Corrected on September 17, 2026

The first version called schema markup "the single biggest AI-search lever" and claimed AI models weight pages with JSON-LD higher, without a source. It also said AI engines quote FAQPage text verbatim, that Perplexity and ChatGPT still read HowTo markup, that a wrong wordCount lowers trust, and that Lumina's validator resolves @id references across pages. None of that is documented, and the last point was wrong about my own tool. The "live audit of 10 articles" had no saved data, so I removed it.

Does schema markup help AI search? Mostly not in the way it's sold. Google says there's no special structured data you need for AI Overviews or AI Mode, and the best controlled study so far found no meaningful effect on AI citations. Schema is still useful, for Google rich results and for describing your pages clearly.

It's part of the GEO cluster. GEO vs SEO covers the comparison, SEO for AI Search the practical tactics, and the Schema Markup Guide the general how-to.

Does Schema Markup Help AI Search?

Google: its documentation on AI features in Search says there's no special schema.org structured data you need to add to appear in AI Overviews or AI Mode. The guide to generative AI in Search goes further: no special files, markup or Markdown versions are needed. A page has to be indexed and eligible to show a snippet.

Ahrefs: in a study published in May 2026, Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 and compared them with 4,000 similar pages that didn't. Citations in AI Overviews fell by 4.6%, AI Mode rose by 2.4% and ChatGPT by 2.2%, which Ahrefs summarized as no major uplift. The study also explains why the popular stat misleads: cited pages are about three times more likely to have JSON-LD, but schema tends to sit on better-maintained sites with stronger content, so the correlation says little about cause.

OpenAI, Perplexity, Anthropic: none of them document using structured data to choose or cite sources. They document their crawlers and how to allow them. That's where the concrete lever is.

So the honest answer: a page without schema isn't penalized in AI search, and adding schema isn't a proven way to get cited. It's worth doing for other reasons, covered below.

Is There a Separate "AI Search Schema"?

No. Many guides split schema into an "SEO stack" and a "GEO stack". The vocabulary is the same. What differs is the claims attached to it.

ClaimWhat's documentedVerdict
Schema gets you cited in AI OverviewsGoogle: no special structured data needed; Ahrefs: no major citation liftNot supported
AI engines quote FAQPage text verbatimNo AI provider documents thisUnsourced
@id entity links raise author trust in LLMsNo AI provider documents this; @id is a Linked Data identifierUnsourced
Schema powers Google rich resultsGoogle's search gallery lists the supported features and their required propertiesDocumented
Markup must match visible contentGoogle's structured data policiesDocumented

If you already maintain good markup for Google, you're not missing an AI-specific layer.

Schema Types Still Worth Adding in 2026

These are worth it because Google documents a use for them, not because of AI citations. The Schema Markup Guide has code and required properties for each.

  • Article or BlogPosting. Helps Google understand the page and show better title, image and date information.
  • Organization. Helps Google understand and disambiguate your organization's details, like logo and contact information. There's no guaranteed knowledge panel.
  • BreadcrumbList. Shows your site hierarchy in results, on desktop only since January 2025.
  • Product, Review snippet, Recipe, Event, LocalBusiness, VideoObject. Where the page really is that thing. These are the types with visible rich results.

Stop expecting much from FAQPage (no rich results since 7 May 2026), HowTo (removed in 2023) and Speakable (still a beta for US-English news publishers).

Five Schema Practices That Hold Up

1. Mark up only what's on the page. Google's structured data policies say markup has to represent the visible content. Hidden or misleading markup can lead to a manual action and loss of rich result eligibility.

2. Keep FAQ markup identical to the visible FAQ, or remove it. There's no rich result left, so the only reason to keep FAQPage is to describe the page accurately. A schema that says something the page no longer says is worse than none.

3. Use one JSON-LD block with @graph and @id references. Declaring your Organization and author once and referencing them by @id keeps markup consistent across pages. It's a maintenance pattern, not a ranking or citation trick.

4. Add sameAs to Organization and Person. Links to your official profiles help Google connect the entity to the right organization or person. It's not documented as an AI citation factor.

5. Update dateModified only for real changes. Google asks for accurate dates and warns against changing them without substantial content changes.

