I run LinkDaddy. I was trying to make the company easy for software to identify. Gemini told me the same thing other answer engines tell a lot of owners: create a Wikidata item, upload a photo to Wikimedia Commons, put the Q-id in your schema, and you will have an entity.
I had never done it. I followed the steps. It took hours. I got a Q-id. I put that Q-id on sites. About a week later the item was gone. The notice said promotional. Not notable.
I could not talk my way back in. That is why this registry exists.
What the models leave out
The advice is half a sentence. The missing half is the rule.
Wikipedia’s guideline for companies is blunt. A firm is presumed notable only if it has been the subject of significant coverage in multiple reliable secondary sources that are independent of the subject. Paid media does not count. Press releases do not count. Self-promotion does not count.
Wikidata is a knowledge base, not an encyclopedia article. Its bar is different and still not “you have a website.” The community’s notability page asks for a sitelink, serious public references, or a structural need. Draft language on the same project is clearer about the commercial case: creating an item about your own business is treated as a conflict. Marketing copy is not a source.
I had already paid for distribution. MarketersMedia placed releases on wires that show up next to Yahoo Finance, Business Insider, AP, and Barchart. Those pages look like coverage when you are the person who wrote the check. Volunteers who protect the projects treat them as what they are: you talking about you. That is the correct reading of their rules. It is also why the advice “get a Q-id” fails for a roofer in Tempe.
Notability is not a paperwork problem
Notability, in this context, means independent people already wrote about you because the story was the story. Not because you bought a slot. Kim Kardashian clears that bar. Mark Cuban clears it. A pest-control firm with a real truck and a real license usually does not.
There is no Bloomberg desk assigned to local trades. There is no Forbes interview waiting for the third HVAC company on the mile. That is not an insult. It is the newsroom. Encyclopedias were built to describe the world that already got written down. They were not built to verify every legal entity on earth.
The Wikimedia Foundation, which hosts Wikipedia and Wikidata, does not sit on those deletion decisions. Content is written and removed by volunteers. The Foundation is a nonprofit. In fiscal year 2024–25 it reported about $209 million in revenue, most of it donations, with a typical gift around $11. It does not sell ads on the articles. It also does not owe a local firm an item. That split — donated infrastructure, volunteer judgment — is why a model can send you to Wikidata and a person can take the item down a week later.
How many firms sit outside that bar
The World Bank group and the ILO have used a working range of roughly 400 to 510 million micro, small, and medium firms worldwide, most of them informal. The World Economic Forum cites about 400 million SMEs, around 90 percent of businesses, and about half of global GDP. Definitions move. The order of magnitude does not.
Almost none of those firms will ever collect the independent coverage Wikipedia wants. A slice of them will get a Wikidata item and keep it. A slice will get one and lose it. This newsroom does not have a public census of deletions. We do not need one to see the shape. If the rule is “already written about by someone who was not paid,” the default outcome for a small firm is no item, or an item that does not last.
Meanwhile the same models keep sending the next owner to the same form. The owner embeds the Q-id. The item dies. The schema now points at a ghost. That is not a knowledge graph. That is a broken link with extra steps.
The other half of the problem is the website
Even when the facts on the site are right, software often cannot lift them. A WordPress or Shopify homepage is a brochure: scripts, popups, plugins, three competing headlines. Crawlers have a budget. Answer engines have a time box. They take a paragraph from a list or a thread because the paragraph is already a paragraph.
That is why a hashed public file exists. The badge on the site is not a sticker. It is a door to a JSON document a model can fetch without walking the theme. The homepage can stay a brochure. The file does the document job. The longer version of that argument is the homepage is not the citation.
What I built instead
AI Verified publishes a public identity record for a business: legal name, how to reach it, the site the owner controls, a place or a service area, and the evidence behind those facts. Bronze is a selective snapshot. Gold is the same file after the owner proves the website is theirs and, where they have them, adds government and third-party identifiers.
The record is associated with real concepts already in the graph — organization, knowledge graph, data — not with a fake Wikipedia page invented for the company. We do not mint SME items. That is the rule the encyclopedias already taught me the hard way.
The file does not make you famous. It does not buy a sentence in ChatGPT. It does not replace a Secretary of State. It gives people and software one place to read who you say you are, with a date on it, without waiting for a newspaper that is never coming.
If you are the owner
Search the registry. If a snapshot already exists, read it. If the website is yours, attest it and put the seal on the site you control. Gold is $297 a year. We do not resell the call.
If you want the object explained without the story, start at what AI Verified is and what the public file is. If you want to see a finished Gold record, open LinkDaddy’s file.