Structured data — via schema.org — makes explicit what your content is: a question, a product, an organisation, a procedure. That helps search engines and other parsers with recognition and classification. What it is not: a ticket into AI answers.
What structured data does and does not do
Google documents it itself, unambiguously: there is no special structured data that gets you into AI features, and markup is not required to appear in them. What structured data does deliver: supported rich results in classic search, less ambiguity about what a page is, and explicitly supplied data where a system supports it. Treat it as technical hygiene — well-kept, consistent, correct — and not as a lever that buys AI visibility. No AI channel documents a direct citation or recommendation effect.
The universal base
Four schema types are relevant for almost every organisation — each with an honest expectation attached.
Organization and Person
The basic definition of your organisation and your authors: who writes, on whose behalf, with what identity. The sameAs property unambiguously points to the same identity elsewhere — LinkedIn, Wikidata, official registers. That is valuable for recognition and disambiguation; how individual AI systems make use of it is not documented.
Article and NewsArticle
For editorial content: makes author, publication date and subject explicit. The standard for blog and knowledge articles.
FAQPage
Marks question-and-answer pairs explicitly. Do know: since August 2023, Google shows FAQ rich results mainly for well-known government and health sites. The markup remains valid and can be read by parsers elsewhere, but the real value sits in the visible FAQ content itself — self-contained questions with direct answers are citable material, with or without markup.
HowTo
For step-by-step procedures the same applies, more sharply: HowTo rich results were retired in September 2023. Markup may stay where it correctly describes the content, but expect no display or visibility advantage from it. The procedure itself — clear, step by step, in the user’s language — is what a system can use.
Sector-specific schemas
On top of these, dozens of sector-specific types exist. Which ones you use depends on what your content actually describes. A few examples, not exhaustive: for e-commerce Product, Offer, AggregateRating and Review; for financial services FinancialProduct with subtypes such as BankAccount, LoanOrCredit, MortgageLoan and InvestmentOrDeposit (note: an “InsuranceProduct” type does not exist in schema.org — model insurance offerings with existing types and properties; InsuranceAgency is an organisation type, not a product); for healthcare MedicalCondition, MedicalProcedure, Physician; for professional services LegalService, AccountingService, ProfessionalService; for local players LocalBusiness with its many subtypes; for education and knowledge platforms Course and LearningResource.
The full schema.org vocabulary counts hundreds of types. The trick is not applying everything, but choosing the right types for what your content actually is. One correct, sector-appropriate schema is worth more than five generic ones.
Implementation and validation
Start with your most valuable pages: home page, service or product pages, about and author pages. And validate on two levels, because they measure different things: Google’s Rich Results Test only checks markup that Google supports for rich results; for validation against the full schema.org vocabulary, use the Schema Markup Validator (validator.schema.org). Errors in JSON-LD do not all weigh the same — some invalidate an item, others only limit a property or feature. So validate syntax, vocabulary and the requirements of the specific consumer separately.
Maintenance
Structured data is not a one-off task. When products, prices, opening hours or conditions change, the markup must follow. Inconsistency between visible content and markup undermines your reliability — a product still priced at €99 in markup while the page shows €129 is a trust problem. And the most important rule of all: markup never claims anything the visible text does not say.