Structured data for AI search: which schema types actually matter

For most content sites, four schema types deliver nearly all the AI-search value: Article (or its Blog/News variants), FAQPage, HowTo, and BreadcrumbList — plus Organization once, site-wide. Everything else is situational.

Why schema matters differently for AI than for Google

Traditional structured-data advice was driven by rich results: implement Recipe schema to get recipe cards, Review schema to get stars. The payoff was visual real estate in search results.

AI systems use the same markup differently — as provenance and disambiguation. When a model considers citing your page, schema answers questions it otherwise has to guess: What type of content is this? When was it published and updated? Who wrote it? Which parts are the actual Q&As or steps? Confident answers to those questions make your content safer to cite. That shifts which types matter: visual-treatment types matter less, content-structure types matter more.

The priority list

1. Article / BlogPosting / NewsArticle — the workhorse. headline, datePublished, dateModified, and author give AI systems the freshness and provenance signals they weigh heavily. dateModified especially: in a fast-moving topic, a model choosing between two similar sources tends to prefer the verifiably fresher one. Generate it with our Article & HowTo Schema Generator.

2. FAQPage — marks Q&A pairs explicitly, matching the exact shape of AI-assistant queries. Google restricted its visual FAQ treatment in 2023, but the parsing value for AI systems is untouched (full analysis in our FAQ schema guide). Generate it with the FAQ Schema Generator.

3. HowTo — for genuinely instructional content, named ordered steps are gold: "how do I" queries are among the most common AI-assistant patterns, and pre-structured steps get extracted verbatim.

4. BreadcrumbList — cheap to implement and gives crawlers your site's topical hierarchy, helping models understand that a page lives under /guides/structured-data/ rather than being an orphan.

5. Organization (site-wide, once) — name, logo, URL, and social profiles establish entity identity: which real-world organization this site belongs to. Useful for brand-level queries and knowledge-graph association.

Situational types

• Product + Offer — essential for e-commerce, irrelevant otherwise.
• LocalBusiness — essential for physical businesses (address, hours), irrelevant otherwise.
• Event, JobPosting, Recipe, VideoObject — implement when your content genuinely is one of these; skip the temptation to shoehorn.

Types to skip

• Review/AggregateRating on your own content — self-serving ratings are widely ignored and can look manipulative.
• Speakable — narrow voice-assistant use case that never gained adoption.
• Redundant WebPage/WebSite markup on every page — adds bytes, not signal, beyond what one site-wide declaration provides.

Implementation rules that actually matter

1. JSON-LD, in one or a few script tags — not Microdata attributes scattered through HTML. JSON-LD is what Google recommends and what every tool generates.
2. Match visible content. Schema describing content that isn't on the page is the one reliable way to get markup penalized or ignored.
3. Keep dates honest. Auto-bumping dateModified without real updates is detectable (models compare content) and burns trust.
4. Validate before shipping. A missing required field silently voids the block.

Validate what you have

Run any page — or raw JSON-LD before you publish — through our Schema Validator. It unpacks every schema block on the page, checks required and recommended fields per type, and does deep structural checks on FAQPage and HowTo, separating hard errors from advisory warnings.