Entities, not keywords
A keyword is a string. An entity is a thing: a business with a founder, a location, a set of services and relationships between them. Search engines have matched strings for decades; the shift that made entity SEO a discipline is that answers are now composed about things, and a system has to know which thing is meant before it can say anything about it.
DOCUMENTED Google’s own framing supports the distinction. It describes AI Overviews as “a core Google Search feature, like knowledge panels”, and knowledge panels are Google’s entity surface, assembled from its Knowledge Graph. Google’s advice for its AI features includes keeping Business Profile information up to date and making sure structured data matches the visible text, which are entity-consistency instructions rather than keyword ones.
The ambiguity problem
Take a fictional Perth firm called Meridian. There is a Meridian in construction, a Meridian in financial advice and a Meridian that sells outdoor furniture, and the financial one has left five partially overlapping records behind it:
- its website says “Meridian Advisory”;
- its Google Business Profile says “Meridian Financial”;
- the ABN register says “Meridian Advisory Group Pty Ltd”;
- a 2019 directory listing says “Meridian Financial Services”;
- an old news mention says “Meridian Finance”.
INFERRED A system asked to recommend a financial adviser in Perth now has a resolution problem before it has a recommendation to make. This publication’s working model of the failure, which it labels as inference because no platform documents its confidence logic, is omission through uncertainty: several partial records, no confident way to merge them, and a competitor with one consistent name and address is easier to name. The visibility index is designed to test whether consistency predicts being named, rather than assume it.
How a knowledge graph represents a business
A knowledge graph stores an entity as a node with attributes (name, legal name, address, founding date, description) and edges to other nodes: the people who work there, the services it provides, the places it operates. The edges are what let a system answer a question it was never directly told. Asked who leads a service at a firm, a system with the edges traverses from service to organisation to person and produces a name. A system without them cannot, even if all three facts appear somewhere on the site in prose.
INFERRED Confidence in each attribute comes from agreement. If the website, the ABN record, three directories and an industry publication all state the same founding date, the attribute is reliable. If two disagree, none of them is. Google does not publish how it weighs corroboration; the model here is inference from how entity resolution works in general.
What unambiguous looks like
Four things, in order of importance:
- One canonical name. Pick the exact form, write it down, and use it on the site, every profile, the ABN record where possible, and the email signature. Variants are the most common and most fixable problem.
- One consistent set of facts. The same address form, phone format, founding year, service names and description everywhere. Consistency beats elegance; an awkward description used everywhere is worth more than a polished one used in three variations.
- Explicit relationships. Name the people, state their roles, connect them to the services they deliver and the organisation that provides them, in visible text and in matching structured data.
- Third-party agreement. Independent sources stating the same facts. It is the slowest part, the part the business controls least, and, on this publication’s reading, the part that carries the most weight for exactly that reason.
Entity SEO versus keyword SEO
| Keyword SEO | Entity SEO | |
|---|---|---|
| Optimises for | Matching and ranking for a query | Being identified as one entity |
| Core unit | The page and the phrase | The business, its people and its services |
| Off-site work | Links and authority | Corroboration and factual agreement |
| Success looks like | A position for a term | A system describing the business correctly without hedging |
| Failure looks like | Ranking below a competitor | Being omitted from the answer |
| Who controls it | Largely the business | Largely other people |
INFERRED The two are not rivals. A well-optimised page can rank respectably and still contribute nothing to how an assistant describes the business, because ranking and identification use different inputs. Entity work is the pass that makes the content programme accumulate to a recognisable business rather than to a URL.
How entity SEO applies in Perth
DOCUMENTED The sources a Perth business is described on are Australian: ABN Lookup and ASIC registers, its Google Business Profile, national directories, state and industry bodies, review platforms and professional associations. These exist and describe businesses; that much is documented. Which of them AI systems retrieve for Perth prompts is the subject of the most cited domains study and is not asserted here.
Two checks take five minutes. Ask an assistant, in a fresh session, what the business does, where it operates and who runs it; a hedged or confused answer means the entity is not resolving, and the check should be run more than once because answers vary. Then search the business name and read how the top results describe it, ignoring its own site; if three sources give three descriptions, retrieval systems are working from that same disagreement. The entity consistency audit turns those checks into a full inventory.
What is documented and what is inferred
- Documented: Google’s instructions on structured data matching visible text and current Business Profile data; knowledge panels as Google’s entity surface; the Australian registry and directory sources.
- Inferred: omission through uncertainty; corroboration as the weight-bearing signal; the ranking-versus-identification split.
- Not established: any measured relationship between consistency and being named in AI answers for Perth prompts, pending the observation programme.