Why Has On-Page SEO Moved From Keywords to Entities?

TL;DR

1. On-page SEO used to mean putting keywords in the right places. 
2. Title, headings, first paragraph, density targets. This definition is no longer enough. 
3. AI systems do not read pages by counting words. They read pages by identifying real-world entities and the relationships between them. 
4. The brands that show up consistently in AI answers are those whose entity definitions are clear, consistent across all platforms, and connected to the right related entities. 
5. Keyword optimisation alone produces flat results. Entity optimisation produces compounding visibility.

‍

What does the shift from keywords to entities actually mean?

For years, on-page SEO was a placement exercise. The right keyword in the title. The right keyword in the H1. The right keyword in the first paragraph. Density targets. Synonym sprinkling. If you hit the placement checklist, you had a chance of ranking.

That era is closing. The new on-page discipline is entity optimisation, and it operates on a completely different logic than keyword-based SEO ever did.

An entity is a distinct thing in the world. A company, a person, a product, a concept, a place. It is not a word. It is the actual thing the word refers to. 

When AI looks at your page, it is not counting how many times you used a phrase. It asks what real things this page discusses and how they relate to each other.

Here is how the two approaches compare in terms of what they prioritise and what they actually produce.

This table shows why entity optimisation produces visibility that keyword optimisation cannot:

‍

What is the difference between a keyword and an entity?

The clearest way to see this is with an example most people already understand.

The keyword "apple" is just text. The entity "Apple Inc." is a technology company founded by Steve Jobs and headquartered in Cupertino that manufactures the iPhone and the Mac. The entity "apple" is a fruit that you eat, grows on trees, and comes in varieties like Fuji and Granny Smith. Both share the same spelling. Neither is the same thing as the other.

AI systems understand the difference without needing you to explain it. What they do need is for you to be clear which entity your page is about, and to surround that entity with the right relationships so the system can place it confidently in context.

Keywords are surface signals. Entities are structural ones. Search engines match the first. AI systems reason about the second.

‍

Why does keyword optimisation no longer produce reliable AI visibility?

Because modern AI systems interpret natural language rather than matching text strings. They understand context, semantics, and the relationships between concepts. They do not need your target keyword to appear three times in the first paragraph. They need to understand what your page is fundamentally about.

The shift produces results that look strange from a traditional SEO lens. An article with strong entity connections can be cited by ChatGPT even if it doesn't contain the target keyword in the title at all. An article stuffed with keywords but built on a weak entity structure can fail to appear in AI answers entirely.

Research analysing millions of AI citations between July 2024 and February 2025 found that generative AI traffic grew by more than 1,200% during that window. The brands that captured that growth were not the ones with the most keywords. 

They were the ones whose entity profiles were clear, consistent, and well-connected to the surrounding ecosystem. This is also why authority alone no longer guarantees visibility. A high-authority page with a fragmented entity profile can be skipped over by AI in favour of a lower-authority page with a clearer one.

‍

What does proper entity optimisation actually look like?

Entity optimisation has three core components, and each one builds on the previous.

The first is defining your entity clearly. What is your brand? What category does it belong to? What specific problem does it solve? What separates it from adjacent things? These are not marketing questions. They are machine-readable definitions that AI uses to place you accurately in its understanding of your category.

The second is connecting your entity to related entities. Your brand does not exist alone. It sits in relationship to your industry, your competitors, the problems you solve, the customers you serve, and the methodologies you use. The clearer those relationships are, the more context AI has to draw on when constructing answers that involve your category.

The third is consistency. Your entity description must be identical across your website, social profiles, knowledge graph entries, press coverage, and every other place your brand appears. Inconsistency creates entity fragmentation, which weakens AI's confidence in including you. 

The mechanism is the same as the one that determines which content enters the retrieval layer in the first place. Fragmented entities fail to enter the right retrieval contexts because the system cannot resolve which version of you to use.

Why does Wikidata matter more than most brands realise?

Wikidata serves as a primary source for Google's Knowledge Graph, which in turn feeds into the training sets and grounding sources used by major LLM systems.

This makes a Wikidata or Wikipedia entry significantly more than a vanity asset for large brands. It functions as a machine-readable validator of your entity's existence, category placement, and relationships to other recognised entities. 

A brand with a clean Wikidata entry has a structural advantage in AI visibility that no amount of on-page optimisation can replicate.

Three specific actions worth taking this month if you have not already:

1. Check whether your brand has a Wikidata entry. Search for your company name at wikidata.org and confirm what is currently published.
2. If no entry exists, evaluate whether you meet the notability criteria to justify creating one. Press coverage, founder profiles, and verifiable third-party citations are the foundation of eligibility.
3. If an entry exists, audit it for accuracy. Outdated entries actively hurt visibility because AI systems use them as primary source data.

How do you start applying entity optimisation to your existing site?

Begin with the pages AI reads most often to understand your brand: the homepage and the About page. Both should explicitly state what your company does, the category it operates in, the specific problem it solves, and what makes it different from similar offerings in the same space.

Add Organization schema markup to your homepage. Structured data is the most reliable way to tell machines your name, your type of business, your location, your founders, and your related entities, in a format they can parse unambiguously. Schema.org documentation covers the full specification.

Audit your brand name consistency across every property. Variations like "Acme Inc," "Acme Incorporated," "Acme," and "Acme Solutions" are read as four different entities by machines, regardless of how obvious the connection seems to humans. Pick one canonical version and enforce it across all places your brand appears, including social profiles, directories, and third-party listings.

Consistency is also what protects your brand across personalised AI answers that serve different user contexts. A clear entity definition stays stable across user histories. A fragmented one breaks under the pressure of personalisation.

What does this mean for how content strategy needs to evolve?

The shift this requires is structural rather than tactical. Stop treating on-page SEO as a keyword placement exercise. Start treating it as a machine-readable identity-building discipline.

When entity definitions are clear, connected, and consistent, you are not optimising a single page. You are building an identity that AI can confidently reference across any question type, any user context, and any platform. That identity is what makes visibility compound rather than decay over time.

Search engineering treats entity clarity as a foundational layer because everything downstream, retrieval, fusion, citation, depends on it. Pages without clear entity grounding fail at earlier stages of the answer-construction process, before any keyword-level evaluation even occurs.