea as part of your overall SEO strategy, because it helps with rich result eligibility, one of the enhanced search features Google offers. Google says the same about AI text files and special markup: you don't need them (our post on whether you need an llms.txt file covers that evidence).

A definition helps here. Schema markup is code, usually JSON-LD, that labels what a page's content represents using the shared Schema.org vocabulary, which Google, Microsoft, Yahoo, and Yandex founded. It gives search engines explicit signals about a page and changes nothing readers see. How structured data gets used is up to each system, and Google's one markup rule for AI features is a plain one: make sure your structured data matches the visible page content.

What do Microsoft, OpenAI, Perplexity, and Anthropic say about structured data?

Only Microsoft says schema helps its AI systems. The other three document their crawlers and don't mention structured data.

Microsoft's position comes in two forms. In an October 8, 2025 post on the Microsoft Advertising blog, Krishna Madhavan, a principal product manager at Microsoft Bing, wrote that schema "helps search engines and AI systems understand your content." The same post says AI assistants split a page into smaller pieces instead of reading it top to bottom, and that clear headings mark where one idea ends and the next begins. It also says nothing guarantees selection in an AI answer. Earlier, in March 2025, Fabrice Canel of Bing said at SMX Munich that Microsoft's large language models use schema markup. Search Engine Land reported it on March 20, 2025, based on a LinkedIn post that Canel confirmed in the comments. Treat it as a conference remark, not documentation.

The other three are quieter. OpenAI's bot documentation says OAI-SearchBot surfaces websites in ChatGPT's search features, and it recommends allowing the bot in robots.txt. Perplexity says PerplexityBot is designed to surface and link websites in Perplexity's search results, and that it isn't used to crawl content for AI foundation models. Anthropic's help center describes Claude-SearchBot as a crawler that improves the quality and relevance of search results for users. None of the three pages mentions schema, structured data, or JSON-LD. Silence isn't proof that these systems ignore markup, but it does mean there's no vendor documentation to point to.

What does the research say about schema markup and AI citations?

Mostly correlation and small tests, plus one large before-and-after comparison that found no lift on any platform it measured. None shows that adding schema alone earns AI citations.

Read the vendor studies with their limits in mind. The Ahrefs authors note that their sample held only pages already cited heavily in AI Overviews, that pages adding JSON-LD often changed other things at the same time, that all schema types were pooled, and that 30 days may miss slower effects. They also suggest that cited pages tend to have schema because those sites invest in other SEO work, which makes schema a correlate, not a cause. The OtterlyAI test rests on one brand's site and the Williams-Cook test on a single page. Neither settles the question, but together they say that markup isn't a reliable way to get a fact into AI-generated answers.

One number gets quoted often. GPT-4's accuracy on enterprise questions rose from 16% to 54% when the questions ran over a knowledge graph instead of a raw SQL database, in a 2023 benchmark by Sequeda and colleagues. That benchmark tested a closed enterprise database, not public web pages, so it says little about whether AI search engines or other AI models read the schema on your site. It shows why structured information matters when you control the data, and no more.

The fair summary: no single study settles it, but the signals from multiple sources agree. There's no good evidence that adding markup, on its own, gets you cited, and no clear evidence that it hurts. Schema markup plays a supporting role.

Does FAQ schema still matter?

Not as a search visibility tactic. Google stopped showing FAQ rich results in Search on May 7, 2026.

That change finished a long retreat from rich result types that SEO teams once leaned on.

FAQPage markup isn't wrong, but it no longer earns a rich result, so it isn't a reason to add one. A question-and-answer section is still worth writing for readers. Any system that reads visible text can reuse it, and Microsoft's guidance lists Q&A blocks among the formats that assistants can reuse easily. Treat rich result features as unstable. Build markup around what a page is, not around whichever result type Google shows this year.

Which schema markup is still worth adding to a B2B SaaS site?