JSON-LD vs Microdata vs RDFa

Use JSON-LD. Google recommends JSON-LD where your setup allows it, and supports Microdata and RDFa too. No AI provider documents a preference. If a CMS plugin outputs Microdata and you can't switch it off, make sure the two formats don't contradict each other. The Schema Markup Guide compares the three.

Schema Mistakes That Actually Cause Problems

  • Markup for content that isn't visible. The one schema mistake with a documented penalty (see practice 1).
  • Missing required properties. A Product snippet without a review, rating or offer doesn't qualify for the rich result.
  • Invented properties. Fields that don't exist on schema.org are ignored. The Schema Markup Validator flags them, the Rich Results Test usually doesn't.
  • Expecting AI citations from markup. Time spent adding schema types for AI search is time not spent on crawler access, HTML content and well-sourced answers, the parts that are documented to matter.

How to Validate Schema Markup

The Rich Results Test checks Google eligibility, the Schema Markup Validator checks the full schema.org vocabulary, and Lumina's Schema Validator also checks whether FAQPage markup matches the visible FAQ. The Schema Markup Guide has the workflow.

None of them can tell you whether an AI engine will cite you. For that, look at the measurement options in SEO for AI Search.

FAQ

Does schema markup help AI search engines?+
There's no evidence that it gets you cited. Google says AI Overviews and AI Mode need no special structured data, and an Ahrefs study of 1,885 pages that added JSON-LD found no major change in AI citations. No AI provider documents using schema to pick sources. Schema is still useful for Google rich results and for describing your pages accurately.
What is the difference between SEO schema and AI search schema?+
There isn't a separate AI search schema. Both use the same schema.org vocabulary. The difference is in the claims: rich results in Google are documented, while claims that schema raises AI citations, or that AI engines quote FAQPage text, have no published source. If your markup works for Google, you're not missing an AI layer.
Which schema types should I add first for AI search?+
None specifically for AI search. Add the types Google documents a use for on your kind of page: Article or BlogPosting for editorial content, Organization for your brand, BreadcrumbList for site structure, and Product, Recipe, Event or LocalBusiness where the page really is that. For AI visibility, crawler access and content in the HTML matter more.
JSON-LD, Microdata, or RDFa for AI search?+
JSON-LD. Google recommends it where your setup allows it, and it's the easiest format to maintain. Microdata and RDFa still work for Google. No AI search provider documents a preference, because none documents using structured data for citations at all.
How do I know if my schema is helping AI citations?+
You can't isolate it. Validators only tell you whether markup is correct. To see AI visibility, use the generative AI report in Search Console, the AI Performance report in Bing Webmaster Tools, GA4 referrals from chatgpt.com or perplexity.ai, and manual checks of your target questions. If you change schema, compare against pages you didn't change.
Does llms.txt replace schema markup?+
No, they do different things, and neither is documented as a citation factor. llms.txt is a proposed file that lists pages for language models. Google says it neither helps nor harms in its search. Schema.org markup describes a page for search features like rich results. Neither replaces crawler access and useful content.

Where to Start

If you want your schema in good shape without chasing myths, do this:

Validate your current markup

Run your top pages through Lumina's Schema Validator and the Rich Results Test. Fix syntax errors and missing required properties first.

Schema Validator →
Remove what doesn't match the page

Delete markup for content that isn't visible, and bring FAQ markup back in line with the visible FAQ or drop it.

Check FAQ sync →
Add the types Google uses

Article, Organization and BreadcrumbList on content sites. Product, Event, Recipe or LocalBusiness where they fit.

Schema Markup Guide →
Fix what AI search actually needs

Check crawler access for OAI-SearchBot, PerplexityBot and Claude-SearchBot, plus snippet directives. That's where AI visibility is won or lost.

GEO Readiness Check →
Measure AI visibility separately

Use Search Console, Bing Webmaster Tools and AI referrals in GA4. Don't read validator results as an AI visibility score.

GA4 Dashboard →

Check your structured data

Lumina's free Schema Validator checks JSON-LD syntax, required and recommended fields for Google features, and whether FAQ markup matches the visible page. Paste code or enter a URL, no signup.

Run the Schema Validator →