Contents
Introduction
English version of an article first published in Polish on 7 August 2026. Read the Polish original.
Entities come back into SEO conversations every time a new layer of search appears, and every time the conversation ends with the same conclusion: "we need to implement structured data". That shortcut costs people a lot of time and does not give them what they took it for.
Because an entity is not code. Code is only a declaration. I can write in JSON-LD that I am an SEO business from Częstochowa, Poland, in five minutes. What makes me an entity is that the rest of the internet says so about me, and says the same thing everywhere.
I take the topic apart from the ground up: what an entity really is, how it differs from a keyword, what Google bases recognition on and what the hard data says. No promises that implementing schema will lift your rankings, because the search engine itself explicitly denies it.
You will learn:
- How an entity differs from a keyword and why that difference changes the way you write content
- The three signals Google relies on to recognise a business as an entity
- Why Google states plainly that you need no special schema to be visible in AI features
- What the Ahrefs study of 75,000 brands showed and how to read the numbers without fooling yourself
- How a Google patent measures entities and why a patent must not be read like documentation
- What Bing says about entities, and why it says it far more concretely than Google
- How to build a consistent identity for your business on the web, step by step
- How to check whether Google recognises your brand as a separate entity at all
What an entity is
An entity is a thing that can be unambiguously identified and told apart from other things. A business, a person, a place, a product, an event, a concept. It has an identity of its own, independent of the words you happen to use for it.
Google described this itself when it launched the Knowledge Graph in May 2012. It called it a model that understands real-world entities and their relationships to one another, and summed it up in a phrase that has stuck to the topic for good: things, not strings. At launch the graph contained more than 500 million objects and more than 3.5 billion facts about them and relationships between them.
It is worth noting that this is not a novelty of the last two years. Entities have been sitting in the search engine for well over a decade, and in that time they have been promoted from a curiosity to the layer that query understanding rests on. What has changed is that the same layer now feeds AI-generated answers, and that is why the topic is back.
What an entity is made of
Identifier
Something that tells this entity apart from every other one with a similar name. In practice: a website address, an entry in a knowledge base, a set of company registration data.
Type
The category the entity belongs to: business, person, place, product. The type decides what information the system expects in the first place.
Attributes
Its own properties: name, address, year founded, range of services. They are what fills a knowledge panel, if one ever appears.
Relations
Links to other entities: a city, an industry, people, clients. The most underrated part, and usually the hardest to build.
Entity versus keyword
This difference is the heart of the whole topic, so I will break it down with a concrete example. I live in Częstochowa, so I will take something everyone in Poland knows: Jasna Góra, the monastery that is the country's main pilgrimage site.
"Jasna Góra", "klasztor w Częstochowie" (the monastery in Częstochowa) and "sanktuarium na Jasnej Górze" (the Jasna Góra sanctuary) are three different strings of characters. For the search engine of years ago they were three separate matching problems. For a search engine that works with entities they are one entity described by three labels.
Keyword approach
three problems to solve
separate phrase = separate optimisation
Entity approach
one entity, many labels
Jasna Góra
place · Częstochowa · historic site
all three labels lead here
The consequence for writing content is concrete and rather liberating: you do not have to cram every variant of a phrase into the text to be matched. What you do have to do is write so that it is unambiguous which entity you mean, because the Polish word "zamek" can be a castle, a door lock or a zip on a jacket.
What this does not mean
Entities have not abolished keyword research. You still need to know what people search for and how, because that determines which content makes sense at all. What changes is what you optimise for: instead of thirty variants of the same phrase, clarity about what and whom the page is about.
How Google recognises an entity
Recognising a business as a separate entity rests on three signals. The order matters, because most implementations stop at the first one, which is the cheapest and at the same time the weakest.
JSON-LD code in which you say outright: this is a business, this is its name, this is where it is based, these are its profiles. The cheapest and fastest of the three signals, and on its own the weakest.
Mentions, descriptions and references on sites you do not control. They turn a declaration into something that can be checked independently.
The same name, the same description and the same profile URLs everywhere. Without that, the two previous signals describe two different entities instead of one.
