Knowledge graph

A knowledge graph is a store of entities, such as people, places, organisations, products and events, held as nodes with typed relationships between them, so that a system can reason over the relationship instead of matching text. Google launched its Knowledge Graph on 16 May 2012 with the framing "things, not strings". The announcement said it held more than 500 million objects and more than 3.5 billion facts about and relationships between them.

The difference from an index is the whole point. An index answers "which documents contain this phrase". A graph answers "what is this thing, and what is it connected to", which no single document has to have written down.

How do entities get into a knowledge graph?

By corroboration, not submission. Structured markup, reference sources, business listings and consistent descriptions across independent sites all act as evidence that an entity exists, has a type and has certain attributes. None of them is decisive alone, which is why "we added schema markup and nothing happened" is the expected outcome, not a bug. Markup is a machine-readable assertion, and assertions are weighed against other sources.

Reading is more limited than people assume. Google's Knowledge Graph Search API returns ranked entity matches typed with schema.org types such as Person, Organization, LocalBusiness and Event, and Google states in the same documentation that the API is not suitable for production-critical use. Treat it as a lookup, not infrastructure.

What happens when a brand name collides with another entity?

We have a live example in our own data. Attensira tracks the questions "what is Attensira" and "Attensira reviews" on ChatGPT and Perplexity for our own site, and both engines name us when asked: ChatGPT in all 33 of its answers to the first question in the 30 days to 4 October 2026. The sources around that name are less settled. The Perplexity answer at the top of this page, stored on 4 October, describes us correctly and lists help.attensi.com, the help centre of a similarly named organisation, as its first source, ahead of our own site. Its answer to "Attensira reviews" the same day cited help.attensi.com again, and bienestar.atenzia.com. ChatGPT's answer to "Attensira reviews" ended by asking whether the user meant a different Attensira, "for example a supplement". Pulling the other way, ChatGPT's 33 answers to "what is Attensira" cited e-krediidiinfo.ee, an Estonian company-information site that describes the registered company, 11 times.

That is an identity problem, not a visibility problem. The engines found us; they also found other things with nearly the same name and could not be sure which one the asker meant, so they hedged, or borrowed a source from the wrong one. A knowledge graph exists to settle exactly that: one node for the software company and one for each namesake, with types and attributes that keep them apart. Where the evidence for the company is thin, the system has less to separate them with.

It is also a warning about measurement. Attensira's own mention detection is a case-insensitive whole-word match with no entity resolution, as its limitations state, so an answer that only discussed a namesake would still count as a mention of us. We report that limit rather than hide it.

Why does entity identity matter for AI answers?

Because selection and attribution both depend on it. AI search systems retrieve passages and then attribute them to someone. If the entity is unresolved, both steps can go wrong in ways that look like a ranking problem and are actually an identity problem: a claim about another company attached to you, or your page skipped because the system was not sure it was about the thing asked. Semantic search covers the retrieval half; source citation covers the attribution half.

How do you strengthen your entity?

• One canonical name, used the same way everywhere: legal name, trading name and stylised name reconciled, not scattered.
• Organization markup on your home page with sameAs links to the official profiles that describe you.
• Matching third-party descriptions on the directories and profile pages that answers already cite about you. In the collision above, our own answers leaned on G2, Capterra and LinkedIn as well as our site, so those descriptions carry real weight.
• A clear category, stated in plain words near the top of your own pages, so the type of thing you are is never left to inference.

What goes wrong with knowledge graphs?

• Assuming markup equals inclusion. Acceptance requires corroboration from sources you do not own, with no queue and no timeline.
• Inconsistent naming across properties. Three names give a graph three weak candidates instead of one strong entity.
• Treating a knowledge panel as the graph. A knowledge panel is one rendering in one surface; its absence does not prove absence from the graph.
• Reading entity presence as a ranking signal. Being resolved decides whether a system knows what you are, not whether it cites you for a given question.

How does Attensira help with entity confusion?

It shows you the confusion in the answers themselves and works the sources behind it. Attensira keeps every answer verbatim, so a collision like the one above is visible as text, with the domains each answer cited. Its deciding sources view ranks the pages that shaped those answers, and the crew's off-site work prefills listings on directories such as G2 and Capterra from your confirmed brand profile, leaving any field it cannot answer blank rather than guessing. Every listing waits for your approval, and a placement counts only once the page shows you and an answer cites it.