ChatGPT doesn’t know your business because it doesn’t know any business. AI assistants don’t keep a directory of companies — they build each answer from whatever they can read in the moment: your Google Business Profile, directories, review sites, your website. If those sources are thin, stale, or contradict each other, the assistant has nothing solid to work with, and its answer skips you or gets you wrong. The fix isn’t inside ChatGPT. It’s in the sources it reads.
Here’s how that actually works — and what you can do about it.
Assistants don’t remember. They look things up.
An AI model is trained once, on an enormous snapshot of text, and then it answers from two places: what stuck from that training, and what it can retrieve right now.
What stuck from training is a kind of long-term memory, and for a local business it’s close to empty. Training snapshots skew heavily toward the most-written-about things on the internet, and a plumbing company in Boise or a dental clinic in Des Moines simply isn’t written about enough to leave a durable imprint. You can see this yourself: ask an assistant about your business with web browsing turned off, and it will usually tell you it doesn’t have information — or, worse, guess. That’s not the assistant failing. That’s what its memory honestly contains.
What it can retrieve right now is different. When web search is on — and on the assistants your customers actually use, it usually is — the assistant runs searches on your behalf, reads what comes back, and builds its answer from that. The industry word for this is grounding: the answer is grounded in sources fetched seconds before you read it.
So when a customer asks “who does ceramic tint near me,” the assistant isn’t consulting what it knows. It’s doing a fast, automated version of what the customer would have done: searching, reading a handful of results, and summarizing. The businesses in those results can be in the answer. The businesses that aren’t, can’t.
What an assistant actually reads
You don’t have to take this on faith — the assistants show their work. Ask one a question about businesses in your area and look at the citations attached to the answer. You’ll see the same kinds of sources over and over:
- Google Business Profiles and the maps results built on them
- Review platforms — Google reviews, Yelp, and whatever is strong in your industry
- Directories and aggregators — general ones, and vertical ones for your trade
- Local press and “best of” lists
- Business websites — sometimes yours, often a competitor’s
That citation trail is the whole story. An assistant’s answer about your business is only as good as the best sources it found about your business — and if it found none, the answer is built entirely from other people’s.
The ways a business ends up missing
When we trace why an assistant skipped or misdescribed a business, the cause lands in a handful of gaps. The common ones:
- The assistant isn’t sure which business you are. Duplicate listings, an old address still live somewhere, a name too generic to pin down. A machine that can’t tell your two listings are one business — or your business from a similarly named one — can’t recommend you with confidence. (We’ve watched a single misspelled name in one review conjure a practitioner who doesn’t exist — that story is in our case studies.)
- You’re absent from the sources it reads. The assistant cites a directory for your kind of question, and you’re not in it. No amount of quality on your own site fixes an answer built from a list you’re missing from.
- Your facts are stale somewhere. The assistant read a source with your old hours or a dead phone number, and repeated it. It has no way to know that source is wrong — it can only compare sources against each other.
- Nobody has answered the question. Customers ask assistants specific things — “does anyone do mobile tint on weekends?” — and the assistant searches for an answer. If no page anywhere answers it for your business, you can’t be the answer.
- Your site says it, but not in a form machines parse. Your hours might be in an image, a PDF, or a line of styled text. Structured markup — schema.org — is how a website states its facts in a form machines read reliably.
We go deeper on all five on our services page, but the pattern is the point: every one of these is a property of the sources, not of ChatGPT. Which is good news, because sources can be fixed.
See it for yourself first
Before fixing anything, find out where you actually stand. You can do a real version of this check yourself, free, in about twenty minutes — we wrote up the exact questions to ask and what to look for. Ask the assistants what a customer would ask, read the citations, and note what’s missing or wrong.
What to do about it
The work is unglamorous and specific:
- Consolidate your identity. One Google Business Profile, one exact business name, one address and phone number, everywhere. Google’s own documentation covers managing your Business Profile; duplicate listings are worth resolving before anything else.
- Get onto the sources assistants cite. Not every directory — the ones that actually showed up in your citation trails.
- Correct the stale facts at their source. Old hours on a directory you forgot existed will keep resurfacing in answers until the directory itself is fixed.
- Publish answers to the questions customers actually ask, on pages you control.
- Add structured markup so your site states its facts machine-readably — Google documents the supported formats in its structured data guidelines.
None of this makes an assistant promise to mention you — nothing does, and anyone who guarantees a mention is guessing. What it does is remove every reason the assistant currently has to skip you or get you wrong.
If you’d rather have it measured and fixed with a paper trail — dated transcripts before and after, every finding cited to its source — that’s the service we run. Either way, start by asking the assistants about your business today. What they say may surprise you; where they got it from will explain it.