Google’s Knowledge Graph, LLM retrieval mechanisms, and ranking algorithms are not the same system. Most entity optimization advice treats them as one. This conflation leads to wasted effort because each requires fundamentally different tactics.
Knowledge Graph Entry: The Wikipedia Problem
Your brand entity exists in Knowledge Graph only if Google has imported it from trusted structured sources. Schema markup does not create KG entries. It connects your page to an existing entity. The actual sources: Wikipedia, Wikidata, Crunchbase, government business registries, major industry databases.
Test this yourself. Search your brand name. If no Knowledge Panel appears, schema on your site changes nothing. You need external entity establishment first. Local businesses get this through Google Business Profile. Everyone else needs Wikipedia notability or equivalent institutional presence.
LLM Citation: Retrieval Beats Entity Structure
When ChatGPT or Perplexity cites a source, entity consistency plays almost no role. RAG systems chunk your content, embed those chunks, and retrieve based on semantic similarity to the query.
What actually drives citation:
- Assertive statements over hedged language. “X costs $500” gets retrieved. “X might cost around $500 depending on factors” does not.
- Numerical specificity. Concrete figures survive summarization.
- Claim density per chunk. More extractable facts per paragraph means higher retrieval probability.
Entity salience optimization is orthogonal to this. A page with perfect entity structure but vague prose loses to a messy page with quotable claims.
Google Ranking: Entity Understanding vs. Entity Ranking
Google uses entity recognition for query interpretation, not ranking. When someone searches “apple,” Google determines whether they mean the company or fruit. This affects which results are eligible, not which eligible result ranks first.
Ranking still runs on links, user behavior, and topical authority. A site with weaker entity signals but stronger backlink profile outranks the reverse. Entity optimization provides diminishing returns once basic disambiguation is solved.
Schema Markup: When It Matters
Schema affects ranking in exactly zero documented cases. Google has stated this repeatedly. What schema does:
- Rich snippet eligibility. FAQ, HowTo, Product schema can trigger enhanced SERP features. This improves CTR on existing rankings.
- Disambiguation for ambiguous entities. If you write about Mercury the planet, Mercury the element, and Mercury the Roman god on different pages, schema helps Google classify correctly.
- Edge case interpretation. Niches where Google’s NLP models have limited training data benefit more.
For mainstream topics, Google’s BERT and MUM extract entity relationships from unstructured text. Schema becomes redundant signal.
The Salience Mechanism
Entity salience is computed through:
- Syntactic position. Subject of sentence > object of sentence > prepositional phrase.
- Document position. Title > H1 > first paragraph > body text.
- Coreference chain length. How many pronouns and references resolve to this entity.
- Frequency relative to document length.
Practical application: Place your primary entity in sentence subject position. “Nike designed this shoe for overpronation” registers higher salience than “This shoe was designed by Nike for overpronation.” Avoid pronoun substitution in the first 200 words.
Over-optimization triggers quality filters. Natural co-occurrence patterns exist in Google’s training data. Deviation flags manipulation.
AI Overviews: Different Selection Criteria
Appearing in AI Overviews requires:
- Existing top 10 ranking for the query. AIO sources from organic results, not the entire index.
- Factual consistency with other top results. Contradicting consensus reduces selection probability.
- Statement extractability. Prose that can be quoted directly without modification.
Entity optimization is downstream here. You need ranking first. Entity signals might marginally help interpretation, but they don’t compensate for weak traditional SEO.
The Real Shift Nobody Discusses
Informational queries are migrating to AI interfaces. “Best running shoes for flat feet” increasingly gets answered by ChatGPT without a click. The queries remaining in traditional search are navigational and transactional.
Entity optimization matters most for transactional queries where brand recognition drives clicks. Nike’s entity graph richness helps when someone searches “Nike Pegasus buy” because entity understanding confirms commercial intent. It does nothing for “best stability shoes review” where Google deliberately excludes brand sites as biased sources.
Focus entity work on branded and transactional terms. For informational content, optimize for LLM retrieval characteristics instead: quotable claims, specific numbers, assertive tone.
Topical Coverage vs. Entity Relationships
Connecting pizza to dough fermentation to New York style on your page does not teach Google these relationships. Google already knows them from Knowledge Graph. What you signal is topical coverage: this page addresses multiple facets of the pizza topic.
Topical coverage works through internal linking architecture, not individual page entity density. One comprehensive page underperforms a cluster of interlinked focused pages. The mesh structure creates more entry points and distributes ranking signals.
Build clusters. Link contextually. Entity relationship building is Google’s job, not yours.