AI systems do not parse Schema.org markup at response time. But the markup still matters for AI Search because it feeds the discovery-layer signals search engines use to build knowledge graphs, and AI retrieval pulls from those graphs indirectly across every model in the candidate set. The same logic applies to semantic HTML and to entity-rich language at the sentence level. Each one carries indirect signals. They compound. This chapter covers the schema types that matter most for AI Search, the semantic HTML patterns that improve retrieval, the entity-rich language that helps AI systems attribute claims correctly, and the worked schema builds for the content types where this matters most.
Why This Technique Matters
The mechanism is layered, not direct. AI Search systems retrieve content that traditional search engines have already indexed, ranked, and tagged for rich results, which means traditional search shapes the surface AI systems pull from and the signals that affect traditional search end up affecting AI Search by the same mechanism. Schema markup, semantic HTML, and entity-rich language all work at this discovery layer. Retrieval itself is text-based. But the surface retrieval pulls from has been pre-curated by these signals. Schema matters here.
The practical effect is large. Pages with proper schema setup feed the knowledge graphs and entity signals search engines build, and they remain eligible for the rich-result features Google still supports, including featured snippets and Product rich snippets. The pages that carry those discovery-layer signals are the pages AI Overviews most often pull from (seoClarity February 2025: 97% of AIO citations come from the top 20 organic results, and the pages search engines surface most are over-weighted in that top 20). The AI model does not retrieve schema. Google retrieves the schema. Google then uses it to understand and surface the page. That makes the page more likely to appear in the candidate set the AI model retrieves from.
Entity recognition works the same way. AI systems sort out brands, people, products, and topics during retrieval. They use named-entity recognition signals to do it. Pages with consistent entity-rich language earn citations attributed to the right entity. Those signals are full canonical names, schema sameAs links, and consistent author and brand references across pages. Pages with vague references (the agency, the founder, the framework) earn a lower share. The model cannot tie the substance to a specific entity when the references are loose.
Semantic HTML carries the smallest discovery effect of the three. But the benefits stack up with accessibility and maintainability. Semantic elements signal content boundaries to crawlers, screen readers, and (indirectly) to retrieval indexes. The cost is near zero. Writing semantic HTML takes the same time as writing div-soup. That makes the small AI Search benefit pure upside.
Schema Types That Matter Most for AI Search
Five schema types produce the strongest discovery signals for AI retrieval. Most pages need only two or three. Site-wide coverage takes consistent setup across content types. For teams new to the markup itself, a complete guide to structured data for SEO covers the format, the syntax, and the validation basics this chapter builds on.
Organization Schema
The base entity for the brand. It goes on the homepage or a dedicated brand page. It sets the brand's canonical name, logo, contact info, social profiles, and sameAs disambiguation links. Without proper Organization schema, AI systems retrieve content about the brand but cannot tie it to the right entity. That scatters citation share across vague mentions.
The canonical setup includes a stable @id (a URL fragment like https://example.com/#organization that other schemas reference). It also includes the legal name, the operating name if different, the logo URL, contact info, and an address with full postal detail for local SEO. The sameAs array links to Wikipedia, Wikidata, Crunchbase, LinkedIn, X, and any other authoritative directories the brand appears on.
Person Schema
The backbone for named operators (founders, CEOs, SMEs, authors). The schema attaches the operator's bio, role, and authority signals to a stable @id. Pages that reference the operator (as author, contributor, or video creator) link back to the Person via @id. They do not restate the operator's traits.
Person schema's sameAs property matters most. It links the Person to the operator's LinkedIn, X (Twitter), Wikipedia article (where applicable), Crunchbase profile, and personal site. That set of links sorts the operator out in knowledge graph terms. AI systems retrieving the operator's content can then tie the substance to the right Person entity.
Article, BlogPosting, and Chapter Schemas
Content-level schemas that tie pages to their author and publisher entities. Article is the general type for editorial content. BlogPosting covers blog content. Chapter covers book-style content like the MERIT Framework Playbook chapters. Each one carries author and publisher references (usually to the Person and Organization schemas via @id) plus properties like headline, dateModified, and mainEntityOfPage.
