Schema markup for AI search: a 2026 reference

Classic Google ranking uses schema as a *rich-result* eligibility signal — get the star ratings, the FAQ accordion, the recipe carousel. For AI engines, schema is upstream of retrieval itself: the engine parses your FAQPage, lifts the Q&A pairs verbatim, and uses them to ground answers. A page with a well-formed FAQPage schema is dramatically more likely to be cited because the engine can extract a clean answer-snippet without having to re-parse the rendered DOM.

1. FAQPage — Q&A pairs lifted verbatim by AI engines. Use on every page with two or more user-facing Q/As.

2. HowTo — Step-by-step procedural content. AI engines preferentially cite these for "how do I X" queries.

3. Article with mentions + citation — Linked entities + sources. AI engines treat well-cited articles as authoritative.

4. DefinedTerm — Single-concept definition pages. Glossaries with DefinedTerm are the AI engine's favorite content type for definitional queries.

5. Dataset — Marks original research and survey results. AI engines prefer to cite primary sources, and Dataset is the structured way to claim primacy.

Organization (every page), WebSite (homepage), BreadcrumbList (every page), Product + Offer (commerce/pricing pages), Review + AggregateRating (review pages). These improve Google rich-result eligibility but don't materially change AI citation likelihood. Ship them anyway — they're cheap.

• Invalid JSON-LD. An unclosed bracket or missing @context and the entire blob is silently dropped by both Google and AI engines.
• Stuffing irrelevant types. Tagging a blog post with Recipe because it has steps is a quality signal *against* you.
• FAQPage with marketing content. AI engines downweight FAQs that read as ads ("How is Acme better than the competition?" with marketing copy). Use Q&A for genuine user questions only.
• Hidden schema. Schema for content that doesn't exist on the rendered page is treated as cloaking.
• No @id anchors. Without @id, multi-node graphs aren't linked; engines see disjoint islands of metadata.

Use:

• Google's Rich Results Test — https://search.google.com/test/rich-results for Google-eligibility.
• Schema.org Validator — https://validator.schema.org for strict spec compliance.
• Crawlmind — site-wide JSON-LD audit on every crawl. Surfaces both validation errors and "missing on pages where it would help" gaps.