Key Takeaways
- Stop picturing one database and start picturing four sources. Assistants assemble local answers from structured listings, your own site, third-party writing, and a live search. Different sources, different update speeds, different levers.
- Fix the sources you own before chasing the ones you do not. Your listing and your website are the two you can change this afternoon, and they are the two most often left half-finished.
- Independent writing is the source you cannot fake. A mention on a site you do not own confirms you exist. It is slower to earn and worth more per line than anything you publish about yourself.
- Expect a lag between fixing a source and seeing the answer change. Live retrieval can reflect a change quickly. Anything drawn from training data reflects a picture assembled long before you asked, so re-check quarterly rather than daily.
- Wrong information usually means a stale source, not a broken assistant. When an answer gets your hours, location, or offer wrong, the fix is finding which source still says the old thing and correcting it there.
There is a picture most business owners carry around, and it is wrong. It imagines a big list of businesses somewhere inside the assistant, and a process for getting your name onto it.
No such list exists. There is no submission form, no inclusion fee, and nobody to email.
What exists instead is four kinds of public source, read at different times and in different ways. Once you can name them, the whole problem stops being mysterious and starts being a to-do list.
Where does ChatGPT get information about local businesses?
From four places: structured listings such as map profiles and directories, your own website, third-party writing about you, and a live search run at the moment someone asks.
It has no private database of businesses. OpenAI's own description of ChatGPT is a model trained on large amounts of public text, and current products add the ability to search while answering. Both halves of that sentence matter, because they behave very differently.
So when an assistant says something about your business, it read it somewhere public. Your job is to work out where, and to make that somewhere correct.
The four source families
Two you control outright, one you influence, and one you cannot touch at all. Knowing which is which stops you spending a month on the wrong one.
- Map and business profiles
- Industry directories
- Booking platform pages
- Service and pricing pages
- About and credentials pages
- Anything marked up as structured data
- Local news and roundups
- Association and club pages
- Forum threads and community posts
- A search run at the moment you ask
- Pages fetched and read on the spot
- Whatever ranks that day
In text, so nothing lives only in the figure. Structured listings, which you control, include map and business profiles, industry directories, and booking platform pages. Your own website, which you control, includes service and pricing pages, about and credentials pages, and anything marked up as structured data. Third-party writing, which you influence, includes local news and roundups, association and club pages, and forum threads. Live retrieval, which you cannot touch, is a search run at the moment someone asks, pages fetched and read on the spot, and whatever ranks that day.
The structured listings do disproportionate work because they are built to be parsed rather than read. Category, area served, and opening hours are not marketing copy. They are the machine-readable answer to "which business is this", written in a shared vocabulary, and the local boot camp post walks through filling those fields in for a real business.
Training data and live retrieval behave differently
One reflects a picture assembled in the past. The other reflects the web this morning. Most confusing AI answers come from mixing the two.
Training data is fixed at a point in time. If you rebranded last spring and an assistant still uses the old name, that is the old picture surfacing, and no amount of updating your own site changes what was already learned.
Live retrieval is the opposite. It reads current pages, so a corrected listing can show up quickly, provided the assistant can reach the page at all. That last condition catches people out: OpenAI documents its crawlers and a site that blocks them removes itself from the half of the system that updates fastest. The same failure mode is covered in why ChatGPT cannot find your business.
The practical consequence: fix the source, then wait, then re-check. An answer that has not moved in a week has not necessarily failed to move.
The sources compared
How fast each one updates, how much control you have, and what it contributes to the answer.
| Source family | Examples | Your control | How fast it updates | What it contributes |
|---|---|---|---|---|
| Structured listings | Map profiles, directories, booking platforms | Full, if you claim them | Days to weeks | Category, area served, hours, contact facts |
| Your own website | Service pages, pricing, about, structured data | Full | Immediately, once crawled | The sentences an assistant can quote about you |
| Third-party writing | Local news, roundups, association pages, forums | Influence only | Weeks to months | Independent confirmation that you exist |
| Live retrieval | A search run while the answer is being written | None directly | Same day | Freshness, and whichever pages rank at that moment |
Read the control column downward and the strategy writes itself. Two rows are yours today, one is a request you make to other people, and one is a consequence of the first three rather than a thing you act on.
What to do with this
Find which source is wrong before you change anything. Most bad AI answers trace to one stale page, and guessing wastes the month.
Ask the assistant your question, then ask it where that came from. Many will name or cite sources, and Perplexity's help center documents a product built around showing them. The cited page is your target. Correcting it beats publishing something new that has to compete with it.
Then work outward in control order: your listing, your site, then the outside mentions. Reviews sit inside the third family and are worth their own effort, which the reviews post covers in detail.
One caution about sequence. There is no point earning outside mentions while your own listing still states the wrong category, because the mentions will be read against a record that contradicts them. First make the sources you own agree with each other. Then go and get quoted.
If you want to see which sources are shaping your answers without doing the digging: the free check is twenty-five questions, $0, and a plain-English readout within the hour. It runs as soon as you ask for it.
Questions we hear the most
Where does ChatGPT get information about local businesses?
From four places: structured listings such as map profiles and directories, your own website, third-party writing about you, and a live search run at the moment someone asks. It has no single private database of businesses.
Does ChatGPT have its own directory of businesses?
No. It has a model trained on large amounts of public text plus, in most current products, the ability to search the web while answering. Local facts come from those public sources rather than a proprietary listing of businesses.
Why does AI have out-of-date information about my business?
Because at least one source it reads still says the old thing. Training data reflects a moment in the past, and stale directory entries or old articles can outlive the change you made on your own site.
Which source matters most for a local business?
The structured listing, because it states your category, area, and hours in a machine-readable form and it feeds other sources. Your own website matters nearly as much, since it is the only place you write the full sentences.
Can I submit my business directly to ChatGPT?
No, and anyone selling that is selling nothing. There is no submission form or inclusion fee. The only route in is making the public sources it already reads say the right thing about you.
How do I find out which sources AI is using for my business?
Ask it, then ask it where the information came from. Many assistants will cite or name their sources, and the ones that do will show you exactly which stale page is shaping the answer.
What does the AnswerHalo free check include?
It costs $0. We ask ChatGPT 25 real buyer questions from your niche, then email a plain-English readout within the hour: how many answers named you, who was named instead, and the first fix. It runs as soon as you ask for it.
Does the AnswerHalo audit tell me which sources to fix?
Yes. The $500 audit names the pages being quoted instead of yours: the domains the assistants cited in the answers to your own questions, so you can see what they are actually reading.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →