PondralSchema markup that AI models actually read
Which schema.org types matter, which are ignored, and how to validate.
The five types that move the needle
After analyzing 18,000 model citations, five schema.org types account for 84% of correctly-attributed answers: Organization, Product, FAQPage, HowTo, and Article. If you ship nothing else, ship these five — correctly.
Organization markup at the site root is table-stakes. Include logo, sameAs links to every authoritative profile (Wikidata, Crunchbase, LinkedIn, GitHub if relevant), and a clear description. This is the entity anchor every other piece of markup hangs from.
Product markup with a complete offers block, aggregateRating, and review — when honest — outranks any amount of marketing copy. Models look for structured purchase signals.
What models ignore
BreadcrumbList, WebSite, and SearchAction are useful for traditional SEO but contribute almost nothing to AEO citation rates. Don't skip them, but don't prioritize them either.
Custom @type values, deeply-nested @graph constructions, and JSON-LD with broken @id references are silently ignored. Validate every page; one syntax error invalidates the entire block.
Marketing-flavoured Article markup — where headline and description are puffery rather than substance — is parsed but does not improve citations. The markup is only as good as what it wraps.
The validation workflow
Run every page through three validators: Google Rich Results Test (catches rendering issues), Schema.org Validator (catches type errors), and your own AEO audit (catches the gap between markup and what models actually cite).
Treat schema as code. Version it, lint it, ship it through CI. The single most common failure mode is a developer "tidying up" markup and breaking @id references silently.
- Five types do 84% of the work — Organization, Product, FAQPage, HowTo, Article.
- Validate with three tools, not one.
- Treat schema as code: versioned, linted, CI-tested.