The compounding effect with Chapter and Book schemas is large for playbook content. Each chapter's Chapter schema references the parent Book schema via isPartOf. That signals to discovery systems that the chapters form one body of work rather than stand-alone articles. The signal flows into knowledge graph structures that AI systems pull from at retrieval time.
FAQPage Schema
This schema has the most direct tie to AI Search citations. FAQPage markup turns question-answer pairs into machine-readable Q-and-A entities that feed the discovery and grounding layer AI systems pull from. The retrieval layer pulls those Q-and-A entities out cleanly. The structure matches how AI systems retrieve and assemble answers, which is what earns the citation rather than any Google search feature.
Every page with an FAQ section (per Chapter 7's structural rule) should carry FAQPage schema covering the questions and answers in the section. The setup is mechanical. Each question becomes a Question entity with a name property. Each answer becomes an Answer entity nested under acceptedAnswer with a text property. Every MERIT Playbook chapter carries FAQPage schema for its FAQ section.
HowTo and Product Schemas
HowTo schema covers step-by-step procedural content. It turns procedures into machine-readable Step entities that feed the discovery and grounding layer AI systems pull from. The sequenced, machine-readable structure helps the retrieval layer pull each step cleanly. Pages covering setup steps, recipes, or any task with sequenced steps benefit from HowTo markup.
Product schema covers products and services. It attaches Offer (pricing), AggregateRating (review scores), and Review (individual review content) sub-schemas. Together they produce Product rich results with pricing, ratings, and review snippets visible in search results. Comparison content benefits when multiple Product schemas appear on a page representing the products being compared.
Semantic HTML Patterns
Semantic HTML elements signal content structure to crawlers, accessibility tools, and (indirectly) AI retrieval indexes. The patterns are standard HTML5 practice. The AI Search benefit is modest. But it stacks on top of clear gains in accessibility and maintainability.
The page-level semantic skeleton. Header (site navigation and branding), Nav (the navigation block), Main (the page's primary content), Article (an individual article or chapter wrapping the main content), Section (logical content blocks within the article), Aside (extra content like the MERIT-map sidebars), Footer (site-wide footer). Crawlers and retrieval indexes read these elements as structural cues. They improve content boundary detection.
Content-level semantic elements. Figure and Figcaption for images with captions. Blockquote with Cite for quoted material. Time for dates and time references. Dfn for definitions. Mark for highlighted text. Code and Pre for technical content. Em and Strong for emphasis. Each carries meaning that pure visual styling does not.
Heading hierarchy. One H1 per page (the page's main title). H2s for major sections. H3s for sub-sections within H2s. The hierarchy reads cleanly to crawlers. It helps retrieval indexes grasp the content scope. Pages with broken order (H3 before H2, multiple H1s, skipped levels) confuse the parse and lower retrieval quality.
Lists for enumerated content. Ordered lists (ol) for sequenced steps. Unordered lists (ul) for non-sequenced groups. Description lists (dl/dt/dd) for term-definition pairs. Each carries meaning that lists styled as paragraphs cannot match. Chapter 7 covered the citation impact of lists. Semantic markup amplifies the effect.
Entity-Rich Language at the Sentence Level
This is the discipline that compounds with schema and semantic HTML. Entity-rich language references brands, people, products, and topics by their full canonical names. It does not lean on pronouns or generic phrases. The pattern signals to AI retrieval that the substance is about those named entities.
The Full-Canonical-Name Rule
First reference in any content unit (page, section, paragraph) uses the entity's full canonical name. Later references in the same unit can shorten. But they should keep enough specifics to hold the entity signal. For brand entities, the full canonical name uses the official spelling and capitalization (Searchbloom, not searchbloom; the MERIT Framework, not Merit framework). For person entities, the full canonical name includes the middle initial when the operator uses one (Cody C. Jensen, not Cody Jensen).
The rule works per section, not per page. A 5,000-word page with 8 H2 sections introduces the entity by full canonical name at least once per section. The pattern guards against AI retrieval that may pull a single section out without the page-level intro. The sentence-level discipline works best when each core entity also has a designated canonical page that holds its definitive reference; the anchor set is the small group of pages the rest of the corpus should point back to.
