Somewhere out there, someone is trying to figure out what your company is all about, but they’re only getting bits and pieces. They might look at your website’s About page, but it’s not even finished. Or they’ll check out your LinkedIn profile, only to find that the job titles are outdated since the last reorganization.
Then there are those directory listings – and you’ve got two different phone numbers listed in three different places. And if they dig deeper, they might stumble upon an old Reddit thread from 2023 that’s no longer relevant. Or maybe they’ll come across a press release announcing a product you don’t even sell anymore. It’s like trying to put together a puzzle with missing pieces.
That image is essentially your identity. It’s what Google uses to answer searches for your brand, and it’s what a large language model looks at when someone asks for a recommendation in your field. You didn’t create most of it, and you can’t erase the parts you don’t like.
But what you can do is make the parts you do control very consistent, clear, and easy to understand, so that the other pieces of information stop sending mixed messages. This way, you can make sure that your brand is represented accurately and positively, even if you can’t control everything that’s out there. By being consistent and clear, you can make your brand’s image stronger and more reliable, which can help build trust with your audience.
That is entity SEO.
The story is old but the consequences are new
It is important to be absolutely clear about the timeline here, because a lot of current commentary presents this as a discovery of the AI era.
Google bought Metaweb, the company behind the Freebase database, in July 2010. The Knowledge Graph went live in May 2012 with more than 500 million objects and over 3.5 billion facts about the relationships between them, introduced by Amit Singhal with a line that has since been repeated into meaninglessness: things, not strings. By May 2020 Google reported roughly 500 billion facts covering around five billion entities.
So search stopped matching strings and started resolving things fourteen years ago. What has changed is not the mechanism but the tolerance for ambiguity.
In June 2025, across two closely spaced updates, Google’s Knowledge Graph contracted by 6.26%, removing more than three billion entities in a single week. The preceding twelve months had seen steady growth of 2.79%. The figures come from Kalicube’s daily tracking of a sample of entities and were reported in Search Engine Land; Google has confirmed nothing, so treat them as a measurement rather than an announcement. The direction seem clear though, as entities filed under the vague catch-all “Thing” type were cut the hardest.
What is also somewhat vague is the concept of “entity”
What is an entity?
An entity is something singular, uniquely identifiable and definable. A person, a place, an organisation, a product, a concept. Three properties separate it from a keyword, and Semrush sets them out cleanly.
- You can tell them apart: the company Tesla is different from the person Nikola Tesla, and you can figure out which one is being talked about by looking at the details and the situation, not just the name.
- It’s got certain characteristics that define it, like who started it, where it’s based, what kind of business it is, and what it makes or sells. These characteristics are like the threads that link it to other things.
- It stands on its own, regardless of the language being used. For instance, Nintendo remains the same well-known company whether you’re talking about it in English, Japanese, or Portuguese.
A keyword is the string of text someone type onto a search engine bar, while entities are the thing they meant. One’s content can carry the string in every heading and still fail to establish the thing (pretty common actually!)
This is where a popular strand of advice goes wrong. The version circulating on LinkedIn suggests that mentioning well-known entities near your brand builds a semantic network that raises a depth score and lifts you in AI results.
The thing to keep in mind, though, is that there’s no set standard to measure how well you’re doing in terms of semantic depth. And the way it works is actually the opposite of what you might think. Just because you’re associated with someone or something that’s well-known and respected, it doesn’t automatically give you more credibility. Instead, you earn credibility when other people consistently talk about you in a similar way. It’s not about how many big words or important-sounding names you can cram onto a page, either.
In fact, using too many irrelevant terms can actually make your content less focused and less effective. Using irrelevant entities can water down your message rather than add depth to it.
The work which actually works
Often, the reason entity optimisation projects stall is that teams start at the end. They chase a Knowledge Panel or a mention in ChatGPT before the underlying facts are stable, which is like commissioning a painting without knowing what to paint.
These are the stages of the “entity-optimization” sequence one should follow:
Start with what you own
This is the only part you can fix. Inventory every editable asset: the website, the corporate site, About and Contact pages, the footer, author profiles, social channels, Google Business Profile, marketplace listings, Wikidata where it applies. Then agree an internal source of truth for the brand name, description, address, phone number and leadership, and enforce it across all of them. This is unglamorous work with a boring failure mode. Being “ABC Marketing” on your homepage, “ABC Marketing Agency” on LinkedIn and “ABC Mktg” on a directory does not confuse a human reader for a second, and makes recognition needlessly harder for everything else.
Make the relationships explicit
It should be done page by page. Does each page connect with one clear entity? Does the surrounding content credibly support that ? Are internal links using consistent anchor text, and do topic hubs actually gather the depth you have published? Organization schema defines the brand, Article and Person schema corroborate the page and its author, sameAs stitches your properties together. On large sites, internal linking is still the most scalable improvement available and the most commonly neglected.
Prove the people are real
Schema and text help a system recognise your entity, but E-E-A-T signals decide whether it earns visibility. Recognition without trust means you are understood, yet ignored. Every person who matters needs one verifiable identity: a genuine author page, a matching LinkedIn profile, a publication history, stated credentials, a bio explaining why they are worth reading. Publish your methodology. Show the testing. Name the sources.
Go and earn agreement
Third-party evidence divides in two. Entity corroboration answers who you are: accurate brand mentions, consistent descriptions, correct directory and association listings. Topical corroboration answers what you are known for: commentary, interviews, bylines and citations that repeatedly tie your name to the subjects you claim. Claiming to be the leading voice on a topic the rest of the web does not associate you with is a losing position.
Conclusion
So, what you get from all this is a kind of map that shows your organization and all the things connected to it. This includes people, products, services, locations, and topics. All these connections are clearly labeled, like “founder of”, “based in”, “specialises in”, and “publishes”. And each connection has some kind of identifier or evidence attached to it. This map is called an entity map, and it’s like a big diagram that shows how everything is related to your organization. It’s pretty useful for getting a clear picture of what’s going on and how all the different pieces fit together.
You can create a basic version using just the data from Search Console and page queries. But for bigger sites, it’s better to use a special tool called named entity recognition through an LLM API. This tool can quickly sort through tens of thousands of rows and figure out what’s what. However, it’s not perfect and sometimes makes mistakes without warning, so it’s a good idea to double-check a sample by hand. If you’re working with multiple markets, it’s best to make a separate map for each one. This is because people in different places often think about things differently, and the differences can be really useful to understand.
Now, compare that to how different models and search engines portray you. Do you like what you see?
