citelity.

Schema markup and AI Overviews: what Google actually says

Google's May 2026 guide says no special schema is needed for AI Overviews, and Google Search doesn't use llms.txt. What markup still does, which rich results remain after more than ten retirements, and how to measure AI visibility.

By Ed Grows · Updated 12 min read

"Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." That's Google, in the guide to generative AI features it published on 15 May 2026. The next sentence matters as much: "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."

So the short answer is no: you don't need schema to appear in AI Overviews or AI Mode, and no markup type has been shown to cause citations. Markup is still worth maintaining for the rich results that remain, which is a shorter list than it was in 2023.

What the evidence says

The studies that exist are correlational. Cyrus Shepard's review of AI citation ranking factors (Zyppy, May 2026) scored structured data 5.6 out of 10, noting that "practically every study that looks at schema and AI citations finds a positive relationship. The effect is typically small, but it's amazingly consistent across studies." None of those studies isolates schema as the cause. Sites that maintain good markup tend to maintain everything else well too.

You may also meet a claim that pages with schema are "roughly 35% more likely to be cited". We couldn't find a primary study behind it, so treat it as unsourced until someone produces one.

What markup does reliably is remove ambiguity about what a page contains:

WITHOUT MARKUP: THE PARSER INFERS<h1>Acme Tasks</h1><p>From $12</p><span>Jane Smith</span>A product? A company? A film?Is $12 the price, or a discount?Is that the author, or someonethe page mentions?WITH MARKUP: IT IS STATED"@type": "SoftwareApplication""offers": { "price": "12.00" }"author": { "@type": "Person" }The entity, the number and theperson are each labelled, and theperson can link to profileselsewhere on the web.

That's a real job, and it's the one rich results depend on. It just isn't the citation lever many AEO guides describe.

What Google has retired since 2023

More than ten rich-result features, in five rounds. Several still appear on AEO checklists written before the latest ones.

What still produces a search feature

STILL PRODUCE A SEARCH FEATURE (SELECTION)Article / BlogPostingProfilePageOrganization (logo)BreadcrumbList (desktop)ProductReview snippetSoftware appVideoRecipeLocalBusinessEventQ&A / Discussion forumAlso: Job posting, Course list, Movie, Vacation rental and others. Dataset markup now serves Dataset Search only.RETIRED SINCE 2023FAQPageMay 2026HowToSep 2023Sitelinks search boxNov 2024Practice problemJan 2026ClaimReviewJun 2025Course InfoJun 2025Book ActionsJun 2025Estimated SalaryJun 2025Learning VideoJun 2025Special AnnouncementJun 2025Vehicle ListingJun 2025Retired types remain valid schema.org and can stay on the page. They just produce no search feature.

The full, current list is Google's search gallery. Two entries are often misread. Person is not a rich-result feature in its own right; author and profile markup counts through ProfilePage. Organization is a gallery feature for logos and knowledge-panel details, not a visual snippet in the results list.

Retired markup isn't invalid. Google said in 2023 that structured data it doesn't use "does not cause problems for Search" (Google, August 2023), so there's no need to strip it out. Just stop adding it for a result that no longer exists. The FAQ case is covered in detail in FAQ rich results are gone.

AI
Free tool · No signup
Free Schema Validator
Paste any URL → find the structured data your page is missing, with ready-to-paste JSON-LD fixes.
Check your schema →

Three types worth implementing first on a content site

Article or BlogPosting with a Person author. Google lists no required properties for Article; it recommends headline, image, datePublished, dateModified and author (with name and url). The author field is where most sites fall short. Code and the matching author-page markup are in Article schema with author.

Organization, once, site-wide. Name, URL, logo and sameAs to your official profiles, referenced by every article's publisher. Google uses it for logo and knowledge-panel details.

BreadcrumbList. It still produces a rich result on desktop; Google stopped showing breadcrumbs on mobile in January 2025. Google's rules are looser than most guides say (the homepage and the last item's URL are optional, and multiple trails are allowed). A working implementation is in BreadcrumbList JSON-LD in Next.js.

Markup must match what readers see

Google's structured data guidelines don't allow marking up content users can't see. Q&A pairs that aren't on the page, ratings that aren't shown, or an author byline that doesn't appear are all mismatches, and they can cost a page its rich-result eligibility.

Validators help only partly here. Google's Rich Results Test, when given a URL, fetches and renders the page, and the Schema Markup Validator checks against schema.org. Both check syntax and eligibility; neither checks that the markup says the same thing as the visible text. Read the two side by side after any significant edit.