Disambiguation Through Specificity
Entities that share names with other entities need disambiguation cues in the surrounding language. A piece that mentions Apple without saying Apple Inc. (versus the fruit) may attribute correctly from context. Surrounding sentences may make clear which Apple is meant. But the same passage may attribute wrong when retrieval pulls it out alone. The discipline is to add enough specifics in the entity's local context that the disambiguation is clear.
For brands with common-word names, the cues include category indicators (Searchbloom, the SEO agency; the MERIT Framework for AI SEO). For people with common names, the cues include role and affiliation (Cody C. Jensen, CEO of Searchbloom). The cues do not need to appear in every sentence. One cue per section is usually enough.
Avoiding Vague References
Pronouns and vague phrases (the agency, the founder, our framework) carry weak entity signals. AI retrieval that pulls a passage with vague references may tie the substance to the wrong entity or fail to tie it at all. The discipline is to use entity-rich language where attribution matters. Pronouns are fine only when the prior sentence has the full entity reference.
The pattern is not absolute. A conversational tone benefits from pronoun variety. Pages that read as a list of full canonical names every sentence read poorly to humans. The working balance: every paragraph contains at least one full-canonical-name reference to the primary entity. Pronouns and shortened references fill the rest of the prose.
First reference in any content unit (page, section, paragraph) uses the entity's full canonical name.
The Entity Coherence Score
Entity-rich language at the sentence level needs a measurable diagnostic. Most brands write entity-rich on some pages and drift toward vague references on others. The drift is invisible without a number. The Entity Coherence Score is a Searchbloom-coined metric that captures how consistently the brand's primary entities are referenced across the site.
ECS = (count of correct canonical entity references) / (total entity references) x 100
Measurement workflow. Pull the brand's three primary entities (the Organization, the lead named expert, and the flagship product or framework). Run a site search or programmatic scan for every reference to each entity. Categorize each reference. Correct canonical references count toward the numerator (Searchbloom, Cody C. Jensen, the MERIT Framework). Vague or non-canonical references count against (the agency, the founder, our framework). Pronouns following a canonical reference in the same paragraph count as neutral. Pairing the ECS scan with an intra-site embedding audit shows whether pages that reference the same entity actually cluster together in embedding space, which is the retrieval-side proof that the sentence-level discipline is landing.
Reading bands:
- ECS above 95%. Strong entity coherence. AI retrieval attributes the brand's substance to the correct entity reliably. Knowledge graph signals consolidate around the canonical entity. Citation share attributes correctly across most pages.
- ECS 85 to 95%. Moderate coherence. Some pages drift. Drift usually clusters in older content, product pages written by the marketing team without an entity discipline, or pages written by guest contributors. Quarterly audits catch the drift and rewrite the worst offenders.
- ECS 70 to 85%. Scattered references. Entity attribution leaks across vague references. AI retrieval may attribute substance to no entity or to the wrong entity. The discovery-layer signal is weakened. A site-wide editing pass is the right move before adding more content.
- ECS below 70%. Critical drift. The brand's content base has fragmented references at a scale that prevents knowledge-graph consolidation. This is one driver of semantic relationship drift, where the connections between an entity and its references decay across the corpus over time. The fix is structural: editorial template updates, content-team training, and a full-corpus rewrite pass on high-value pages.
The ECS pairs with the Schema Coverage Index later in this chapter. ECS measures the sentence-level entity signal. SCI measures the markup-level entity signal. Together they describe whether the brand's entity work compounds or fragments. Most brands have one strong and one weak. Schema is mature but sentence-level discipline drifts (high SCI, low ECS). Or sentence-level is tight but schema is patchwork (low SCI, high ECS). Bringing both above 90% is the discipline that unlocks the compounding entity authority knowledge graphs reward.
Worked Schema Implementations
The base pair for any brand with a named expert as the public face. JSON-LD lives in the page head. The @id references let downstream schemas point back to the base entities without repeating the data.
Organization setup. @type Organization, @id https://example.com/#organization, name and legal name, logo URL with width and height, address as PostalAddress, contactPoint, and a sameAs array. The sameAs array holds Wikipedia, Wikidata, Crunchbase, LinkedIn company page, X handle, Facebook page, and any industry-specific authoritative directories.
