Entity Disambiguation and Knowledge Graph Groundwork

Before an assistant retrieves anything about you, it has to decide which you the question means. If your practice shares a name with a clinic in Perth, a US software company and a wound-up trust, that resolution happens on corroboration across sources that do not link to each other, not on what your homepage says about itself. The workable target is agreement: the same legal name, address, founding year, practitioner list and service description repeated in six or more independent places, with no contradictions left behind in the ones you forgot about. Where we start, which is subtraction: A twelve-year-old practice usually has three trading names, two former addresses still live on directories, a rebrand half the web missed, a practitioner who left in 2021 still listed on four sites, and an ABN registered to a name nobody uses. Each of those gives a resolver a reason to hedge, and hedging looks like your competitor being named instead of you. The first fortnight is spent finding them, and it is the part clients underestimate. What we build: Organization, LocalBusiness, MedicalClinic or RealEstateAgent markup with sameAs pointing at both the profiles you control and the registers you do not: ABN Lookup, ASIC, the AHPRA public register, state property licence registers, Google Business Profile, Bing Places, Apple Business Connect, the relevant industry body. Person entities for the people who do the work, an author page each, credentials, registration number, sameAs to the register entry. In health this moves citation more than anything on the service pages, because an assistant answering a clinical question is looking for who is accountable for the claim. Then the corrections: registry records amended with evidence attached, and directory cleanup aimed only at the sources that actually appear in answers. A Wikidata item where notability genuinely supports one. What we refuse: We do not create Wikidata items for businesses that fail notability. They get deleted, the deletion log is public and permanent, and a deleted item is a worse signal than no item. We do not write or edit Wikipedia articles about clients; the conflict-of-interest rules are explicit and the fallout lands on the client's name, not ours. We do not submit to two hundred directories, because directory sprawl is exactly how the contradictions we spend the first fortnight cleaning up arrived in the first place. On Knowledge Panels: You will ask. Entity work makes a panel possible; it does not make one happen. Panels appear months after the underlying agreement is in place, if they appear at all, and we will tell you where you currently sit without putting a date on it. How we measure it: Resolution prompts run before and after, of the form who is this business, is this business the same as that one, who founded it, who works there, what does it do. We record what each assistant asserts, and the same prompts run again at ninety days. Correct answers where there were confident wrong ones is the result; there is no score. Who this is not for: Businesses mid-rebrand or mid-acquisition. Settle the name first, because entity work on a name that changes in six months is paid for twice and the second pass is harder than the first, since the web now holds both versions. Also not for anyone whose main interest is a Knowledge Panel by a fixed date, and not for a business that would rather its true ownership or trading history stayed vague. This work makes an organisation legible. If legibility is not what you want, nothing here helps you.