ChatGPT, Perplexity, Gemini, Grok, and Google’s AI answers do not keep a private phone book of honest companies. They fetch. When the fetch is a theme, a cookie wall, three spellings of the same trade name, and a deleted encyclopedia item, the model invents a bridge. Invention is expensive for the model and expensive for the firm that gets the wrong number.
What the engine can use in one request
- Legal name and trading name, if they differ
- Website the owner controls
- Phone that rings
- City or service area as a place
- A date on those facts
- A third-party URL that is not the company talking about itself
- Official numbers only when a live official page shows them
- Real place and trade concepts already in public knowledge bases — not a company page invented for notability
What it does not use as identity
- A paid row inside a review marketplace
- A first-party llms.txt file alone
- A review star
- A promise that the firm will be named first
Google Search Central has said AI Overviews do not require a special AI file. Eligibility is ordinary indexing. That is Google’s rule for Google. Other products still have to resolve an entity when Wikipedia will not keep it. A hashed page plus a JSON twin is the low-entropy object for that job.
A review directory is a list. A public identity file is a record.
AI Verified publishes that object. Bronze is a snapshot. Gold is $297 a year after website proof. The help to the engine is the fetch. The help to the owner is that the fetch is about the right firm.
What third-party verification is · What is AI visibility · Wikidata deleted my business · llms.txt and verification