AI Search

Schema for AI Overviews: Which Markup Still Helps

Read Time7 Min Read
Length1,747 Words
CategoryAI Search
Schema for AI Overviews: Which Markup Still Helps
AI Generated

Readers hunting schema markup for AI search usually want a type, a JSON-LD snippet, and a citation bump. Google’s own guidance is the opposite of that shopping list. Structured data is not required for generative AI search, there is no special schema.org markup to add for those features, and eligibility for a supporting link in AI Overviews or AI Mode is the same as a normal Search snippet: indexed, with no extra technical requirements.

That does not make markup useless. It makes most “AI schema” projects a bad trade. Google still recommends structured data for rich-result eligibility, and it still asks that markup match the visible text on the page. Bing’s Copilot side has said schema helps its models understand content — comprehension, not a citation promise. ChatGPT, Perplexity, and Claude have not published a schema requirement for being cited.

The useful test is fail-closed. Keep markup that describes facts already on the page and that machines still misread: who you are, who wrote this, what the price is, how the page sits in the site. Refuse schema that exists only to impress an answer engine, including FAQPage used as a citation tactic after FAQ rich results left Google Search. The next edit that actually moves visibility is a standalone answer passage a model can quote, not another checklist of types.

Summary
  • Google says structured data is not required for AI Overviews or AI Mode, and there is no special schema.org type to add.
  • Adding JSON-LD has not been shown to cause more AI citations; an older “cited pages have more schema” finding is correlation, not cause.
  • Keep markup that matches visible facts and still supports rich results or disambiguation: Organization, Article or BlogPosting, Product and Offer, BreadcrumbList.
  • Do not mark up content that is not on the page. Visible Q&A still matters; FAQPage JSON-LD is not a confirmed AI citation tactic.
  • Refuse schema-only AI SEO projects and vendor citation multipliers. Spend the next edit on a clear answer passage.

Does Schema Markup Help You Get Cited in AI Overviews?

Schema markup script compared with an AI Overview citation landing on a visible paragraph

No. Google says structured data is not required for AI Overviews or AI Mode, and there is no special schema.org type to add for those features. A supporting link has the same eligibility as a normal Search snippet: the page must be indexed, with no extra technical requirements.

Google still recommends schema, but that advice is about rich-result eligibility in Search. It is not an AI citation requirement, and it is not a reason to commission markup solely to appear in generative answers.

1,885
Pages that added JSON-LD, vs about 4,000 controls
4.6%
AI Overview citation drop versus controls
+2.4%
AI Mode change, treated as noise

Ahrefs tracked those 1,885 pages from August 2025 to March 2026 and found no meaningful AI citation increase. ChatGPT citations rose 2.2% on the treated pages, also treated as noise aside from the small Overview decline. Do not buy a schema project to get cited. Spend the edit on a visible answer a reader can quote, and connect that passage to citation patterns rather than another markup list.

Key Takeaway

Not a citation lever — Schema does not buy AI Overview citations. Keep markup only when it still describes visible page facts for Search rich results, and put the next edit into a quotable answer.

Citation shifts after schema (%)

Sources

Why Cited Pages Have Schema If Adding It Does Not Help

Diagram of schema markup correlation: cited pages already had schema, and adding markup creates no new citations

The higher rate is correlation, not a lever you can pull. Ahrefs' scan of 6 million URLs found AI-cited pages almost three times more likely to carry JSON-LD than uncited pages. The later intervention did not support that gap as cause and effect.

Every page in that schema test already had more than 100 AI Overview citations in February 2025, before JSON-LD was added. The study does not speak to pages AI has never cited.

A Semrush analysis of 304,805 URLs, as reported by Radiant Elephant, associated structured data with a 22% citation lift, but ranked it behind clarity, E-E-A-T, Q&A format, and section structure. Search Atlas's study of 5.5 million responses, reported on the same page, found no measurable correlation between schema coverage and citation rates on OpenAI, Gemini, or Perplexity.

Stronger pages do more SEO hygiene. Adding a script tag to a weak page does not recreate the cited set.

Key Takeaway

Correlation — Cited pages carry schema because they are already stronger, not because the markup is what earned the citation.

Sources

Which Schema Types Are Still Worth Adding?

