In short: An entity is a thing a system can identify, not a string it can match. Before anything can recommend you, it has to settle which "you" is meant, what kind of thing you are, and what you sit next to.
Those three decisions happen before retrieval, and mostly from what the model already holds rather than from your website. If they resolve to nothing, or to the wrong thing, the rest of your work never gets reached.
This is the layer that is slowest to change and hardest to fake, which is exactly why it is worth understanding before spending a quarter on page templates.
A String Is Not A Thing
Search engines spent two decades getting better at treating text as more than characters. A query for "jaguar speed" is not a request for the fastest sequence of those letters. It is a request about one of two things, and which one decides the whole answer.
The systems behind AI answers inherit that work and lean on it harder. They are not matching your page against a query. They are trying to decide what the question is about, and then what it can say about that thing.
An entity: a distinct thing with attributes and relationships, which a system can tell apart from other things with similar names. Your company is an entity. Each product is a separate entity. You are one. So is every competitor you get compared to, and so is the category all of you sit in.
The practical consequence is that your brand name is not the unit of work. The unit is whether a system holds a coherent, unambiguous record of what that name refers to.
Three Decisions, In Order
Resolution is not one judgement. It is three, and failing any of them ends the process.
Which one
Ambiguity is usually self-inflicted. A company that calls itself Acme, Acme Inc, Acme Software and getacme.com across its own properties has produced four candidate strings for one thing, and left the work of merging them to a system that has no reason to try.
Genuine collisions exist too. If a well-known band, a town and a chemical share your product name, you are not disambiguating a variant, you are competing for a slot against things with far more written about them.
What kind
This is the step that decides whether you appear at all. A question like "what should I use for X" is answered by first assembling the set of things that are an X. If you are not in that set, nothing about your pages matters, because your pages are never retrieved.
This is why a brand can be described accurately when named and be entirely absent when the category is asked about. Those are two different failures with two different fixes, and the second one is not fixed on your website.
Next to what
Association is what lets a system say "similar to", "an alternative to" or "typically used with". It is built from how other people write about you, not from how you describe yourself, which is why it is the slowest of the three to move and the most durable once it has.
Where The Answer Comes From
The lesson on how models work drew a line between what a model absorbed during training and what it retrieves at question time. Entity resolution sits mostly on the first side of that line.
Your parametric presence is what the model can say about you with no search at all. It is formed over years, from the general written record, and it is frozen at the knowledge cutoff of each model release.
Why this matters more than it looks: a system that does not resolve you at all cannot search for you, because it never writes a query containing your name.
Retrieval can only find you once something has decided you are relevant. That makes the entity layer the one place where doing nothing is genuinely expensive, and where the cost of having done nothing arrives years late.
What Actually Breaks It
Almost every entity problem that can be fixed from your side is one of three things, and none of them are technical.
Name variation you created, and never consolidated. Category language nobody outside your company uses, so the classification step has nothing to attach you to. Relationships that are implied by your product but stated nowhere, so association has no material to work from.
The next lesson is about removing each of those. The one after it is about the part you cannot fix on your own site, and how to measure where you currently stand.
Key Takeaways
- An entity is a thing a system can identify, not a string it can match. Your brand name is not the unit of work; whether it resolves to one coherent record is.
- Resolution is three decisions in order: which one, what kind, next to what. Failing any of them ends the process before your pages are considered.
- Absent from category questions but accurate when named is a classification failure, not a ranking failure, and it is not fixed on your website.
- Most of this sits in what a model already knows rather than in what it retrieves, which makes it slow to change and slow to lose.
- The fixable causes are self-inflicted: name variants you produced, category words nobody uses, and relationships you never stated.
Check yourself
Before you move on
Not scored, not recorded, and not part of the certificate. Both answers are settled by a sentence in this lesson, and the reasoning appears whichever option you pick.
- 01
An assistant describes your brand accurately when named, and never mentions you in answers to your category's buying question. What has failed?
- 02
Why is doing nothing about the entity layer expensive in a way that is easy to miss?