You declare
structured data on your site
Others confirm
mentions outside your domain
← most drop out here
Google joins it up
one entity instead of scattered traces
Why schema alone is not enough
This is where it gets uncomfortable, because I have to say something part of the industry will not like. Structured data is not a ranking factor, and Google has been saying so consistently for years. It does not lift rankings. It makes you eligible for rich results and makes it easier for a machine to read what is on the page anyway.
For AI features, Google's documentation goes further still and says it outright: there is no special schema.org structured data that you need to add to appear in them. There are no new machine-readable files to create either. The same conditions apply as in regular search: the page has to be indexed and eligible to be shown with a snippet.
You can print "the best roofer in town" on your business card. What makes you an entity is that other people say so, and that they are talking about the same roofer.
None of this means schema is not worth doing. It means something else: it is a foundation, not a lever. You do not skip the foundation, but nobody expects it to put up the house by itself.
What structured data really gives you
Disambiguation
A lower chance of being confused with a business with a similar name at the other end of the country.
Eligibility
Access to rich results, which raise click-through rate without moving your position.
Anchor point
The place where you point to your profiles and allow them to be joined into one identity.
One caveat that Google gives separately and that deserves to be taken seriously: structured data has to match what is visible on the page. Describing in code services you do not offer, or reviews you have not collected, is not optimisation. It is a straight road to losing your rich results.
What really correlates with visibility
If not schema, then what? The most interesting material on this was published by Ahrefs in May 2025: an analysis of 75,000 brands and millions of AI Overviews. It checked which signals go hand in hand with a brand's presence in those answers.
The pattern is telling: the three strongest signals are things that happen outside your domain. Brand mentions came in at 0.664, branded anchors at 0.527 and search volume for the brand name itself at 0.392. The classic link metrics, referring domains at 0.295 and number of backlinks at 0.218, were clearly left behind.
How to read these numbers without fooling yourself
These are correlations, not causation, and Ahrefs itself points out that all the factors it studied showed moderate to very weak correlations. Brands that get written about are usually brands that have invested in their product and in PR, so cause and effect are hard to separate. Treat this as a pointer to the direction, not as a recipe.
The conclusion I take from this into my work is a cautious one: if I have to choose between another link from a directory and getting a client written about on an industry site without a link, I choose the latter. Five years ago I would have chosen the other way round.
What Google patents and Bing say
Sooner or later, two arguments come up in conversations about entities: Google's patents and what Bing does. Both are worth going through, but with very different weight, because one of them is much stronger evidence than the other.
The patent that measures entities
Google patented a way of scoring entities for search results: filed in December 2012, granted in March 2019. What is interesting about it is not that the system recognises entities, but that it tries to compare them with one another using four metrics.
Relatedness
How often the entity's name co-occurs on the web with its type. "Empire State Building" next to the word "skyscraper".
Notable type
The entity's popularity set against its position on the list of types. In other words: are you recognisable as what you are.
Contribution
Reviews, rankings and similar assessments, where highly rated ones weigh more in the score than average ones.
Prizes
Awards and distinctions, weighted by their standing: an Oscar counts differently from a local industry award.
The four values are added up with weights which, in the description, depend on the type of entity. That makes sense, because awards mean more for an actor than for a builders' merchant. The direction of thinking is telling: not "do you exist" but "how distinct are you within your category".
A necessary caveat
A patent is not proof that something works in the search engine. Google says so itself: according to Google's John Mueller, the mere fact that something is patented, even by someone who works on Search, does not mean it is used in Search. Other Google employees had said the same before him. I read patents as a pointer to a way of thinking, never as a description of the mechanism.
Bing says it outright
Microsoft has its own knowledge graph: Satori, launched in 2013 and conceptually a twin of the Knowledge Graph. Far more interesting, though, is what Bing wrote in February 2026 when it introduced its report on visibility in AI answers. Its recommendations for publishers include a sentence that Google does not have in this form:
Align text, images, and video so they consistently represent the same entities, products, or concepts.
Bing calls this reducing ambiguity across formats and adds that pages cited for specific queries often reflect clear subject focus and domain expertise. That is exactly the idea this whole article stands on, only said by a search engine, in a current document, without beating about the bush.
