What an entity actually is, beyond the dictionary line
An entity is a thing Google can name, disambiguate and store: a person, a place, an organisation, a product, a concept. The textbook line says it is « a uniquely identifiable object », which is true and useless on its own. What matters operationally is that an entity has a stable identifier inside Google's systems, a Knowledge Graph node often mapped to a machine ID inherited from the old Freebase project, and that identifier survives the exact wording you use. « Citroën », « the French carmaker » and « a mass-market European brand » can all resolve to the same node. A keyword cannot do that. A keyword is a string; an entity is a referent.
This is the shift the top results all gesture at without nailing down: search moved from matching strings to resolving references. Google's own framing, from the 2012 « Things, not strings » announcement that launched the Knowledge Graph, has not changed in intent, only in depth. If your content mentions a brand, a model, an ingredient or a regulation, Google maps each mention to a node and then weighs how strongly your page is associated with that node.
If you want the ground-level version before we go deeper, this short primer frames entities well:
For a netlinking professional this reframes everything. A backlink is not only a vote of trust passing through an anchor, it is also a co-occurrence signal: the linking page sits next to your brand entity in a topical neighbourhood. A link from a page that Google already associates with your entity's field reinforces that association. This is why building recognised authority on a single subject beats scattering links across unrelated domains, even at equal DR.
How Google detects and uses entities in 2026
The pipeline is well documented if you read the right sources. Google runs natural language processing over your text, performs named entity recognition, then entity linking, also called entity disambiguation, to attach each detected mention to a Knowledge Graph node. You can watch a simplified version of this yourself: paste a page into the Google Cloud Natural Language API demo and read the « salience » score it returns for each entity. Salience is Google's estimate of how central an entity is to the document, and it is the single most useful number most SEOs never look at.
This walkthrough on connecting entities is a useful companion to the detection mechanics:
One patent is worth knowing by number rather than by blog summary: US8682892B1, « Ranking search results based on entity metrics » (patents.google.com), which describes how entities are scored on relatedness, notability and contribution. You do not need to reverse-engineer it, but it confirms the operational point: Google ranks the relationship between entities, not just the presence of words. Structured data accelerates this. A case study reported by Digital Applied, drawn from Schema App enterprise experiments, measured a 336% increase in click-through for a primary query after entity linking was added in structured data, with a 390% lift on a query variant. One case study is not a law, but the mechanism is sound: explicit sameAs and @id markup hands Google the disambiguation it would otherwise have to infer.
In 2026 this matters more because the same entity graph now feeds AI Overviews and LLM-driven answers. When a generative engine assembles a response, it pulls from entities it trusts, not from pages it merely indexed. Being a well-defined entity is now the price of entry for showing up across the spread of AI-generated answers, not a nice-to-have layered on top of classic SEO.
Where entity SEO sits against keyword SEO in a real operation
The honest answer to « what is the difference between keyword SEO and entity SEO » is that one is a tactic and the other is the terrain. Keyword SEO targets the strings users type; entity SEO targets the concepts those strings point to. You still need keywords: they are how you discover demand and how you title pages. But planning a site around a flat keyword list in 2026 leaves authority on the table. Planning around entities means organising content into clusters that cover a subject and its related concepts, so Google sees a site that owns a referent rather than one chasing isolated queries.
Concretely, keyword density is dead as a lever and entity salience replaced it. The question is no longer « did I repeat the term enough » but « is this entity unambiguously central to the page, and are the supporting entities present ». A page about cold brew that never mentions coffee, extraction time or caffeine concentration is thin in entity terms even if it repeats « cold brew » twenty times. A page that names the related entities reads as authoritative to the NLP layer.
This is also where Stringer fits, briefly and operationally. We run 28 owned French media in-house, which means when we place an editorial link we control the surrounding entity context: the article that hosts your link already lives in the right topical neighbourhood. That is a different model from a link marketplace where you rent a slot on whatever page will take it. Placing a contextual article inside an owned editorial environment is, in entity terms, co-citation engineering, not just link buying.
Tactics and tooling for the working SEO
Start with the things you control directly. Mark up your organisation with Organization or LocalBusiness schema, include sameAs pointing at your Wikidata item, your Crunchbase, your LinkedIn and your verified social profiles: these are the edges Google uses to confirm your entity. Keep your name, address and contact consistent everywhere, because inconsistent identity data forces a disambiguation Google may resolve against you. If you have any legitimate claim to a Knowledge Panel, a Wikidata entry you can honestly create is the cheapest lever available.
The tactical layer below maps closely to what follows:
For measurement, three tools earn their place. The Google Cloud Natural Language API demo for salience, as mentioned. InLinks or a comparable entity tool to map which entities your content covers against the ones the SERP leaders cover, which surfaces your missing concepts relative to the pages that already rank. And plain SERP observation: search your brand and watch whether Google returns a Knowledge Panel, the clearest public signal that you have crossed from string to entity. On the netlinking side, pacing acquisition around the entities you want to own over a quarter beats a one-off burst, because entity associations strengthen with repeated, varied co-occurrence, not with a single spike.
Common mistakes we see go wrong
The first is treating schema as a checkbox. Dropping Article markup with no sameAs, no author entity and no @id linking gets you valid structured data that says nothing about who you are. Validation passing is not the goal; disambiguation is.
The second is inconsistent identity. A brand spelled three ways, two different addresses, a logo that changes per platform: every inconsistency is a reason for Google to keep your node fuzzy. Entity building rewards boring consistency.
The third is the panic narrative that « SEO is dead ». It is not, and the data is blunt: Reboot Online's 2026 statistics report that 49% of business owners rate SEO as the best-ROI channel they have, and 58.7% feel that ROI is increasing. What died is the version of SEO that treated Google as a keyword-matching machine. The discipline evolved toward entities and intent; the professionals who kept reading « keyword » where Google now reads « concept » are the ones feeling the ground move. Content creation remains the scaling bottleneck, cited by 42.6% of SEO professionals (Foursets, 2026), which is exactly why a structured, entity-led editorial plan outperforms volume for its own sake.
The last mistake is chasing entity SEO while ignoring the link graph that still anchors authority. Entities and backlinks are not rival strategies. A well-defined entity with no authoritative links is a tidy node nobody trusts; a strong link profile attached to a blurry entity wastes signal Google cannot attribute. The operation that wins does both: clear identity, clean structured data, and links acquired in the right topical neighbourhood.