AI Is Getting Your Business Facts Wrong. Here Is How to Fix It

When AI states wrong facts about your business, the cause is almost always a stale source it trusts more than your website. Fix it in this order: your own site and schema first, then your Google Business Profile, then the big aggregators feeding everyone else, then the specific pages the engines cite. Corrections show up in live search within days and in the models themselves over months.

Being found is not the same as being right

Most of the advice about AI search, ours included, is about getting named at all. That is the right first problem. But there is a second one waiting behind it, and it is quieter, because nothing looks broken from where you sit.

Someone asks an AI where to go. It names you. Then it tells them you close at 6, and you have closed at 8 for two years. It gives a phone number you dropped when you changed providers. It says you do not do the exact thing they are calling about.

They never call. You never find out. Your ranking looks fine.

We ran into this on our own business earlier this year. We changed our main number, updated Google the same afternoon, and assumed that was that. Months later the old number was still living on a dozen directories, and the AI answers were still handing it out.

Where the wrong facts actually come from

Four places, roughly in order of how much damage they do.

1. A source the engine trusts more than you. AI engines weigh third party sources heavily, because a business describing itself is not independent confirmation. If a directory with real authority says you close at 6, that can outrank your own site saying 8. This is the same mechanism that gets you cited in the first place, working against you.

2. Your own site, quietly out of date. A footer nobody has touched since the site launched. An old location page still live at a URL you forgot. A PDF menu from 2023 sitting in your uploads folder. Machines read all of it and have no way to know which one you meant.

3. Training data that predates the change. Part of what a model knows was baked in months or years ago. Even when it searches the web live, it can blend what it retrieves with what it already believes. This is the part you cannot fix directly, only outweigh.

4. Nobody ever said the right thing loudly enough. Sometimes the answer is not wrong so much as invented from thin context. If no reliable source states your hours in a machine readable way, the engine fills the gap with what businesses like yours usually do.

When two sources disagree about you, the engine picks one. It does not always pick yours.

Find every wrong fact in fifteen minutes

Open a private window so your own history does not color the results, then ask each engine the same four questions. ChatGPT, Perplexity, Google AI Overviews, and Gemini is a good spread.

• What are the hours for [your business name] in [your city]?
• What is the phone number and address for [your business name]?
• What services does [your business name] offer?
• Tell me about [your business name].

Write down every answer verbatim. Do not correct them in your head as you read, which is the easy mistake. You are looking for four categories:

What to check	What wrong looks like
Hours	Old hours, missing holiday hours, "permanently closed"
Contact	A dead number, an old suite, the wrong side of town
Services	Something you dropped, or a flagship service missing
Description	The wrong category, the wrong size, the wrong specialty

Then ask one more, which is the one owners tend to skip: Where did you get that? Most engines will name the source. That is your repair list, handed to you.

Fix them in this order

The order matters more than the effort. Fixing the long tail first is how people spend a weekend on this and see nothing change.

First, your own site. Put the correct hours, address, phone, and service list in real HTML text, not inside an image and not only in a PDF. Then add LocalBusiness schema so the facts are machine readable rather than inferred. This is the one source you fully control, and every other fix points back to it.

Second, your Google Business Profile. It feeds Google AI Overviews and Gemini directly, and plenty of other engines lean on it too. Hours, holiday hours, categories, services, phone. If you only do two things, do this one and the one above.

Third, the big aggregators. Apple Business Connect, Bing Places, Yelp, Facebook, and the data brokers your industry sits on. These syndicate outward, which is why they are worth more of your time than any single small directory. One inconsistent listing here quietly contradicts you in a hundred places.

Fourth, the specific sources the engines named. You asked where the answer came from. Go correct those, whatever they are. A single wrong directory that happens to be the one Perplexity cites is worth more of your attention than fifty nobody reads.

What you cannot fix, and what to do instead

You cannot edit a model's memory. If something wrong was baked into training, no amount of correcting directories reaches back into it.

What you can do is outweigh it. Make the correct version of the fact true in more places, more consistently, and in more machine readable form than the wrong one. Engines that search live will find the corrected sources. Engines relying on memory get retrained, and the corrected web is what they get retrained on.

Expect the live search side to catch up within days to a couple of weeks. Expect the memory side to take months. That gap is normal and it is not a sign your fixes failed.

Keep it from happening again

The reason wrong facts persist is that nobody owns checking. Two habits fix that:

• Run those four prompts on the first of every month. Ten minutes, no tools, same routine as checking whether AI recommends you at all.
• Treat any change to hours, phone, address, or services as a distribution job, not an edit. The change is not done when your site is updated. It is done when the aggregators match. Our Google Maps and local SEO checklist has the 30 minute audit for exactly this.

Consistency is unglamorous and it is still the highest return work in this whole discipline. The businesses AI describes correctly are almost never the ones with the cleverest content. They are the ones that say the same true thing everywhere.

FAQs

How long does it take for AI to stop repeating the wrong information?

Engines that search the web live usually reflect corrections within days to two weeks once the underlying sources are fixed. Answers drawn from a model's training data can take months, because they only change when the model is retrained.

Why does AI trust a random directory over my own website?

Because a business describing itself is not independent confirmation. Engines weight third party sources to avoid taking marketing claims at face value. The fix is not to argue with that, it is to make the third party sources agree with you.

What if a listing I do not control has the wrong details?

Most major platforms have a "suggest an edit" or claim flow, and claiming the listing is usually faster than reporting it. For a listing with no owner path at all, focus on outweighing it: get the correct fact into the higher authority sources that more engines actually cite.

Does fixing this help my Google ranking too?

Yes. Consistent name, address, and phone across the web is one of the strongest local ranking signals there is, and it is free. The AI visibility benefit and the traditional local SEO benefit come from the same work.

Is it worth fixing small directories nobody visits?

Only after the big ones. Nobody reads them, but machines do, and a contradiction is a contradiction. Fix your site, Google, and the major aggregators first, then work the list the engines actually cited.

---

Author: Roger Wong Won, founder of Capture That Media, San Antonio's AI Visibility specialist. Want to know what the engines are saying about you right now? Run a free AI visibility audit.