So the practical difference between the two sources is this: the patent shows how Google once thought about it, and Bing says what it expects today. If you are going to read only one, pick the second. I take the Bing reports themselves apart in a separate article.
Entities and AI answers
Entities came back into favour not because someone discovered something new, but because the entity-understanding layer started feeding answers generated by models. When an assistant composes an answer, it has to settle who is being talked about before it decides whom to cite.
A business whose identity is scattered across the web (one name on the website, another on its Google Business Profile, a third variant on its social profiles) makes that harder to settle. Not because the model "dislikes" it, but because from the model's point of view these could be three different organisations, none of which has enough confirmation.
Except that, and this is the finding that spoils a neat thesis, not every assistant behaves the same way. In the December part of the Ahrefs research, covering ChatGPT, AI Mode and AI Overviews, ChatGPT came out with the weakest correlations for almost every traditional brand authority signal. Branded anchors: 0.511 against 0.628 in AI Mode. Branded search volume: 0.352 against 0.466. Domain Rating: 0.266 against 0.285.
Who this is good news for
I read it this way: ChatGPT is less attached to established brand dominance than Google's products and reaches more often for brands with a varied profile on the web. For a small business that will never beat a corporation on recognition, that is the best news in this whole comparison. One last detail: the strongest single signal for ChatGPT turned out to be YouTube mentions, at around 0.737.
The practical upshot is that AI visibility today has two layers and both need to be handled. The first is content that can be cited, which I wrote about when covering the BLUF method and citations in AI Overviews. The second is an identity that can be recognised, which is exactly the subject here.
This can be measured, no guesswork needed
Whether assistants reach for your content today is something you can check for free in the citations report in Microsoft Clarity, which I take apart in a separate article. It is the only sensible starting point: first see where you are, then build.
How to build your own entity
The order below is not accidental. The first three steps are entirely under your control and cost nothing but time. Only the fourth depends on other people, and that is why it takes longest.
- 01Settle on one version of the name. With the legal form or without, with a space or with a hyphen: it does not matter, as long as it is one. Then stick to it everywhere, including invoices and your email signature.
- 02Build a page that anchors your identity. An "About us" or "About me" page with specifics: who, since when, where, what you do, whom you have worked for. Every other signal will lead to it.
- 03Add structured data with
sameAs. TheOrganizationorPersontype, full contact details and a list of the profile URLs you actually maintain. Not every one you have ever set up. - 04Put your Business Profile and other profiles in order. The same name, the same address, the same phone number, the same description of what you do. A mismatch in one place can break up everything else. More on this in my guide to the Google Business Profile.
- 05Earn mentions outside your own domain. Industry sites, podcasts, expert comments, directories with real editorial oversight. A mention without a link counts too, and that is the most important change in thinking of recent years.
- 06Wikidata, only once you have the backing. An entry based on independent sources helps. An entry forced into existence for a business nobody has written about will not build recognition that is not there.
Profiles are not a separate channel
Social media accounts are often treated as a completely different patch from SEO. From the entity point of view it is the same problem: every profile is one more place where your identity is either confirmed or pulled apart. That overlap is what I work on as part of Social SEO.
Myths and costly mistakes
"We'll implement schema and rankings will go up"
They will not, and that is not my opinion but Google's consistent position. What will go up is click-through rate, if you become eligible for rich results.
"Let's put everything we can into sameAs"
A list of dead profiles from a decade ago is not an identity signal, it is noise. Point to the places you actually maintain and where the data matches.
"We need a Wikipedia article"
Wikipedia has its own notability criteria and deletes articles that do not meet them. Forcing it can leave a trace in the form of a deletion discussion, visible in searches for your brand.
"An entity will make up for missing content"
Being recognised as an entity does not replace answering your customers' questions. Without content to reach for, identity alone sets nothing in motion.
These four mistakes have one thing in common: each of them tries to declare something that has no backing. A search engine does not examine declarations under a magnifying glass, but it does not build anything lasting on them either until someone else confirms them.
How to check whether Google sees you
There is no single button that displays the message "you are an entity". There are, however, four tests that together give a fairly good picture.
- 01Search for your own name. Are the results arranged around you, or do they mix you up with someone else? This is the simplest and most underrated test.
- 02Check for a knowledge panel. If one appears, the entity is recognised beyond any doubt. If not, that settles nothing, because panels go to entities with a high level of recognition.
- 03Look for yourself in Wikidata. Being there is not a requirement, but it can be a good indicator of whether there is anything about your business on the web worth citing.
- 04Look at the citation reports. Clarity and Bing Webmaster Tools show whether assistants reach for your content. I have covered those reports separately.
A separate note for those who were looking for a programmatic shortcut: the Knowledge Graph Search API still exists, but Google directs new implementations to Cloud Enterprise Knowledge Graph. The old version remains available but gets no new features, which is worth knowing before anyone builds a tool on it.
Limitations
- This is work for quarters. Consistency and structured data take a week. Confirmation in independent sources builds up over months, and a budget cannot speed it up.
- Google does not confirm the mechanism. We know how the search engine behaves and we have correlation data. We have no documentation describing exactly how the decision to recognise an entity is made.
- A small business will not get a knowledge panel. And it does not need one. The realistic goal is to be recognised unambiguously within your niche, not to have a separate card in the results.
- No activity, no entity. You cannot build recognition of something that leaves no trace beyond its own website. That is a business limitation, not a technical one.
None of these limitations undermines the point of working on your entity. They only undermine selling it as a quick optimisation, because it is not quick, and it is not optimisation in the classic sense either.
FAQ: common questions
How does an entity differ from a keyword?
A keyword is a string of characters, an entity is a thing. "Jasna Góra", "klasztor w Częstochowie" (the monastery in Częstochowa) and "sanktuarium na Jasnej Górze" (the Jasna Góra sanctuary) are three different strings describing one entity, and Google treats them as one entity, not as three separate phrases. The practical consequence is that you do not have to cram every variant into the text to be matched. What you do have to do is write content that makes it unambiguous which entity you mean.
Is structured data a ranking factor?
No, and Google has been repeating that for years. Schema does not lift rankings by itself. It makes you eligible for rich results, makes content easier for machines to read and helps settle which entity is meant. That is a lot, but it is not the same as higher rankings. If someone sells you a structured data implementation with a promise of higher positions, they are selling you something the search engine itself explicitly does not confirm.
Do I need special schema for AI to cite me?
No. Google says in its AI features documentation that there is no special schema.org structured data you need to add, and no new machine-readable files to create. The same rules apply as in regular search: the page has to be indexed and eligible to be shown. What is worth watching is that your structured data matches what is visible on the page, a point Google makes separately.
So why implement structured data at all?
Because it is the cheapest way to tell a machine outright something it otherwise has to infer from the content. It makes you eligible for rich results, lowers the risk of your business being confused with another one with a similar name, and gives a clear anchor point when your profiles are being joined into one identity. Treat it as the foundation the rest stands on, not as a lever.
Do I need an entry in Wikipedia or Wikidata?
You do not, and in most cases you should not force it. Wikipedia has its own notability criteria, and an article about a company that does not meet them will be deleted, sometimes leaving behind a deletion discussion you would rather not have in the results for your brand name. Wikidata is more open, but an entry with no backing in independent sources adds little. It is a result of recognition you have built, not a shortcut to it.
Where do I start if I run a small business?
With consistency, because it is the only element fully under your control and it costs nothing but time. Settle on one version of the name, one description of what you do and one set of contact details, then bring your website, your Google Business Profile and every profile you actually maintain into line with them. Only then add structured data and start working on mentions. The reverse order means describing in code something that is itself inconsistent.
How long does it take for Google to recognise my business as an entity?
The honest answer: it depends on how many independent sources talk about you, not on how quickly you ship the code. Consistent data and structured data take a week. The mentions that confirm them build up over months. For clients with a recognisable local brand the effects show after a quarter. For businesses nobody has written anything about it can take a year, and then the problem is not SEO but the fact that there is nothing to build recognition from.
Do Google patents explain how entity recognition works?
They show a way of thinking, not the mechanism. Google patented scoring entities with four metrics (relatedness, notable type, contribution and prizes), but a patent is not proof of implementation. Google says so itself: the fact that something has been patented, even by someone who works on Search, does not mean it is used in Search. I read patents as a pointer to the direction and never build a client recommendation on them.
