Brand entity checker

An assistant answering "who is X" has to resolve X to something. The three surfaces below are the ones it can actually look up, and this tool reads each of them directly rather than inferring anything about them.

• Wikipedia. We ask the public MediaWiki API at en.wikipedia.org/w/api.php two separate questions: does a page exist at this exact title, and what does the search index return for this term. The first is a fact — and when the API refuses the string as a title at all, we report that as an unknown rather than as an absence. The second is a list of candidates, and we label it that way: a search hit is not proof the article is about your company.
• Wikidata. We search www.wikidata.org/w/api.php for items matching the term and report each one by id, label and description. There is no query that returns "the item for this domain", so we do not pretend there is. Picking the top hit and calling it yours would put a Q-number on this page that you would then go and cite.
• Your homepage markup. We fetch the homepage, extract every <script type="application/ld+json"> block, parse them, and look for a node whose @type is an Organization or an Organization subtype. If we find one we report the types it declares, the name it declares, and every entry in its sameAs array. If we do not, we say the markup declares none — we never fill in what a page "probably has".

Then we request each sameAs URL. That last step is the point of the tool: sameAs is a set of claims, and a claim nobody checked is how a homepage ends up pointing at a company profile that was deleted two years ago.

We hold no Crunchbase or LinkedIn API access, so there is no Crunchbase lookup and no LinkedIn lookup here, and we are not going to imply otherwise. What there is: if your own markup lists a profile on one of those hosts, we request that exact URL and report the status it returned — labelled presence-only, because hosts that wall automated requests can refuse a page that exists.

The labelling matters more than the check. Three outcomes are kept deliberately apart:

• Answered. The URL returned a success status. That proves the URL answers. It does not prove the profile is yours, current, or complete, and the row says so.
• Not there. The host answered 404 or 410. This is a real absence and the only outcome we treat as a failure of the sameAs entry itself.
• Refused us. The host answered with a wall — a 403, a 429, or a non-standard status of its own. That is a fact about the host and not about the page. The row is amber and reads unchecked, not disproved, because colouring it red is how a reader ends up deleting a profile that exists.

Everything else in this toolbox is something you can fix this afternoon by changing your own site. Robots rules, structured data, heading structure, an llms.txt — all of it is yours to edit. A Wikipedia article is not. Neither is a Wikidata item with real references behind it. Both are written by other people, about you, under notability rules you do not control, and creating one about your own company is against Wikipedia's conflict-of-interest guidance.

We would rather say that plainly than sell you a checklist item you cannot complete. If this tool reports no Wikipedia entity, the honest reading is not "go and make one". It is: the surfaces you do control now carry more weight, because they are the only place an assistant can learn who you are. That means an Organization node whose sameAs array is accurate and resolves, consistent naming across every profile you own, and coverage on third-party sites that write about companies in your category.

Do not invent a sameAs. Adding https://en.wikipedia.org/wiki/Your_Company to your markup when no such article exists does not create an entity — it creates a broken claim in your structured data, which this tool reports as not there. A page has to be earned. That is the rule we hold ourselves to on our own site, and it is why our own markup has no Wikipedia sameAs in it.

• No Organization node in the homepage markup at all. This is the first thing to fix and the cheapest. One JSON-LD block with @type: Organization, your legal name, your canonical URL and a sameAs array of profiles that genuinely exist. Check the block renders and parses with the structured data validator.
• An Organization node with an empty sameAs. You have told an assistant your name and nothing that corroborates it. Add only the profiles you actually maintain and can point at today.
• A sameAs entry that returns 404. Remove it or repoint it. A dead claim is worse than a missing one: it is the one piece of your markup that can be demonstrated wrong.
• A sameAs entry that redirects somewhere else. The row names the final URL. Put that URL in your markup instead, so the claim and the destination are the same string.
• The name in your JSON-LD is not the name anyone searches. The result header shows exactly which string was searched and where it came from. If that string is wrong, the whole report is answering the wrong question — re-run it with the optional name field filled in, then fix the name in your markup.
• Several plausible Wikipedia or Wikidata matches. Name collisions are the ordinary case, not the exception. If an entity with your name exists and is somebody else, the risk is not absence but misattribution, and the fix is unambiguous naming everywhere you control.

Retrieval and citation both work on entities, not strings. When an assistant is asked a question in your category, it is matching a concept to sources it can attribute. A brand that resolves to a described entity — with a consistent name, a canonical URL and corroborating profiles — is a thing it can name with confidence. A brand that resolves to nothing is a string it saw once on a page.

That is also why this check is worth re-running rather than doing once. Profiles get renamed, company pages get merged after an acquisition, a social handle changes and every sameAs pointing at the old one quietly starts 404-ing. None of that shows up in your analytics, and none of it breaks a page. It just makes your markup slightly less true each time, until the claims and reality have drifted apart.