A schema markup strategy for a B2B SaaS site can stay short: the types that describe the company, the site, and the content, and that Google still supports. Those are Organization, WebSite, BreadcrumbList, and Article.

Two rules apply to all of it. Google says your structured data must be a true representation of the page content, and you shouldn't mark up anything readers can't see. A violation, such as irrelevant markup, can bring a manual action that removes rich result eligibility, though Google says that action doesn't affect ranking in web search. Accuracy matters more than volume, and comprehensive schema that doesn't match the page can cost you eligibility. And when a business controls the reviews about itself, Google says its pages with Organization or LocalBusiness markup can't earn star ratings, so curated customer feedback and testimonials on your own site won't produce them. That, plus the rating requirement, is why most B2B SaaS sites can skip SoftwareApplication.

Google recommends JSON-LD if your site setup allows it. A minimal Organization block looks like this:

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Example SaaS Inc.",
  "url": "https://www.example.com/",
  "logo": "https://www.example.com/images/logo.png",
  "sameAs": ["https://www.linkedin.com/company/example-saas"]
}

Put one Organization block on the home page and keep the name, url, and logo identical wherever you repeat them, so every page describes the same entity. Google lists sameAs as a recommended property for pointing to other pages about the organization, such as social profiles. For posts, Article schema is the lightest addition: describe the author with a Person schema block, since Google lists the author's name and url among the recommended properties. Validate the markup with Google's Rich Results Test while you build, then watch the rich result reports in Search Console after launch. Well-implemented schema doesn't have to cover the entire site. It has to be accurate and maintained, like the rest of technical SEO.

What matters more than schema for AI search optimization?

Content worth citing, crawl access, and facts in visible text. Each has a source behind it.

Start with the content. Google's guide asks for non-commodity content: unique expert or experienced takes, as opposed to generic tip lists that repeat common knowledge. It also warns against publishing large volumes of pages to manipulate rankings. The one peer-reviewed study in this post points the same way, because the edits that helped most on its 10,000-query benchmark were citing sources, adding quotations, and adding statistics. They are changes to the text, not to the markup.

Then access. A page has to be crawlable, indexed, and eligible for a snippet before any of this counts, because Google says its models use publicly accessible, crawlable web content. OpenAI says each robots.txt setting is independent, so you can allow OAI-SearchBot for search while disallowing GPTBot, the crawler tied to AI training. Our post on AI crawlers and robots.txt walks through the choices.

Then the page text itself. If a price, a feature, or a date matters, put it in the copy and not only in the markup, because the OtterlyAI and Williams-Cook tests suggest that a fact stored only in JSON-LD may never reach AI search results. Clear headings and answer-first paragraphs help readers first. Microsoft says assistants split pages into pieces, so a clear content structure helps AI understand where one idea ends and the next begins. Google says there's no need to chunk content for its systems. Write for the reader and you cover both positions, because a clear structure serves search engines and AI systems alike.

A content strategy built on these three also gives you something to measure, which is the next question.

How can you tell whether structured data is helping your AI visibility?

Validate it, then watch rich result eligibility and AI search visibility as two separate things.

Google's Rich Results Test checks markup during development, and the rich result status reports in Search Console show whether pages stay valid after launch. For AI features, Google points site owners to the Generative AI performance report in Search Console, and says that traffic from AI features is counted in the Web search type, so it sits inside your regular search performance data. That covers Google. For other AI engines and assistants, the practical method is a fixed set of buyer prompts that you check on a schedule, which our 30-prompt method describes, plus the referral traffic that GA4 can show.

Ahrefs' own caveat points to the main lesson: when structured markup ships together with a rewrite, a redesign, or a new set of pages, you can't tell which change moved the numbers. If you want to know whether markup matters on your site, ship it on its own and compare the same prompts before and after.

We implement structured data and AEO for B2B SaaS companies, so we have an interest in this topic. For how we approach AI visibility, see our AEO agency page.

Questions about schema markup and AI search