Crawler access matters more than markup

Markup on a page an AI system can't fetch does nothing for it. Two details are commonly confused:

  • Google's AI Overviews and AI Mode are part of Search. They use pages Googlebot can crawl. Google's crawler documentation says the Google-Extended token "does not impact a site's inclusion in Google Search", so blocking it doesn't remove you from AI Overviews. To limit what appears there, Google points to the usual preview controls such as nosnippet.
  • Other AI engines use their own crawlers, and most of them don't execute JavaScript. JSON-LD added only in the browser is invisible to them, so render it on the server.
AI
Free tool · No signup
Free AI Crawler Checker
Paste a domain → see which AI crawlers your robots.txt allows and which it blocks, with training and retrieval bots told apart.
Check your crawlers →

llms.txt and "AI-specific" markup

llms.txt is a proposed file at a domain root that describes a site for language models. Google settled its own position on 15 June 2026 by adding a note to its AI guide: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." Its updates log adds that such files "won't negatively or positively impact your visibility or rankings." Outside Google, Zyppy's review scored llms.txt 2 out of 10, the lowest factor examined: "we're unable to find any credible evidence or experiments showing LLMs.txt files influence AI citations in any way."

The only case for publishing one is tools and agents outside Google that choose to read it. Your server logs will tell you whether anything requests it. If something does, our llms.txt generator builds the file; if nothing does, skip it.

AI-specific schema doesn't exist as a supported standard. Google says there's "no special schema.org markup you need to add", and since 15 May 2026 its spam policies explicitly cover attempts to manipulate generative AI responses in Search. Anyone selling markup that "tells AI how to cite you" is selling a proposal at best.

How to measure AI visibility instead

Since 31 August 2026 every site has Search Console's Generative AI performance report. It launched on 3 June as a beta for a subset of UK sites (Google's announcement). It shows impressions only for AI Overviews and AI Mode, by page, country, device and date. There are no clicks, no queries and no citation text, so it tells you that a page appears in AI features, not for which prompts or alongside whom.

For that, you still need to sample queries yourself, or with a tracker, and record which sources each answer cites. The method, and how to tell a real loss from normal churn, is in AI Overview citation disappeared.

Common mistakes

  1. Markup injected only by client-side JavaScript. Google can process it; most other AI crawlers can't. Server-render it.
  2. Wrong case in type names. "FaqPage" is not the schema.org type FAQPage. Run markup through the Schema Markup Validator.
  3. The same block twice. A plugin and a theme both emitting identical Article or BreadcrumbList markup is noise. Different blocks describing different things are fine.
  4. Adding retired types for a rich result. HowTo hasn't produced one since 2023, FAQ since May 2026.
  5. Broken URLs in a BreadcrumbList. They send users and crawlers to dead pages.
  6. Changing dateModified without changing the content. The date should describe a real update; a reader comparing it to the text will notice.

Where to start

Check the author field on your ten most visited articles, then confirm the crawlers you care about can fetch those pages. Both are quick, and both matter more than adding new markup types.

FAQ

Can I have several schema types on one page?
Yes, and most article pages should: BlogPosting for the post, a Person for the author, an Organization for the publisher and a BreadcrumbList for the trail. Several entities of one type are also fine when they describe different things, such as two breadcrumb trails. Avoid duplicate blocks describing the same thing, usually from two plugins.

Sources cited in this piece

Updated October 1, 2026: added Google's May 2026 AI guide and its June llms.txt note, corrected the retired-feature list (now eleven, including the sitelinks search box and practice problems), fixed the Article and BreadcrumbList details, and added the Search Console Generative AI report.

Keep reading

aeo · 8 min readClaude-SearchBot vs ClaudeBot vs Claude-User: what to allowAnthropic runs three crawlers, and Claude's web search also draws on Brave. What each bot does, what blocking it costs according to Anthropic, and how to check Claude can see you.aeo · 7 min readGoogle-Extended vs Googlebot: what actually controls Gemini citationsBlocking Google-Extended doesn't touch AI Overviews, but it can keep you out of the Gemini app's grounded answers. Google's three controls, the surfaces each one governs, and how to test Gemini.aeo · 10 min readAEO content checklist: 47 checks, from crawler access to quarterly upkeep47 AEO checks in the order you do them: retrievability (now including Search Console's AI control), planning, structure, schema, authorship, validation, upkeep.