Person setup. @type Person, @id https://example.com/#cody-c-jensen, name "Cody C. Jensen" (with middle initial per Searchbloom convention), jobTitle, worksFor pointing to the Organization @id, and image URL of the operator's professional photo. The sameAs array holds LinkedIn profile, X profile, personal site, Wikipedia article (where applicable), Crunchbase profile, and any speaker-circuit directories.
Compounding effect. Article schemas on content pages reference both @ids via author and publisher. The retrieval-layer entity graph sees Cody C. Jensen authoring content for Searchbloom across many pages. That strengthens both entities' authority signals.
The pattern used across every MERIT Playbook chapter. Each chapter's JSON-LD includes a Chapter entity (the chapter itself). It also references a Book entity (the playbook) via isPartOf.
Chapter setup. @type Chapter, @id with chapter URL fragment, name "Chapter N: Title", headline matching the chapter title, position N, datePublished and dateModified, author and publisher referencing Person and Organization @ids, and isPartOf pointing to the Book @id at the playbook home page.
Book setup on playbook home. @type Book, @id https://example.com/ai-seo/#playbook, name "The MERIT Framework Playbook", description, author and publisher references, and datePublished.
Discovery-layer effect. Search engines see the chapters as parts of one book, not as stand-alone articles. Citations to one chapter build the authority of the broader playbook entity. That lifts the retrieval odds for nearby chapters when the brand's broader content surfaces.
The schema with the most direct AI Search relevance. Every page with an FAQ section carries FAQPage markup. The markup mirrors the on-page Q-and-A content.
Setup pattern. @type FAQPage at the section level (not the whole page; the page holds other content beyond FAQs). The mainEntity property holds an array of Question entities. Each Question carries name (the question text exactly as it appears on the page) and acceptedAnswer (an Answer entity with a text property holding the answer body).
Mirror-the-page rule. The Q-and-A text in the schema must match the Q-and-A text visible on the page. Google penalizes schemas with content not visible to users. The schema is a machine-readable mirror, not an enhancement.
Citation effect. FAQPage schema makes the question-answer pairs machine-readable for the discovery and grounding layer AI systems pull from, which feeds AIO citations. AirOps measured FAQ-marked content earning +40% citation lift compared to similar content without schema. The schema is mechanical. The underlying content quality still has to back up the citation.
The Markup-Audit Workflow
Most brands have schema layered up over years in patchwork form. The audit-and-consolidation workflow brings the setup to a coherent state across the site, and it is a standard line item in a technical SEO engagement.
Step 1: Inventory existing schema. Crawl the site and pull every JSON-LD, microdata, and RDFa block. Note which pages have which schema types. Most audits find a mix of valid schemas, broken schemas, and pages with no schema at all.
Step 2: Validate against Schema.org and Google rich results. Run each schema through validators (Searchbloom's schema validator, Google's Rich Results Test, Schema.org's validator). Document errors and warnings. The most common errors are missing required properties, type mismatches, and broken @id references.
Step 3: Set the entity backbone. Build clean Organization and Person schemas as the base. Place these on the homepage and the operator's bio page. Every downstream schema references them via @id.
Step 4: Add content-level schemas in a systematic pass. Article and BlogPosting on every editorial page. FAQPage on every page with an FAQ section. HowTo on every procedural page. Product on every product or service page. A clean sweep produces site-wide consistency that compounds the discovery-layer effect.
Step 5: Verify entity coherence. The author and publisher references across all schemas should point to the same @ids. The Organization @id appears on every Article schema as publisher. The Person @id appears on every Article schema as author when the named operator wrote the content. Loose references scatter the entity signal across many ambiguous IDs.
Step 6: Refresh and maintain. Schema is not set-and-forget. The dateModified property needs updating when content changes. New schemas need adding when new content types appear. Quarterly audits catch schema drift and validation errors that pile up as the content base grows.
The Schema Coverage Index
The Schema Coverage Index measures how well the brand's schema setup covers the eligible content. Most brands have schema on the most-trafficked pages and patchwork or no schema on the long tail. The SCI is a Searchbloom-coined diagnostic that converts the patchwork into a single number.
