Dense Schema.org: why 22 types per page is the new standard
Twenty-two Schema.org types per page is no longer a geek SEO quirk. In 2026 it became the reference threshold for B2B GEO-native sites. Why and how to reach it.
1. Why Schema density is exploding
Schema.org has existed since 2011 and was long used minimally: Organization, WebPage, Article was enough for Google Rich Snippets. The arrival of LLMs in 2023-2024 changes the game. Models consume JSON-LD as raw material to understand a page’s nature, author, organisation, subject, target audience.
The useful threshold for LLMs is no longer five types, it sits between fifteen and twenty-five. Twenty-two types is the sweet spot observed on the GEO Rocket portfolio: dense enough to spell out each entity, not so dense that file weight becomes a problem.
Twenty-two Schema types per page is the 2026 GEO reference threshold. Below it, citability suffers. Above it, you hit diminishing returns.
2. The twenty-two types to embed
Here is the typical list for a blog post or category hub on the GEO Rocket portfolio. It is not dogmatic: it adapts to whether the page is editorial, transactional or conversational.
| Schema type | GEO role |
|---|---|
| Organization | Brand identity |
| Person | Author identity, sameAs Wikidata |
| WebSite | Root site |
| WebPage | Current page |
| Article or BlogPosting | Editorial nature |
| BreadcrumbList | Contextual navigation |
| ImageObject | Structured og:image |
| ItemList | Article sections |
| SpeakableSpecification | Voice-read selectors |
| Service | Business service |
| Course or EducationalOrganization | Educational corpus |
| FAQPage or QAPage | Structured questions |
| DefinedTerm | Glossary terms |
| ContactPoint | Contact channels |
| PostalAddress | Postal address |
| OfferCatalog | Offer catalogue |
| Offer | Individual offer |
| Review or AggregateRating | Ratings if applicable |
| VideoObject | Embedded video if any |
| AudioObject | Embedded audio if any |
| HowTo | Procedure if any |
| NewsArticle or OpinionNewsArticle | Press variant if any |
3. The @graph architecture, not separate blocks
The common mistake is stacking twenty distinct JSON-LD blocks, each in its own script. LLMs parse that structure poorly. The 2026 practice groups all types in a single JSON-LD with a root @graph and stable @id values.
The benefit is dual: one parse for the LLM, and the ability to cross-reference entities via their @id. A brand that points its Article to its Person author via {“@id”: “#person-lorenzo”} produces an entity graph that LLMs understand in one pass.
4. Density traps
First trap: duplication. If the same Organization repeats across five distinct blocks with different descriptions, the LLM cannot pick. Solution: single @graph with stable @id.
Second trap: weight. Twenty-two unfactored complete types can exceed 80 kB of JSON-LD. Practice uses $ref or @id to cross-reference rather than duplicate.
Third trap: non-compliance. Schema.org’s Schema Markup Validator (validator.schema.org) remains the reference verification tool. Every Schema deployed to production must pass the validator before publishing.
5. How to migrate in four weeks
Week 1, audit existing Schema and inventory present types. Week 2, design target @graph architecture with stable @id. Week 3, integrate on two pilot pages and pass the validator. Week 4, generalise to priority pages (homepage, category hubs, author pages, service pages).
Migration averages three to five person-days on a 100 to 300 page site. Typical citation rate gain observed after 90 days is four to nine points on Perplexity and two to five on ChatGPT.
Dense Schema.org is no SEO goldplating. It is the grammar LLMs use to understand a brand. No grammar, no sentence. No sentence, no citation.
source: PGSM v1.0 · dense Schema.org section
The PROEMA GEO diagnostic audits Schema density on fifteen criteria and delivers the exact list of missing types versus the standard. Recommendations in under ten minutes.