What ChatGPT says about your business — and where it gets it
More and more enquiries begin with someone asking a language model. It's worth knowing what it answers about you — and where that comes from.
Start with the simplest thing: ask it. Open ChatGPT and type your business name together with the town. Then do the same in Gemini. No settings and no expertise required, and within ten minutes you know what picture of you exists in these systems.
What you find will be one of three things, and each means something different.
Case one: it knows nothing about you
This is the most common, and not in itself a disaster — but it does mean that when somebody asks for a recommendation in your trade, you aren't in the answer. You aren't ranking badly; you are absent. That's a different problem from search optimisation, and it needs different handling.
The cause is usually that the business isn't documented in enough places. A model knows about a company when several independent sources say the same thing about it. One website on its own isn't enough for that.
Case two: it knows you, but says something outdated
An old address, a discontinued service, previous opening hours. This is the awkward case, because the answer sounds confident, and the person asking has no reason to doubt it.
The key thing to understand here is that the model cannot be corrected directly. There is no interface where you type in “that's not true”. What you can influence is the source it works from.
- trained knowledge
- live search
- business databases
- the answer someone receives
the lower two refresh — the top does not
The lower two boxes are where your work lies. Live search means the system looks at web pages at the moment of the question; if the correct information is there, it can reach the answer within days. Business databases mostly means the Google Business Profile — which you edit, and which many systems read directly.
The top box, trained knowledge, only changes with the model's next update. You have no influence over it, and no reason to wait for it.
Case three: it confuses you with someone else
The rarest and most irritating: the model merges you with a similarly named company, or with the same trade in another town. The cause is almost always the same — not enough distinguishing information. If a name isn't firmly tied to one town, one set of services and one contact, the system picks the most likely match.
The fix isn't to write more text, but to have the same information in several places, in agreement. Name, town, contact and scope of services identical everywhere — on the website, in the business profile, and in every trade listing where you appear.
The sequence that actually matters
One — the Google Business Profile. If you don't have one, create it; if you do, go through it. It is the one surface where you write the data about yourself, and which many systems read directly. Opening hours, address, services, category.
Two — put it in text on your own site. Opening hours, address and the service list often live only in an image, or in an element that opens on click. What isn't text isn't read by a machine. This is the most common silent failure.
Three — make them agree. If the website shows different opening hours from the business profile, the system can't tell which is true — and it will often use neither.
Four — give it something to quote. A model lifts specific passages. A page full of generalities (“quality service at fair prices”) gives it nothing. A page describing what you do exactly, for whom, and how, does.
These four steps are the foundation of visibility, but not the whole picture. On what makes content machine-readable, quotable and verifiable — structured data, clean hierarchy, self-contained passages — there's a separate piece: on AI visibility.
What isn't worth doing
Arguing with the model in the conversation isn't worth it. If you correct it, it remembers within that conversation, but the next person asking will get the same answer — the correction doesn't feed back into any knowledge.
And it isn't worth buying a service that promises to “get you registered” in a model. No such thing exists. What does exist is putting the sources in order — and that is verifiable work, not magic.
Questions on this topic
Can I get a language model to correct what it says about me?
Not directly. There is no interface where a business can rewrite what a model says about it. What can be influenced is the source: if the publicly available, authoritative places carry the correct information, and several sources agree, the model has less and less reason to repeat the old version.
Why do ChatGPT and Gemini answer the same question differently?
Because they draw on different material. Gemini leans on Google's search index and business profile data; ChatGPT on its own search and on the text seen during training. The same business can therefore be documented differently in the two systems — and where sources are scarce, the gap is wider.
What's the fastest step if I find wrong information?
Sorting out the Google Business Profile, because you edit it yourself and many systems draw on it directly. Your own site comes next: opening hours, address and the list of services should be there as text, not only in an image, and should match the profile.
How long until a correction shows up in the answers?
It isn't uniform. Answers that rely on live search can follow a change within days, whereas trained knowledge only shifts with the next model update. That's why it pays to fix first the sources these systems query live.