Entity SEO: Why Google Ranks Things, Not Keywords

Strings vs things: what actually changed

In 2012 Google published a post titled “Introducing the Knowledge Graph: things, not strings.” It is the single most quoted and least acted-upon sentence in SEO.

The old model was lexical. A page containing “plumber Bangalore” eleven times was a page about plumbers in Bangalore. Matching was string matching, and the game was getting the right strings onto the page in the right density.

The new model is referential. Google maintains a graph of things — each with an identifier, a type, a set of attributes and a set of relationships to other things. When it reads your page, it is trying to work out which things in that graph the page refers to. Your keywords are evidence for that decision, not the decision itself.

This is why two things that used to be confusing now make sense. It is why a page that never uses the phrase “iPhone repair” can rank for it, and why a page that uses it forty times can fail to. And it is why you can rank for your own name without ever having optimised for it — or fail to, despite owning the matching domain.

The useful reframe: you are not trying to rank a page for a phrase. You are trying to get Google confident about what your page is about, and about what you are.

What counts as an entity

An entity is anything with a distinct identity that can be described and related to other things. Concretely:

• People — a named individual, with a job, a location, an employer, a set of published work.
• Organisations — companies, practices, brands, agencies, non-profits.
• Places — cities, buildings, regions, service areas.
• Products and services — a specific model, a named offering.
• Creative works — a book, a film, a course, an article.
• Concepts — “technical SEO”, “conversion rate”. Fuzzier, but still nodes in the graph.

What is not an entity: a keyword, a page, a URL, a search query. Those are all ways of pointing at entities. Confusing the pointer with the thing is the root of most bad entity advice — including the advice to “add your keyword to alternateName”, which is exactly backwards.

How Google decides two names are one thing

This is the mechanism everything else depends on, and it is worth understanding properly. Google is doing entity resolution: given several strings that might refer to one thing, decide whether they do.

It is a confidence problem, not a lookup. Roughly, confidence goes up when:

1. The claim is made explicitly. Your structured data says these names belong to one node. This is necessary and almost worthless on its own — anyone can claim anything.
2. The claim is corroborated by sources you do not control. Your LinkedIn, a conference bio, a podcast page, a guest article byline all say the same thing. This is where confidence actually comes from.
3. The claim is consistent over time. The same name, role and relationship, unchanged across years and platforms.
4. The relationships resolve. If you claim to work for a company, and that company’s own site names you, both nodes get stronger.
5. Nothing contradicts it. A stale bio with a different job title, a dead profile in your sameAs, two people with your name in the same city — each of these lowers confidence.

The practical consequence is uncomfortable: the half of entity SEO that matters most is off your own website. You can spend a day making your schema perfect and move the needle less than one properly-written author bio on a site Google already trusts.

The six signals you control

1. Naming consistency

Pick one canonical form of every name and use it everywhere it is displayed. Variants exist — platform handles force them — but they belong in machine-readable name fields, not scattered through your copy. If your site calls you three things in three places, you have given Google three weak claims instead of one strong one.

2. Structured data with stable identifiers

One Person node, one Organization node, each with an @id that never changes, repeated identically on every page. Every other schema block on the site references those @ids rather than redeclaring the entity. This is the difference between one confident node and 120 vague ones.

3. One canonical hub page per entity

For a person, that is a real about page — typed as ProfilePage, named by mainEntityOfPage from everywhere else. Exactly one. Two pages competing to be the profile page for the same person is a genuine own-goal, and a surprisingly common one.

4. Internal links with descriptive anchor text

The name as anchor text, pointing at the hub page. A byline that reads “More about Thejas Acharya” teaches the association; one that reads “click here” teaches nothing. Contextual links inside content are worth more than the same link in site furniture.

5. Off-site profiles, linked both ways

Your schema’s sameAs points out at your profiles; each profile links back to your site. The loop is what makes it corroboration rather than assertion. Same display name, same photo, same one-sentence description of what you do.

6. Co-occurrence in real content

Every time your name appears near your brand, your job title and your subject matter — in a post you wrote, a case study, an interview — that co-occurrence is evidence. This is the signal you cannot shortcut, and the one that ultimately decides the outcome.

Where structured data fits (and where it doesn’t)

Structured data is how you state the claim unambiguously. It is not how you make the claim true.

The minimum viable entity graph for a person-led business is smaller than most guides suggest:

Notice what is missing: no aggregateRating, no invented award, no foundingDate you cannot support. Schema you cannot back up is worse than no schema, because a contradiction actively lowers confidence.

Four ways entity work goes wrong

Treating variations as separate brands

If you publish as Acme Studio, AcmeStudio and acme_studio across platforms, those are one brand with three handles. Building three separate identities — three schema nodes, three about pages — splits the signal three ways. Declare one entity with three names.

Stuffing queries into name fields

“Acme Studio | Best Web Design Agency in Pune” is not an alternateName. It is a title tag that wandered into the wrong field. Name fields are for names.

Declaring relationships nobody else confirms

worksFor an organisation whose site does not mention you. sameAs a profile that 404s. alumniOf a university with no record. Each unconfirmed claim is a dangling edge, and dangling edges do not strengthen a graph.

Expecting it to be fast

Entity confidence is cumulative and conservative by design — that is what makes it resistant to manipulation. Snippets and titles can change within days of a recrawl. Genuine name recognition takes months of consistent, corroborated signals. Anyone promising a knowledge panel in six weeks is selling something.

How to tell whether it is working

Entity work is hard to measure because there is no “entity confidence” metric. These are the proxies worth watching:

• Branded query impressions in Search Console. Filter queries containing your name or brand. Rising impressions on the name itself is the cleanest signal available.
• Which page answers the name query. If someone searches your name, is your hub page the result? If your homepage or a random blog post outranks your about page for your own name, the hub is not established.
• Snippet quality on the hub page. A snippet stitched from your footer and nav labels means Google could not find a confident description. A clean one means it could.
• What AI assistants say. Ask a few “who is X” questions. Assistants surface entity associations quickly and are a useful early read on whether the claim has landed.
• Rich results and sitelinks. Not guaranteed, and not the goal, but their appearance usually means confidence has crossed a threshold.

FAQ