SCI = (count of pages with valid schema for their content type) / (total pages eligible for schema) x 100
The denominator includes only pages where schema fits the content type. Privacy pages, contact pages, and login pages do not need content-level schema. They count against the eligible pool only as boilerplate (sitewide Organization schema in the head). The numerator counts pages where the schema validates against Google's Rich Results Test and matches the page's actual content type.
Reading bands by content-type segment:
The composite SCI across all content types is the headline number. Programs at 85%+ composite SCI typically earn rich results on 60%+ of eligible pages. Programs below 60% composite SCI rarely earn rich results consistently. The gap shows up in the AI citation share gap because the discovery-layer signal is missing for most of the content base.
Track SCI quarterly. Use the Markup-Audit Workflow above to identify gaps. Prioritize the audit fix work by traffic. The high-traffic pages without schema are the highest-impact SCI gains. Pair the SCI tracking with rich-result yield (the share of eligible pages actually showing rich results in Google search). Together the two numbers tell the brand whether the schema is correctly implemented and whether Google is using it.
The Schema Maintenance Calendar
Schema is not set-and-forget. Schema.org's vocabulary changes quarterly. New types arrive. Old properties get deprecated. Google rich-result expansions change which properties produce which features in search results. A schema setup built correctly will drift out of date without active care. Brands earning consistent AI citation share treat schema as a living system. They set a maintenance cadence rather than running a one-time technical build.
Brands earning consistent AI citation share treat schema as a living system.
The cadence has three layers. First, a quarterly audit. Second, an annual full-site audit. Third, a set of trigger-based audits that fire on specific events, no matter the calendar. Each layer catches a different class of drift before it compounds into the citation losses that surface well after the underlying schema breaks.
The Quarterly Schema Audit
The quarterly cadence catches validation drift before it spreads across the corpus. The audit covers four areas. First, validation status across the top 50 pages by traffic. Run each page through the Google Rich Results Test API (or an equivalent validator). Document validation pass, warning, and error counts. The top 50 pages usually account for 60 to 80% of organic traffic. Validation drift on these pages drives a large share of citation impact.
Second, deprecated property usage. Schema.org publishes a list of properties marked as deprecated in each release. The properties keep validating. Schema.org rarely removes properties outright. It deprecates them and signals migration paths instead. But search engines weight the deprecated forms less over time. The quarterly audit pulls any deprecated properties in production schemas and queues the migration work for the next round.
Third, new schema types that fit the brand's content patterns. Schema.org adds new types each quarter. Google announces new rich-result features that use specific types. HowTo schema, Recipe schema, Course schema, JobPosting schema, Event schema, and the long tail of category-specific types (Software Application for SaaS brands, Service for service businesses, MedicalEntity for healthcare, FinancialProduct for financial services) each open rich-result features the brand may not have looked at in the original schema scope. The audit reviews the brand's content inventory against the current schema-type catalog. It flags additions worth implementing.
Fourth, entity-graph coherence checks. The Person and Organization @id references should resolve the same way across the site. The audit pulls every Article, BlogPosting, Chapter, FAQPage, and other content-level schema across the top 100 pages. It verifies that the author and publisher @ids match the canonical Person and Organization @ids. Inconsistencies surface where a CMS migration, a developer handoff, or a content-team error introduced drift in the entity attribution chain.
The Annual Full-Site Audit
Once a year, a broader audit covers structural changes the quarterly cadence does not catch. The annual audit reviews sitemap updates against schema coverage; keeping sitemaps and schema markup in sync is its own discipline. Any new sections or URL patterns added since the last audit need schema. It checks that all content-type schemas line up with current content patterns. Content types that shifted in editorial direction may need new schema templates. The audit also reviews competitor schema setups for benchmarking. The competitor review is useful. Pulling the schema from the top three competitors in the brand's category reveals which schema types and properties the category as a whole is using. That signals what search engines have set as the discovery-layer baseline for the category.
The annual audit also updates the canonical schema templates the team uses for new content. The template document captures the current canonical setup for each content type the brand publishes (Article, BlogPosting, Product, FAQPage, HowTo, and any category-specific types). Content creators and developers reference the templates when adding schema to new pages. Without the template document, schema setup drifts over time. Different team members make different calls on edge cases.