What does Bing say about entities?
More than Google, and more concretely. Microsoft has its own knowledge graph called Satori, running since 2013, and in February 2026, when it introduced its report on visibility in AI answers, it recommended outright aligning text, images and video so that they consistently represent the same entities, products or concepts. Bing calls this reducing ambiguity across formats. It is a rare case of a search engine talking about entity consistency in plain text.
Does a knowledge panel mean I have an entity?
A knowledge panel is strong confirmation, but its absence settles nothing. Google recognises a great many entities for which it will never show a separate panel, because the panel is reserved for those with enough recognition. A better test for a small business is to check whether a search for your brand name returns results arranged around you, and whether they mix you up with another business with a similar name.
Key takeaways
- An entity is a thing, not a string of characters. Google has been telling the Jasna Góra monastery apart from the phrase "jasna góra" since 2012, when it launched the Knowledge Graph.
- Structured data declares identity. Only independent sources confirm it, and that is the whole difference between implementing schema and building an entity.
- Google says plainly in its documentation that AI features require no special schema. Structured data is not a ranking factor either.
- In the Ahrefs study of 75,000 brands, brand mentions correlated with presence in AI Overviews at 0.664, and referring domains at 0.295.
- Ahrefs itself points out that all the correlations it studied range from moderate to very weak. That is a pointer to the direction, not a recipe.
- In February 2026 Bing recommended outright that text, images and video should consistently represent the same entities. Google has written no such sentence: patents show a way of thinking, but they are not proof of implementation.
- Assistants differ from one another. ChatGPT showed the weakest correlations with traditional brand authority signals of the three surfaces studied, which happens to favour smaller businesses.
- You cannot declare an entity faster than the rest of the internet confirms it. This is work for quarters, not for a sprint.
Summary
Entities are often sold as a technical trick, when they are something far less glamorous and far more durable: making sure you say the same thing about yourself everywhere, and earning the same from others.
Structured data is a declaration. An entity begins where someone confirms that declaration.
Do one thing today: search for the name of your business and check whether the results are arranged around you or around someone with a similar name. It takes a minute and tells you more than many a structured data audit.
Just do not mistake it for the solution. That minute is only for making a diagnosis: it tells you whether you have a problem and nothing more. The work starts the next day and looks far less glamorous: unifying the name in a dozen or more places, tidying up profiles nobody has touched for years, and months of work so that anyone besides you has something to say about your business. There is no shortcut here, because the shortcut would be a declaration, and a declaration is where this article began.
Sources and further reading
The documentation and studies behind the data and positions cited here:
- Google Search Central: AI features and your website (documentation)
- Google: Introducing the Knowledge Graph: things, not strings (16 May 2012)
- Ahrefs: An Analysis of AI Overview Brand Visibility Factors (75,000 brands)
- Ahrefs: Top Brand Visibility Factors in ChatGPT, AI Mode and AI Overviews
- Google Search Central: introduction to structured data
- Patent US10235423B2: Ranking search results based on entity metrics (Google LLC)
- Search Engine Journal: Google on patents not always being used in Search
- Bing Webmaster Blog: Introducing AI Performance in Bing Webmaster Tools (10 February 2026)
- Google Cloud: migrating from the Knowledge Graph Search API
Not sure whether Google sees your business as one entity?
I will check how consistent the data about your business is across the web and show you where the identity comes apart.
See the SEO auditHow this text was written
I will not pretend it was written entirely by hand, by candlelight. For the research, the proofreading and the code of this page I used a language model. It is a good tool for digging through documentation and catching the typos you no longer see on the third read of your own text. The English version was also prepared with a model, from my Polish original.
What the model did not do: it did not choose the topic, form the arguments or decide what this data means for your site. That stays with me, as does the last word on what goes into the text and what comes out of it.
Where the model helped
- digging through sources and organising data
- proofreading and the English version of the text
- code, page layout and infographics
What stays with me
- the choice of topic and what is worth saying about it
- the arguments, the reading of the data and the conclusions
- fact-checking and the decision on what goes into the text