What schema types matter most for AI search in 2026?

Six types form the core: Article (or BlogPosting/NewsArticle), Person (for the named author), Organization (for the publisher), LocalBusiness (for any geographically-bound business), FAQPage (matching visible FAQs word-for-word), and BreadcrumbList. HowTo when the page has visible numbered steps. Service when the page is a service-tier money page. The rest of schema.org's vocabulary is situational. These six are foundational.

Why use a connected @graph instead of separate JSON-LD blocks?

Google parses schema as a connected graph, not as isolated objects. When Article references Person via @id, and Person references Organization via @id, and Organization links to its sameAs cross-references, Google reads the entire entity relationship in one pass. Separate disconnected JSON-LD blocks force Google to re-parse and re-link, with lower confidence. The connected @graph pattern is consistently more effective at producing rich results and AI Overview citations.

Does schema markup directly improve rankings?

Indirectly. Schema does not raise organic rank by itself (Google has stated this). It enables rich results (FAQ snippets, breadcrumbs in SERP, sitelinks), which improve CTR. It improves Knowledge Graph entity recognition, which improves AI Overview and Gemini citation rate. The cumulative effect on visibility is significant; the direct ranking effect is not. Calling schema 'a ranking factor' is technically wrong, but it is the single most under-invested-in non-content optimization in 2026.

Should I use schema generators or write JSON-LD by hand?

Generators are fine for one-off pages, hand-written for cornerstone pages where the @graph relationships matter. The biggest issue with generators is they produce isolated blocks rather than connected @graph patterns. Schema App, Merkle's generator, and Yoast's schema module all output usable schema; the more complete @graph relationships still need manual review. Validate every output through Google's Rich Results Test before deploying.

Structured Data Cheatsheet for AI Search

Quick Answer: AI search engines weight schema markup heavily for entity grounding and citation eligibility. The six schema types that matter most: Article, Person, Organization, LocalBusiness, FAQPage, and BreadcrumbList, all connected as a single @graph rather than separate JSON-LD blocks. Add HowTo for tutorial pages, Service for money pages. Validate via Google's Rich Results Test before deploying.

Most websites that have schema markup have it wrong. Either the wrong types are implemented, the types are not connected as a coherent @graph, the FAQ schema does not match visible content (a Google penalty trigger), or the @id values are inconsistent across pages so the entity graph never resolves. This cheatsheet is the corrective: which types matter, how to connect them, what to validate, and where the common pitfalls lurk.

The 6 schema types that matter most for AI search

Foundational schema types (in this order of priority)

Article (or BlogPosting/NewsArticle): every content page. Carries headline, author, publisher, datePublished, dateModified, image, inLanguage.
Person: the named author byline. Carries name, jobTitle, worksFor, knowsAbout, sameAs cross-references (LinkedIn, social, Wikipedia if applicable).
Organization: the publisher entity. Carries name, url, logo, address, founder, sameAs.
LocalBusiness: for any geographically-bound business (use the most specific subtype, e.g. ProfessionalService, Dentist, FoundationContractor). Carries address, geo, openingHoursSpecification, priceRange, aggregateRating.
FAQPage: when the page has a visible FAQ section. mainEntity array with Question/Answer pairs that match visible text word-for-word.
BreadcrumbList: navigation hierarchy for the page. Establishes site structure for both Google and AI engines.

Add HowTo when the page has visible numbered steps. Add Service on service-tier money pages with provider, areaServed, offers.

Why a connected @graph beats separate JSON-LD blocks

Most schema implementations use multiple `