Trigger-Based Audits
Some events should trigger schema review right away, no matter the calendar. Five events warrant an immediate review:
- Site redesigns. Any redesign affecting page templates affects the schema rendering and needs full validation across the new templates.
- Major content-type launches. A brand launching a podcast, a video library, a tool, or a research-report series needs schema for the new content type before the launch goes live.
- Schema.org vocabulary updates announced by Google or Schema.org. Significant releases sometimes introduce breaking changes or new required properties.
- Google rich-result feature launches. New rich-result types emerge often. Brands that act quickly often capture early visibility.
- Brand acquisitions or rebrand events that change entity attribution. A brand acquisition changes the Organization entity. A name change requires Organization schema updates plus sameAs updates across the entire web of brand references.
The trigger-based audits do not replace the quarterly cadence. They layer on top of it. A site redesign still requires the regular quarterly audit. The redesign itself triggers a separate audit scoped to its impact on schema rendering.
The Schema-Vendor Option
Schema App, Merkle, Yoast, Rank Math, and similar tools provide managed schema maintenance. The service covers quarterly audits, validation monitoring, and update rollout. The vendor market segments roughly by capability tier. Entry-level tools (Yoast, Rank Math) work as WordPress plugins or similar CMS extensions. They handle baseline schema for Article, BlogPosting, FAQPage, and Product types. Mid-market managed services (Schema App is the category leader) provide template authoring, validation monitoring, and entity-graph management. Enterprise managed services (Merkle and similar consultancies) provide custom schema builds, multi-domain entity graphs, and integration with broader knowledge-graph optimization programs.
The build-versus-buy call depends on three variables. Team engineering capacity drives the first answer. Brands without dedicated engineering bandwidth for schema infrastructure should default to buy. Site complexity drives the second. Simple sites with three or four content types can be maintained by hand through quarterly audits. Complex sites with dozens of content types and entity-graph needs benefit from vendor tooling that manages the work in a systematic way. The rate of content-pattern change drives the third. Brands publishing new content types or new sections often benefit from vendor support that handles new-type rollout without engineering work. Brands with stable content patterns and engineering capacity can build and maintain in-house at lower total cost.
The Schema-Versioning Pattern
For brands doing iterative work in-house, a versioning pattern guards against drift. The pattern is simple. Maintain a schema-templates document. It captures the current canonical templates for each content type. It records the date of last review. It lists any pending changes the team has flagged for the next round. Treat schema templates as code under version control. Require schema review on any content-type rollout the same way the team requires accessibility review or technical SEO review.
The versioning pattern catches the failure mode where a content-team member sees an existing schema on one page, copies it to a new page, and spreads an out-of-date template. The schema-templates document is the canonical reference. Copying from existing pages is forbidden by team workflow. Engineering reviews of new content-type launches verify the launch uses the current template version. The discipline is procedural, not technical. But it produces real long-run schema coherence.
Metrics That Matter for Schema Maintenance Health
Four metrics track schema maintenance health over time. Validation pass rate, with a target of 100% on top pages by traffic and 95%+ corpus-wide. Rich-result yield, with a target of 60%+ of eligible pages showing rich results in Google search. The eligibility threshold depends on content type. Not every product page is eligible for a rich result even with valid schema. Time-to-fix on detected validation failures, with a target of prompt resolution from detection to deploy. Schema-vocabulary currency, with a target of zero deprecated properties in production. Brands tracking these metrics monthly and reviewing them quarterly catch issues before they compound.
Common Mistakes That Defeat Schema and Semantic HTML Work
1. Schema present but not validating. The most common failure mode. The page carries JSON-LD with errors or warnings. Google ignores the broken schema. Counter-test: run your top 10 pages through Google Rich Results Test and document validation status.
2. Schema content not matching page content. The schema claims FAQs that do not appear on the page, or product prices that differ from the displayed prices. Google penalizes the mismatch. Counter-test: spot-check 5 pages where schema claims content. Does the schema-claimed content appear in the visible page?
