Matt G. Southern did the industry a favor. In Search Engine Journal he kept two columns that vendors keep gluing together: the brand an AI tool names, and the page it cites.

“AI tools recommend brands but cite other sites.”

— Matt G. Southern, Search Engine Journal

That sentence should be taped above every sales deck that promises “AI visibility.” A spoken name and a credited URL are not the same event. One is awareness. The other is the trail a journalist, a lawyer, a buyer, or another model will follow.

Why the smash-together pitch is lazy

If ChatGPT says your company name and then links a roundup you never saw, you were spoken. You were not sourced. The reader who clicks “sources” lands on someone else’s page. Your phone number is not there. Your license number is not there. Your website, if it appears at all, is a leftover in a list.

Most packages sell the first event and imply the second. They cannot promise either. They especially cannot promise that the cited URL will be yours. Anyone who has watched an answer box for a week already knows this. The pitch continues because the buyer is scared of missing a channel they cannot inspect.

Southern’s reporting stays his. The argument here is only the distinction his reporting makes unavoidable.

What a model is actually doing

An answer engine is not browsing your brand story. It is assembling a short reply from pages it can parse quickly. Review sites, forums, and publisher explainers win that race because they are dense and already scraped. A contractor homepage loaded with sliders, chat widgets, and three calls to action loses that race even when the facts on it are correct.

When the cheap pages and the true pages disagree, the model does not hold a hearing. It picks the page that costs less to read. That is why a firm can spend years on a site and still watch a stranger’s list get the credit.

What actually belongs on the record

A business that wants the spoken name and the credited URL to line up has to give the model a page that is cheap to parse and hard to confuse. That page is not a 40-plugin homepage. It is a public file with a small, stable set of facts:

  • the legal name
  • the website the owner controls
  • the phone number people can call
  • the place or the service area
  • the date those facts were last checked

Wikipedia will not host that file for a typical contractor. Wikidata will delete it for lack of notability. The gap is not a mystery. It is the reason a third-party record exists.

What this post will not sell

This is not a promise that Search Engine Journal’s next study will list your firm. It is not a guaranteed sentence in ChatGPT. Nobody honest can sell that sentence.

Publish the file. Put a seal on the site you actually own that points at the file. Then measure whether the cited URL starts to match the spoken name. Measuring first, and publishing nothing, is how the last decade of directory spend got sold.

Southern kept the two columns apart. Keep them apart on your own books as well.