Schema markup worth keeping: Organization sameAs links and an Article byline that match visible page text

Keep the types that describe facts a reader can already see, facts machines often misread, and that still support rich results or entity clarity: Organization, Article or BlogPosting, Product and Offer when price and availability are on the page, and BreadcrumbList for hierarchy.

Organization markup belongs on the homepage. Google says it can help the search engine understand administrative details and disambiguate the organization in search results, and it does not need to be on every page. sameAs clarifies which entity you are. It does not change AI visibility by itself.

Article, NewsArticle, or BlogPosting can help Google understand the page and show better title text, images, and date information. There are no required properties. A dateModified value does not get the page quoted.

Author, dates, price, and availability have to match what a reader can see. Do not add a type because a vendor labeled it AI schema.

Key Takeaway

Still worth it — Keep Organization, Article or BlogPosting, on-page Product and Offer, and BreadcrumbList. sameAs and dateModified clarify identity and freshness; they do not buy an AI citation.

Sources

Should You Remove FAQPage Schema After Rich Results Ended?

FAQPage schema rich result gone from search while the same questions remain as visible text on the page

Keep the questions and answers a reader can see. Drop the expectation that FAQPage is still a search feature, or a way to get cited.

Google stopped showing FAQ rich results on May 7, 2026, and said it would drop the FAQ search appearance, the rich result report, and Rich Results Test support in June 2026. FAQPage remains valid Schema.org. It is not a confirmed citation tactic, and it does not belong on a page that is not an FAQ.

Joost de Valk reports no primary source from OpenAI, Anthropic, Google AI Overviews, or Perplexity confirming they favor JSON-LD over rendered HTML. He also flags a 3.2x FAQ-schema claim whose cited URL returns a 404. Treat that kind of multiplier as unsourced.

Does HowTo schema still help?

HowTo markup does not replace the steps. The citable unit is the visible step-by-step text. Ship HowTo JSON-LD only when those steps are already on the page and you still want the type for comprehension, not for an Overview bump.

Key Takeaway

Keep the answers — Visible questions, answers, and steps are what can be cited. FAQPage and HowTo JSON-LD are neither a live SERP feature nor a confirmed AI citation tactic.

Sources

Do ChatGPT, Perplexity, or Claude Actually Parse JSON-LD?

AI answer panels reading visible article text while JSON-LD schema markup sits unread in the page source

Leaving rich results behind does not mean the other engines parse a valid graph. None of ChatGPT, Perplexity, or Claude has published a schema requirement for citations. A searchVIU live-fetch test found none of five AI systems pulled schema data at answer time; they read only the visible HTML.

Otterly.ai added five schema types from December 2025 to March 2026 and tracked 319 prompts across seven platforms. Six of seven could not fetch or interpret the schema when asked. Only Gemini returned the JSON-LD correctly. AI Overviews coverage rose in that window, but control brands rose too, which points to an algorithm change rather than the markup. Google AI Mode also invented a schema type that was not on the page.

Microsoft’s Fabrice Canel said at SMX Munich in March 2025 that schema helps Bing’s LLMs understand content for Copilot. That is comprehension, not a citation guarantee. Mark Williams-Cook put a fake address only inside invalid JSON-LD; ChatGPT and Perplexity returned it, so a script block can be read as plain text. Hidden or fake facts are a liability. llms.txt is not a substitute: Google says you do not need new machine-readable files, AI text files, markup, or Markdown to appear in Search, including generative AI features.

Key Takeaway

Visible text wins — ChatGPT, Perplexity, and Claude have not published a JSON-LD citation requirement. Keep markup as a description of visible facts, never as hidden text or a stand-in for llms.txt.

Sources

What Should Structured Data Match Before It Ships?

Structured data check matching byline and date to JSON-LD, with one mismatched price field flagged before schema ships

Before it ships, structured data has to match what a reader already sees. Google lists matching structured data to the visible text on the page as an SEO practice for AI features, not as a separate AI markup requirement. Every fact in the graph should already be visible text.

Check the name, author, dates, price, availability, and any FAQ or HowTo steps against the rendered page. If a field is not visible, strip it. Do not mark up content that is not on the page.

A green Rich Results Test only means the type is valid for a rich result where that feature still exists. It is not evidence of an AI Overview win, and it will not cover FAQ after Google drops that test support.

On an automated template, compare the JSON-LD payload with the rendered HTML of that URL before publish. A passing validator on a sample page is not a pass for every generated page.

