AI says something wrong about my company. What now?
An old price, a service you have stopped offering or a service area that is not right. This is how to find where an AI system gets its mistake and how to put it right.
- An AI system rarely makes something up about a company out of nothing. Almost always there is a source containing the mistake.
- You cannot edit the answer. You can edit the source, and then the answer follows.
- What comes from live search recovers within weeks. What sits in the model itself takes longer.
Out of curiosity you ask a question about your own company, and the answer is wrong. ChatGPT says you are also open on Sundays. Gemini names a service you stopped offering two years ago. Claude thinks you are based in Zwolle, although you have been working in Kampen for three years.
"AI says something wrong about my company" is one of the most annoying discoveries, because you feel you cannot do anything about it. There is no button for reporting a mistake that is solved by tomorrow.
Still, you can do something about it. This article describes how to find the cause and what to tackle in which order.
Why it matters
A mistake about your company is not just sloppy. It costs you something.
You get the wrong enquiries. People call about a service you do not offer, or from a town you do not go to.
You miss the right ones. If a system thinks you only serve private customers, it does not mention you when a company asks for help.
You disappoint people. Someone counts on the price they read and hears a different amount on the phone. That conversation starts badly.
It damages your credibility. Whoever reads something on your website that differs from what the assistant said no longer knows what to believe.
First: is it really wrong, or was it chance?
AI answers are not the same every time. Before you get to work, you want to know whether the mistake is a pattern.
- Ask the same question three times, each time in a new conversation.
- Ask it in three systems. ChatGPT, Gemini and Claude.
- Note what comes back. If every system says the same wrong thing, there is a source they all read. If one system says it once, it was probably a slip.
A mistake that comes back in several systems is good news. It means there is an identifiable source. And you can edit a source.
Where mistakes come from
An AI system rarely invents something about a specific company out of nothing. In by far the most cases the mistake is written down somewhere. These are the five places where we find it most often.
1. Your own website
It is the source people check last, and where the mistake most often sits.
A page about a service you no longer offer, still online because nobody took it down. An old news item about a promotion with a price. A team page with someone who left long ago. A footer with the old address on half your pages.
You never visit those pages any more. A system does.
2. Your Google Business Profile
Opening hours that were never adjusted after a change. A category that no longer fits. A service area set too wide.
There is something else many business owners do not know. Visitors and Google itself can suggest changes to your profile, and sometimes those are applied. So it may be that something is in your profile that you never put there yourself.
3. Old business directories
Websites that collected company details years ago and never updated them. Your old address, your old phone number, a description from when you still did something else.
You never signed up there and you do not know they exist. To a system they are sources like any other.
4. A company with the same name
There is a plumber Johnson in Deventer and a plumber Johnson in The Hague. A system mixes them up and attributes the services or the reviews of the other to you.
You recognise this by mistakes that make no sense at all. A service you never offered, in a town you have never been to.
5. The knowledge of the model
An AI model was trained at a certain moment. What was on the internet about you then can be in the model. If you have since moved or changed course, the model may still hold the old picture.
This is the hardest kind. More on that further down.
How to find the source
You can have the system itself help.
Ask where the answer comes from. "Which sources do you base that on?" Most systems that search the internet show their sources or can name them. Open them and look for the mistake.
Search for the mistake literally. If a system says you are open on Sundays, search in a search engine for your company name with "Sunday" added. If it says you are in Zwolle, search for your name with "Zwolle". You will find the page where it stands.
Search your own website. Search in a search engine for the word site, a colon, your domain name, and after that the word that is wrong. You see every page of your own website on which that word appears. This nearly always produces a surprise.
Look in your business profile. Go through every field. Including the fields you never use.
If none of these steps produces a source, and the system indicates that it did not search, the mistake is probably in the knowledge of the model.
The repair, in order
Step 1: make your own website the truth
Start here, always. Your website is the source you have full control over, and the source a system leans on most heavily when it is about you.
- Remove or rewrite what is no longer right. Old services, old prices, old team members.
- Write the correct information down literally. Do not only take the mistake away. If a system thinks you are open on Sundays, put on your website: "We are closed on Sundays."
- Say what has changed. "Since 2024 we have been based in Kampen. Before that we were in Zwolle." A sentence like that helps a system reconcile old and new information.
- Check the fixed parts. Footer, contact page, the page about your company.
