Schema markup for AI search: which types actually move the needle

Schema.org has hundreds of types. AI engines use a small subset to ground citations. The types worth implementing first, the ones to skip, and the implementation gotchas that quietly break otherwise valid markup. This is what to ship this week and what to ignore.

Key takeaways

• Five schema types do 80% of the work for AI search citations: Article, FAQPage, Person, BreadcrumbList, and sameAs links inside Organization or Person.
• Article schema is the baseline. Missing headline, datePublished, or author and you're not eligible for most rich-result or AI-citation paths.
• FAQPage schema still helps AI search even after Google deprecated FAQ rich results in 2023.
• Person schema with sameAs linking the author to LinkedIn, Twitter, and an author page is the highest-use attribution signal for AI engines.
• Most other Schema.org types (Product, Recipe, Event, JobPosting) help only specific verticals and don't materially affect AI citation rates outside those contexts.

Why schema matters more for AI search than for Google

Google has decades of crawling data and many signals beyond schema to determine what a page is about. AI engines have less of that history and lean more heavily on structured data to disambiguate entities, attribute content, and assess credibility.

That makes schema markup a higher-use investment for AI search than it is for traditional SEO. A page with complete Article + Person + FAQPage schema can outperform a similar page missing those types in AI Overview citations, even when the unmarked-up page ranks higher organically.

The five schema types that move citations

The five types that do the heaviest lifting for AI search.

Article: tells engines this is a written piece of content with a publication date and an author.

FAQPage: marks up question-and-answer pairs explicitly.

Person: declares the author as a real entity.

BreadcrumbList: establishes the page's position in a site hierarchy.

sameAs: the property that links your entity (Person, Organization) to authoritative third-party profiles.

These five together cover the attribution, freshness, and structure signals AI engines lean on. Most pages need only these; further schema types are vertical-specific.

Article schema: the baseline that gates eligibility

Article schema with headline, datePublished, dateModified, and author is the baseline. Google requires Article schema to include headline, datePublished, and author for rich-result eligibility (Google). AI engines use these fields to determine freshness and attribution.

The implementation: emit a single JSON-LD block per page with @type: "Article" (or BlogPosting / NewsArticle if more specific). Include the required fields plus image, publisher, and mainEntityOfPage for completeness.

The most common error is having multiple Article blocks on the same page with conflicting headline values, often because a CMS auto-generates one and a plugin emits another. Validate every page; pick the better source and remove the duplicate.

FAQPage schema: still useful after the 2023 deprecation

Google deprecated FAQPage rich results for most sites in August 2023 (Google), keeping them only for well-known authoritative government and health sites. Many publishers responded by removing FAQPage schema entirely. That was a mistake.

AI engines still parse FAQPage markup and use it to answer related queries. The rich result is gone; the AI citation signal isn't. Keep FAQPage schema on pages with substantial Q&A content.

The implementation: each Question has a name (the question text) and an acceptedAnswer with text (the answer). Match the question text to real searcher phrasing, ideally pulled from Google's People Also Ask block for the target query.

Person schema and sameAs: attribution that compounds

Person schema linked to your author byline, with sameAs pointing to LinkedIn, Twitter, and an author page, gives AI engines an entity to attribute. Bylined content with linked author entities gets cited at higher rates than anonymous content.

Implementation: emit a Person block for each author on the site, ideally on a dedicated author page. Reference it from the article's author field via @id rather than re-inlining the data per article. The sameAs array should include the author's professional profiles (LinkedIn at minimum), social profiles where they post under their own name, and any institutional affiliation page.

This is the schema type most under-invested. Most sites either don't have Person schema or have it without sameAs. Adding both is a small change with a measurable lift on AI citation rate.

The schema types that don't matter and can be skipped

Most of Schema.org doesn't apply to a content site. Skip Product unless you're an e-commerce site. Skip Recipe unless you're a recipe site. Skip Event unless you're publishing event listings. Skip JobPosting unless you're a jobs site.

For a content publisher, Article + FAQPage + Person + BreadcrumbList covers virtually all of the AI-citation lift you'll get from schema. Adding more types doesn't help; it just adds maintenance burden.

There's a temptation to layer on every applicable type. Resist it. A clean, validated, minimal schema graph performs better than a sprawling one with errors.

Common implementation mistakes that break valid markup

Three recurring mistakes.

First, multiple conflicting JSON-LD blocks. CMS-emitted schema plus plugin-emitted schema plus theme-emitted schema. Pick one source; remove the others.

Second, dates in the wrong format. datePublished and dateModified must be ISO 8601. 2026-05-21T10:30:00Z works; May 21, 2026 doesn't.

Third, missing required fields the validator doesn't always catch. Schema.org marks some fields as "required" only for specific use cases. Google's Rich Results Test catches most of these; run it on every template before shipping.

Validate using Google's Rich Results Test and the Schema.org validator. Both. They catch slightly different errors.

Your next move this week

Pick your top 10 highest-traffic pages. Audit each one for Article + FAQPage + Person + sameAs. Add what's missing. Validate with the Rich Results Test. Ship.

FAQ

What is schema markup for ai search?

Schema markup for AI search is the structured-data layer (typically JSON-LD using Schema.org vocabulary) that AI engines parse to understand a page's content, attribution, and context. It's the same Schema.org vocabulary used for Google rich results, but the AI search use case prioritizes a smaller subset.

How does schema markup for ai search work in 2026?

AI engines crawl and parse JSON-LD blocks on each page. They use the structured data to disambiguate entities (who wrote this, which company is referenced), assess freshness (when was this published or updated), and identify question-answer pairs eligible to be cited verbatim in AI-generated answers.

Why does schema markup for ai search matter for SEO?

Because AI engines lean on schema more heavily than traditional Google search does. A page with complete Article + Person + FAQPage schema typically earns AI citations at a higher rate than an equivalent unmarked page, even when the unmarked page has higher backlink authority.

Do I need JSON-LD or can I use microdata?

Use JSON-LD. It's Google's recommended format, AI engines parse it cleanly, and it's easier to maintain because the markup lives in a single <script> block rather than woven through the HTML body. Microdata still works but offers no advantage and more complexity.

Does FAQPage schema still help AI search after 2023?

Yes. Google deprecated FAQPage rich results for most sites in August 2023, but AI engines still parse the markup and use it to answer related queries. Keep FAQPage schema on pages with substantial Q&A content.

How many schema types should I implement?

For a content site, four to five types cover everything that matters: Article (or BlogPosting), FAQPage, Person, BreadcrumbList, and sameAs links inside Person or Organization. Adding more rarely helps and increases maintenance burden.

Can schema markup hurt rankings if implemented badly?

Yes, in two ways. First, schema that contradicts the visible content (claiming an author who didn't write it, dates that don't match displayed dates) is a manual-action risk. Second, broken schema (invalid JSON, missing required fields) gets ignored entirely, so the implementation effort produces no signal. Validate.