GEO Expert
Let’s talk
← The register

How an AI assistant decides which business to name

What happens in the few seconds between the question and the answer, why directories win so often, and the six ordinary reasons a good business never appears.

Sai Siddesh11 min

People imagine the assistant knows things and recalls them. Something closer to the truth is that it reads the question, goes and finds material, checks the material against itself, and then writes. Each of those four steps is a place where your business is either included or dropped, and knowing which step is dropping you tells you what to fix.

Step one: it works out what is really being asked

A person types is it worth getting a lawyer for a fender bender. The assistant turns that into several narrower questions it can actually look things up for: what counts as a minor collision, when do claims exceed insurance limits, how are fees charged, who handles this locally. Your page is never competing against the question as typed. It is competing against those narrower ones.

This is why pages organised around one clear question outperform pages organised around a service category. A page titled Car accidents matches none of the narrower questions well. A page titled Do I need a lawyer for a minor car accident? matches one of them exactly.

Step two: it goes looking

The assistant issues its own searches, fetches a handful of pages, and reads them. It is impatient. It takes what loads quickly, what parses cleanly, and what has visible structure. A page that renders its main text only after a script runs may return nothing at all.

It also tends to pull from pages it has used successfully before for similar questions. Early absence compounds, which is an argument for starting rather than waiting.

Step three: it checks what it found

Before repeating a specific claim about a named business, the assistant looks for agreement. Does the address on the website match the address on the licensing register. Does the person described as the owner appear as the owner anywhere else. Does the service on the website appear on the licence or the professional profile.

Where the sources agree, the assistant will state the fact plainly. Where they disagree, it does one of three things: hedges, attributes the claim to your site rather than asserting it, or drops the business and names something else. Most invisible businesses are being dropped at exactly this step, and from the outside it looks identical to never having been found.

Most businesses that never get named were found. They were found, checked, and quietly set aside.

Step four: it writes

Now it composes an answer from the passages that survived. Short, self-contained passages survive. Sentences that depend on the paragraph above them to make sense do not, because the assistant is lifting them out of context. A sentence reading in those cases the deadline is two years is useless on its own. In California, a personal injury claim must normally be filed within two years of the injury travels.

Why directories win so often

Open an assistant, ask who to call for almost any service in almost any city, and you will often get a directory rather than a business. This is not favouritism. A directory is an easy thing to be confident about. It is one clearly identified organisation, its structure is predictable, its listings corroborate each other, and recommending it carries almost no risk of being wrong about a specific business.

Naming your business instead is a judgement the assistant has to be able to defend. Everything in this article is about making that judgement easy.

Six ordinary reasons a good business never appears

  • The name does not match across records. A comma, an abbreviation, or an old trading name still on one listing is enough to split you into two entities.
  • The address is stale somewhere. One old suite number on one directory will do it.
  • The site needs a browser to read. Main content injected by script after load is content the fetcher never sees.
  • There are no quotable sentences. Every page opens with positioning and reaches the answer on screen three.
  • No independent source says anything about you. Your own site is the only place the facts exist.
  • The topic is not written about at all. You handle it daily and have no page about it, so nothing can be retrieved.
How to tell which one is yoursAsk an assistant to describe your business and to say where it got the information. If it describes you wrongly, the problem is identity. If it describes you correctly but never recommends you, the problem is quotable content. If it has never heard of you, the problem is reach and corroboration.

What this means in practice

Fix the steps in order. Identity first, because a business the assistant cannot resolve will not be named however good its writing is. Reach second, because unread pages are not in the running. Quotable content third, because that is what converts being eligible into being chosen. Doing them in the other order is the most common way a budget gets spent with nothing to show for it.

What this means for you

Start with the free sweep.

Working out which step is dropping you is the first thing we do and the cheapest thing to get wrong. We will tell you where you are failing before you spend anything on fixing it.

Cite this entry

Siddesh, S., How an AI assistant decides which business to name, Headnote 045 (GEO Expert, 5 Oct 2026).