That last point, explaining what has changed, is often forgotten and works well. A system that finds two different addresses does not know which is right. A sentence explaining that you moved resolves it.
Step 2: update your business profile
Go through everything and adjust what is not right. Then check that it matches your website. Look again in a month to see whether changes from others have been applied.
Step 3: tackle the other sources
For each directory or website with a mistake:
- See whether you can claim the profile and edit it yourself. With many directories that is free.
- If that is not possible, send a short message asking for the details to be corrected or removed.
- If nobody replies, let it rest. One outdated directory against a clear website, a correct profile and recent reviews loses.
Start with the sources a system really quoted. You do not have to sweep the whole internet clean.
Step 4: set yourself apart from your namesake
If there is a company with the same name, make it easy for a system to tell you apart.
- Name your town often and clearly, in your page titles too.
- Use exactly the same spelling of your name everywhere.
- Put on your company page what you do and do not do.
- Consider a sentence like: "We are not connected to the company of the same name in The Hague."
Step 5: give it time and check
After your changes it takes a while. A system that searches live reads your edited page as soon as the search service has visited it again. For an active website that is a matter of days to a few weeks.
Test again after two weeks and after six weeks, with the same questions. More about those terms in how long until AI mentions your company.
When the mistake is in the model itself
Sometimes a system gives a wrong answer without searching, from what it already knew. That is more stubborn.
The knowledge of a model is only updated when a new version is trained. You have no influence over that, and it can take months.
What you can do:
Make sure the correct information is everywhere. When a system does search, it finds the right details and usually uses those instead of its old knowledge. For questions about local companies and current matters the systems search often.
Make sure the next version learns it correctly. What is online now, clearly and in several places, ends up in the next training. So you are working on the future.
Use the feedback button. Most assistants have an option with each answer to give feedback. It does not solve your problem straight away, and we do not know how much weight it gets. It takes ten seconds.
Can I ask the AI companies to change it?
For facts about a company, most providers have no desk where you submit a correction that is guaranteed to be applied. There are forms for requests about personal data, based on privacy law. Those are meant for information about persons and not for a company's opening hours.
If a mistake is about you as a person, or it is harmful and untrue in a way that goes beyond a wrong address, it is wise to get legal advice. That is outside what we can assess.
For ordinary mistakes the route in this article is the fastest: edit the source.
Preventing mistakes
Whoever has had to put a mistake right once does not want to do it again. Three habits help.
Keep one list of your fixed details. Name, address, phone, opening hours, service area, services. And a list of all the places where they appear. If something changes, you work through the list.
Tidy up when you stop doing something. A service you stop offering comes off your website, or you add that you no longer provide it. Do not leave pages drifting around.
Test every quarter. Ask three systems what your company does, where it is and when it is open. Five minutes of work. The full method is in is my company mentioned in AI.
An example
A made up example, built from what we often see. An installation company stopped fitting air conditioning three years ago and now focuses entirely on heat pumps. Even so, two AI systems mention the company for questions about air conditioning.
The search turns up three sources. An "Air conditioning" page that is still on the website but no longer in the menu. A category in the business profile. And a listing in a dealer overview from an air conditioning manufacturer.
The repair: the page is replaced by a short text saying the company no longer fits air conditioning and focuses on heat pumps, with a referral to a colleague. The category is changed. The manufacturer receives a request to remove the listing.
Six weeks later the systems mention the company for heat pumps and no longer for air conditioning. The calls about air conditioning stop.
Most of the work was in the finding. The editing took an hour.
What you can also learn from it
A mistake in an AI answer is annoying, and it is also information. It shows what your company looks like from the outside, to someone who can only read what is there.
Often it turns out that more is unclear than that one mistake. The service area is nowhere. The services are vaguely described. There is no answer to the questions customers ask. The mistake was the visible part of a website that tells too little.
Whoever fixes the mistake and then carries on writing down clearly what they do comes out better than they started. You can read the causes of invisibility in why am I not in AI recommendations.
Would you like to know what AI says about you now?
The free baseline check shows how ChatGPT, Gemini, Claude and Perplexity describe your company, side by side. You see at a glance what is right and what is not. If you have a question afterwards, see the frequently asked questions or get in contact.
Curious how AI mentions your company today?
Enter your website. Within 48 hours the report is in your inbox, free.
