JSON-LD schema for citations in AI engines: a practical guide

When a generative engine evaluates whether to cite you, it first needs to understand who you are: your exact name, category, location, what you offer, and who stands behind the brand. JSON-LD schema answers those questions in a machine-readable format. It’s the lowest-cost, highest-leverage component of a program of AI search visibility — and the one most Mexican websites implement incorrectly.

# The 5 types that drive citations, in implementation order

1. Organization / ProfessionalService — the root entity. It needs a stable @id (https://tudominio.mx#organization), exact name, address, telephone, email, and sameAs pointing to your verified profiles. Everything else connects to this entity.
2. Person — the faces of the brand (founder, specialists). Use jobTitle, worksFor referencing the organization’s @id, sameAs linking to LinkedIn, and knowsAbout for areas of expertise. This is the technical foundation of E-E-A-T.
3. Service — one for each service line, with serviceType, areaServed, and provider pointing to the organization’s @id. This helps the engine associate the brand with the exact category your clients ask about.
4. FAQPage — real customer questions with complete answers (40-90 words). Hard rule: only one FAQPage per page; duplicates invalidate rich results eligibility and confuse extraction.
5. Article / BlogPosting — for every editorial piece: headline, datePublished, author as a verifiable Person, and publisher with the organization’s @id. Without an identifiable author, the content carries less weight as a citable source.

# Complete minimal example (adapt and validate)

# The 4 errors that invalidate the effort

• Contradictions with the visible site: the schema lists one address and the footer another, two different phone numbers, or two LinkedIn profiles. The model detects the inconsistency and lowers its confidence in the entire entity.
• Duplicate entities without @id: each page declares a “new” Organization instead of referencing the same one with @id — fragmenting the graph.
• FAQPage duplicated by a template: the site-wide FAQ is injected on every page alongside each page’s own FAQ. Google treats this as structured data spam.
• Orphaned schema: marking up content that isn’t visibly present on the page. Besides violating Google’s guidelines, it does nothing to help LLMs extract information.

Always validate with Google’s Rich Results Test and the schema.org validator, then rerun your citation audit 60-90 days after deployment: the effect typically appears first in Perplexity and AI Overviews, which read the live web.

# Sources cited

• Schema.org · official vocabulary · Canonical definitions of Organization, Person, Service, FAQPage, and Article.

• Google Search Central · structured data guidelines · Eligibility and validation rules for rich results.