What this article covers: The fix: what to do when AI states wrong facts about your business.
Until recently the big question was whether AI search could find you at all. Increasingly there is a second, more awkward one: it found you, and what it says is wrong. The wrong price, a service you no longer offer, a town you never worked in. Unlike Google, where the customer sees your page and judges for themselves, here they get a statement — with no context and often with no link.
Three reasons this happens
An inaccurate statement is almost never malicious. It usually comes down to one of three entirely technical causes.
- The model is working from an old snapshot of the web. Models are trained on content captured during a particular period. If you changed your prices last year, a model without live access will still quote the old ones.
- Inference from prose. If your page does not clearly state what you offer and where, the system fills the gap by inferring from context. One town mentioned in a news item can become "operates in that town".
- Confusion with a similar business. If your name resembles another and there is no machine-readable data separating you, the details of both get merged.
Tip: All three causes share one root: there was no clear, unambiguous source on your site for the system to draw on. That part you can do something about.
How to check what it says about you
The test takes a few minutes and you can run it yourself. Ask the questions a customer would ask, not the ones you would write:
- "What does [your company name] offer?"
- "How much does [your service] cost at [company name]?"
- "Where is [company name] based and how do I contact them?"
- "Who in [your town] offers [your service]?" — without mentioning your name.
Tip: Repeat the questions in at least two assistants. The answers differ more than you would expect, and it is the differences that reveal where your data is weak.
What you can fix on your own site
You cannot edit the models. You can make sure they have something reliable to copy from — and that works better than it sounds.
- State the facts in machine-readable form. schema.org markup says unambiguously who you are, where you operate, what you offer and at what price. It is not text a reader sees; it is data that does not have to be interpreted.
- Have one place that holds the truth. If a price appears on four subpages and one of them is out of date, that is the one that will get read. A single pricing page that the others link to removes half the problem.
- Answer the questions customers actually ask, in their words. A question-and-answer section is the format an answer is easiest to extract from.
- Keep it consistent off-site too. Your name, address and contact details should be identical everywhere — directories, Google profile, social media. Discrepancies are a common source of mix-ups.
- Remove what is out of date. An old page about a service you no longer offer is just as valid a source to the system as the new one.
Why the fix does not take effect immediately
Once you correct the data, the answers do not change the same day, and that is normal. Two things have to happen: a crawler has to visit the page again, and the system has to process the new version and place it in its index.
For live search that is relatively quick, a matter of weeks. For what a model absorbed during training it can take far longer, or not change until the next version of the model.
So there is no point measuring the effect after a few days. We ourselves wait around forty days after making changes before repeating the test — any sooner and we would mostly be measuring chance.
Tip: If an inaccurate detail is genuinely damaging, do not rely on search engines alone. Correct it on your site, in directories and on your Google profile at the same time — several consistent sources work faster than one.
What you cannot do
Here it pays to be honest, because providers like to promise more than is possible.
You cannot demand a correction. There is no form where you report that an answer is wrong and it is right the next day. Some providers have routes for reporting content, but these are not processes with guaranteed timelines.
You cannot guarantee an assistant will recommend you. Who appears in an answer is decided by a system you do not control and which keeps changing.
You cannot prevent it from ever getting something wrong. You can substantially reduce the odds by making the correct information easier to reach than the incorrect.
Where to start if this affects you
Run the test from the third section and write down what is wrong. Then check whether your site has structured data at all, and whether the disputed detail is even clearly stated anywhere — often it turns out it is not.
If you would like help with that, our quick check of AI search accessibility is free: send us your address and within one working day we will tell you what we found. If everything is in order, we say so and that is the end of it.
If you are also interested in the technical side — why some assistants cannot see you at all — we wrote about that in Why ChatGPT doesn't know you, even though you rank first on Google.
Frequently asked questions
Can I tell the company behind an AI that a detail is wrong?
Some providers offer a way to report problematic content, but these are not processes with a guaranteed timeline or outcome. The more reliable route is to make sure the correct information is stated clearly and machine-readably on your site and in directories.
Why do different assistants say different things about me?
Because they differ in when they were trained and whether they search the web while answering. One may be working from data a month old, another from a year ago. That is exactly why it is worth checking in several assistants.
Does structured data really make a difference?
It guarantees nothing, but it is the least ambiguous way to state facts about your business without the system having to infer them from prose. For prices, opening hours and service areas the difference is greatest.
How often should I check this?
A few times a year, and whenever you change prices, services or contact details. It is precisely around changes that gaps appear between what is true and what the systems still repeat.