Entity resolution so machines agree which company you are

You cannot buy a knowledge panel with sameAs links, and adding twelve of them changes nothing on its own. Machine certainty about a company comes from independent sources agreeing on the same facts, and four identifiers carry most of that weight: a registration number from the national register (KVK in the Netherlands, Companies House in the UK), a Wikidata QID with sourced statements, an LEI where the entity has one, and a legal name written identically everywhere it appears. When those four agree, the rest of the graph resolves. When they disagree, no amount of markup fixes it, and the panel that does appear belongs to somebody else. The question this settles: Which entity a machine believes you are, and whether that entity is the one your revenue depends on. That question gets expensive at three moments: a rebrand, an acquisition, and a market entry where a similarly named local company already exists. If any of those is on your roadmap for the next two quarters, this is worth doing before it happens rather than after. How it usually goes wrong: A trading name and a legal name diverge, and search engines split you into two entities with half the signal each. A subsidiary gets a stronger Wikidata item than the parent, so the parent's panel shows the subsidiary's founding date. sameAs points at profiles the company no longer controls, or at a LinkedIn page for a namesake in another country. A Wikidata editor sourced your employee count to a press release from 2019 and it has been quoted back at you ever since. Each of these is repairable. None of them is repaired by editing your own site, which is the reason most in-house attempts stall. What I actually do: I reconstruct the entity as the machines currently hold it, from Wikidata, the national registers, Crunchbase, Google's own returned panel and the disambiguation behaviour on your brand queries alongside your closest name collision. Then I write the target: one canonical legal name, one canonical description, the identifier set, the parent and subsidiary relations, and the founding and location facts with a citable source behind each one. Corrections go where they are actually read. Wikidata statements get sourced properly, because unsourced edits get reverted and reverted edits make you look worse than silence. Register records get corrected at the register. Your own Organization graph is rebuilt to reference the identifiers rather than assert facts on its own authority. Where I will tell you no: I do not create or edit Wikipedia articles for clients, and I will decline if that is the actual ask. Wikipedia notability is not something a consultant supplies, and paid editing that hides itself gets the article deleted and the brand mentioned in the deletion discussion. I do not remove true information because it is unflattering. I do not claim an entity your legal team has not confirmed you own, which matters more than people expect in group structures. And I will not chase a knowledge panel for a company with no independent coverage at all, because there is nothing for the graph to agree about yet. Not the right buyer: A company two months old with no press, no register history beyond incorporation and no third-party listings. A brand whose name is a common word, where the honest answer is a naming decision rather than an SEO engagement. Anyone who wants the panel changed by Friday, since Wikidata edits and register corrections propagate on their own schedule, typically four to ten weeks, and no supplier controls that. Why the top tier is monthly: Entity signals drift. Editors change statements, aggregators re-scrape stale data, an acquisition rewrites your parent relation, and a panel that was correct in March is wrong in June with nobody inside the company noticing until a journalist quotes it. The retainer exists because this is a standing position, not a project with an end date, and it is how the work is bought by companies that have been burned once. Reporting: Monthly, two pages: what changed in the graph, what I corrected, what is still contested and why, and the brand queries where a competitor or namesake still outranks you in disambiguation. Written for a communications director to read, not for an SEO dashboard.