3. Inconsistent entity references across pages. One page's Article schema attributes to "Cody Jensen." Another attributes to "Cody C. Jensen." A third uses an Organization byline. The entity signal fragments. Counter-test: pull the author references across your top 20 published pages. How many distinct attributions appear?
4. Missing sameAs depth on Organization and Person schemas. The schemas carry name and basic properties but no sameAs array linking to authoritative directories. The disambiguation signal is thin. Counter-test: how many authoritative external references does your Organization sameAs array contain (Wikipedia, Wikidata, Crunchbase, LinkedIn, X, industry-specific directories)?
5. Schema instead of, rather than alongside, semantic HTML. Brands set up schema and skip semantic HTML. They think schema replaces structural markup. The two work at different layers. They compound when used together. Counter-test: do your pages use article, section, aside, and figure semantic elements alongside JSON-LD?
6. Broken heading hierarchy. H3 before H2, multiple H1s, skipped heading levels. The parse confusion lowers retrieval quality. Counter-test: run your top 10 pages through an accessibility validator. How many heading hierarchy warnings appear?
7. Vague entity references in body content. The text refers to "the platform" or "the framework" without naming them. The retrieval extraction ties the substance to no clear entity. Counter-test: pick a 500-word section from your top page. How many times does the primary entity appear by full canonical name versus by pronoun?
8. Schema duplication. Multiple schema blocks per page that try to describe the same entity in different ways. The result is conflicting signals. Counter-test: how many JSON-LD blocks do your top 10 pages each contain?
Questions & Answers
Do AI systems actually parse schema markup directly? No, not at retrieval time. LLMs do not parse Schema.org markup during response generation. Schema is a traditional SEO signal. It feeds knowledge graphs that AI Search retrieval pulls from indirectly.
Which schema types matter most? Organization and Person for brand and named-author entities. Article, BlogPosting, Chapter for content attribution. FAQPage for question-answer content. HowTo for procedural content. Product with Offer and Review for products and services.
Why entity-rich language at sentence level? AI systems sort out entities through named-entity recognition over sentence content. Full canonical names produce stronger signals than vague references. Pages with consistent entity language earn citations attributed correctly at higher rates.
Should I use semantic HTML elements? Yes. Semantic elements signal structure to crawlers, accessibility tools, and (indirectly) AI retrieval. Modest AI Search benefit. Real accessibility and maintainability benefit. Near-zero cost to set up.
What is the sameAs property? A Schema.org property that links a page's entity to canonical references across the web (Wikipedia, Wikidata, Crunchbase, LinkedIn, etc.). It firms up entity disambiguation in knowledge graphs.
How does this relate to traditional SEO schema? It is the same discipline. What people call AEO or GEO is an evolution of SEO, not a separate practice, so the schema work that builds knowledge graphs for classic Search is the same work that feeds AI retrieval. Brands with mature schema need to extend to AI-relevant types and verify entity coherence.
JSON-LD or microdata? JSON-LD. Google recommends it. The broader ecosystem follows. JSON-LD lives in the page head, separate from content markup. That makes it easier to maintain.
Can I use a schema generator? Yes for most cases. Hand-build Organization, Person, and Chapter or Book chains. Generators handle Product, FAQPage, and HowTo boilerplate well.
How This Chapter Closes the Relevance Pillar
The three Relevance chapters together turn owned-domain content into a citation surface. It compounds with the Mentions and Evidence work. Chapter 7 covered the answer-first structural pattern that makes content retrievable. Chapter 8 covered the multi-format surface coverage that earns retrieval across text, video, image, and structured data. Chapter 9 covered the markup and language layer. That layer sorts out the underlying entities and signals content structure to discovery indexes.
The Inclusion pillar follows. Chapter 10 (Entity Optimization) deepens the entity work covered here. It treats brand, person, product, and topical entities as one unified optimization surface. Chapter 11 (Crawler Access) covers the robots.txt and bot-access setup that determines which AI systems can crawl owned-domain content at all. Chapter 12 (Indexing Protocols) covers two protocols. IndexNow covers the Bing-and-beyond ecosystem. Google's Indexing API covers the Google ecosystem. Both notify AI systems of new and refreshed content right away, not at the next crawl cycle.
