Where do AI assistants get information about local businesses?
Short answer
From the text the model was trained on, and from what the assistant looks up on the web when someone asks. For local businesses, the sources that show up include the company's own website, map and directory listings, review sites, local articles and forum discussions. Your website is one source among several.
AI assistants get information about local businesses from the text they were trained on, and from what they look up on the web at the moment someone asks. Which of the two a given answer relied on isn’t always visible, and the companies behind the assistants don’t publish every detail. The general mechanism is known, and so are the kinds of source that appear when an assistant lists them.
Training data and live lookup
They work differently, and the difference explains most odd answers. A language model is trained on a very large body of text collected up to some date. Whatever it holds from training is as old as that collection, so a company that has opened, moved or renamed itself since then may be missing or described as it used to be. That gap is behind the question of whether AI assistants know a new business exists.
Many assistants can also search the web when a question needs current information, read a few of the pages they find, and write the answer from those. Questions about who to hire often trigger a lookup, though not every time. You can usually tell, because the answer comes with sources.
The sources that show up for local questions
The list below shows the kinds of page that commonly appear under local answers. It’s observation, not documentation, and the mix varies by assistant and by question.
- The business’s own website: service pages, and the about and contact pages.
- Map listings such as a Google Business Profile, where the assistant is connected to map data.
- Directories and review sites.
- Local articles that list or compare companies.
- News and community websites.
- Forum and neighborhood discussions where people recommend someone.
Why your own website is only part of it
An answer to “who should I call” is assembled from several pages, and most of them aren’t yours. In practice, a business described the same way in many places tends to be described accurately. A business with conflicting details tends to get an answer that is wrong or vague.
Take a hypothetical roofing company in Escondido. Its website says it serves North County. An old directory entry still shows a previous address in Vista and a phone number that was disconnected years ago. Either version could end up in an answer, and the owner would never know unless they asked. Cleaning that up is part of reviews and citations for AI.
What you control and what you don’t
| Source | Your control over it |
|---|---|
| Your website | Full |
| Google Business Profile | High, though Google and the public can suggest edits |
| Directories | Partial. You can usually claim and correct a listing |
| Reviews | You can ask for them and reply to them. You can’t write them |
| Articles and forums | Little. Mentions are earned |
| Training data | None directly. It reflects what the web said when it was collected |
How to see it for your own business
Ask an assistant what it knows about your company by name, then read the sources it lists. Do this in more than one assistant, and write down any fact that is wrong, with the page it appears to have come from.
One caution. If you ask an assistant where it learned something, its explanation isn’t dependable, because it’s generating a plausible reply and not reading from a log. Trust the listed sources over the explanation. Entity and schema work makes the facts on your own site easy for a machine to confirm.