Key Takeaway

Match the page — Ship schema only when every marked-up fact is already visible on that URL. A green test validates a rich-result type; it does not prove an AI citation.

When a Publishing Gate Should Refuse Schema

Publishing gate refusing schema-only AI projects, mismatched FAQPage markup, and JSON-LD that does not match the page

That visible-text check is the publishing gate. On an automated page, refuse the payload when it does not describe what the template already shows — not after a citation fails to show up.

01
Citation claim or vendor multiplier
Fail closed when the only reason to ship the block is an AI citation claim or a vendor’s promised lift.
02
Type does not match the brief
Refuse FAQPage on a page that is not an FAQ, and any type the template was not written to support.
03
Facts missing from the render
Refuse dates, offers, or any asserted fact absent from the output, including a graph built with no visible-text check.

Measure the passage, not the script. Google’s query fan-out is a set of concurrent, related queries the model generates to request more information — its lawn-weeds example is the pattern — so a standalone answer has to exist for those sub-questions. After a content edit, watch citations, mentions, and share of answer. After a markup edit, watch rich-result eligibility and entity consistency only. Do not expect a citation bump.

Key Takeaway

Next edit — Point the next edit at the citation-patterns piece and the passage-structure guide, and use the AI visibility article for the measurement set. Do not open another schema checklist.

Key Takeaways

[01]
No citation purchaseGoogle does not require structured data for AI Overviews or AI Mode, there is no special schema.org type, and adding JSON-LD did not raise citations versus controls.
[02]
Correlation, not causeCited pages are more likely to carry JSON-LD because strong pages already use it, not because markup creates the citation.
[03]
Keep only matching markupOrganization, Article types, Product and Offer, and BreadcrumbList are worth keeping when they describe visible facts and still support rich results or entity clarity.
[04]
FAQPage is not a tacticAfter FAQ rich results ended, FAQPage stays valid Schema.org but is not a confirmed citation lever and does not belong on non-FAQ pages.
[05]
Engines do not require JSON-LDChatGPT, Perplexity, and Claude have published no schema citation requirement, and llms.txt is not a substitute for visible HTML.
[06]
Ship only visible factsEvery JSON-LD value must already appear in rendered text, and a publishing gate should refuse citation-only schema, vendor multipliers, and facts missing from the page.

Measure citations after content edits, and spend the next revision on passage structure and visibility measurement rather than another schema checklist.

Frequently Asked Questions

Do I need special schema to appear in Google AI Overviews?
No. Google says structured data is not required for generative AI search, and there is no special schema.org markup to add. You also do not need new AI text files, extra markup, or Markdown just to appear in Search, including those features.
Does adding JSON-LD increase AI citations?
Not in the test that actually added it. Ahrefs tracked 1,885 pages that added JSON-LD against about 4,000 controls and found no meaningful citation gain. AI Overview citations on the treated pages fell 4.6% relative to controls. An earlier scan that found cited pages were almost three times more likely to carry JSON-LD is correlation, not proof that schema caused the citations.
Should I still use FAQPage markup for answer engine optimization?
Not as an AI citation tactic. Google stopped showing FAQ rich results on May 7, 2026. FAQPage remains valid Schema.org, but the JSON-LD does not replace visible questions and answers on the page, and there is no primary source from OpenAI, Anthropic, Google AI Overviews, or Perplexity confirming they favor JSON-LD over rendered HTML.
Which schema types are still worth maintaining?
Keep types that describe visible facts and still help classic Search: Organization with sameAs on the homepage, Article or BlogPosting with a real author and dates, Product and Offer only where price and availability are on the page, and BreadcrumbList for hierarchy. Google says Article markup can help it understand title text, images, and dates, and Organization markup can help it disambiguate the business.
Will schema get me cited by ChatGPT or Perplexity?
There is no published schema requirement from those products for citations. A live-fetch test found five AI systems read visible HTML rather than the schema block at answer time. A separate check found six of seven AI platforms could not fetch or interpret added schema when asked; only Gemini returned the JSON-LD correctly.
What should I do instead of a schema-only AI SEO project?
Write a standalone answer passage that states the fact in plain language on the page, then measure whether that passage is what systems quote. Link that work to citation patterns and passage structure, not to another schema checklist or a vendor multiplier.
Sources

You Might Also Like

View All Articles