AI Is Giving Customers the Wrong Information About You

First, find out which ChatGPT you are arguing with

That test takes five minutes, and almost nothing published on this topic tells you to run it first. It is the difference between a Tuesday afternoon of edits and a problem you cannot buy your way out of.

With search on, ChatGPT shows its citations. Open every one of them. The wrong fact has a home address, and it is sitting on one of those pages.

Calling this a hallucination is usually wrong, and the word does real damage because it makes the problem sound unfixable. On local service businesses we almost always find the model retrieving accurately from a page that simply stopped being true. Google's own documentation describes its AI features as rooted in core Search ranking systems, using retrieval and query fan out, not a separate magic index. The practical consequence carries across engines: answers get assembled out of pages, so a wrong answer means a wrong page.

The wrong fact is probably on a page you own

BrightLocal ran 800 manual local business searches in ChatGPT and logged the first ten sources behind each answer. Business websites were the source 58 percent of the time, ahead of business mentions at 27 percent and online directories at 15 percent.

Read that ordering before you email a single directory. The most likely home for your dead phone number is your own domain.

In rough order of how often we find them:

1. A footer nobody has touched since the last redesign
2. An old contact or location page that is still indexed
3. A PDF rate sheet, service menu or holiday flyer with the number printed inside it
4. LocalBusiness markup in the theme with hours hardcoded by a developer who left
5. A blog post announcing new summer hours, three summers ago

Old pages are not a fringe risk here, they are the norm. Ahrefs' analysis of 1.4 million ChatGPT prompts found the median age of a cited search result was roughly 500 days. Seer Interactive's study of 7,683 cited pages found 75 percent had been updated within the previous year, and that far more cited pages looked fresh by their update date than by their publish date. Engines favor maintained pages. A five year old page you never maintain is a page that gets quoted back at your customers with 2019 in it.

The one correction path that reliably works: claim the feeds

There is no listing inside ChatGPT. There is, however, a company whose listing data ChatGPT uses, and it has a claim form.

Foursquare's own business listings product page states that Foursquare Places powers leading consumer apps like OpenAI and Apple Maps, and that a claimed account lets you keep contact information, hours and website details current across its ecosystem. BrightLocal's later testing across multiple assistants found directories and citations feeding business information in every industry it checked, which is why the feed layer earns an afternoon even though you cannot see inside it.

That single fact reorders the whole job. Nearly every article ranking for this question opens with Google Business Profile and never mentions Foursquare once. For Google's AI Overviews and AI Mode, Google Business Profile is correctly the first move. For ChatGPT specifically it is not the feed being named on the record.

Apple is unusually blunt about what this tool is for, which is why it belongs above most of the tactics being sold alongside it.

Apple describes Business Connect as a free tool that lets a business claim its place card across Apple Maps, Messages, Wallet, Siri and other apps. Yelp states plainly that a business page costs nothing and lets an owner change phone number, address, website and hours. Google documents the full list of editable Business Profile fields and says it reviews changes before they go live. Microsoft moved Bing Places to bing.com/forbusiness in late 2025 and describes it as a free platform for creating and managing listings.

Google Business Profile, Apple Business Connect, Bing Places and a Yelp page are each free to claim on the record. Foursquare puts a verification step in front of the listing before you can manage it. One afternoon covers the list, and that is the whole confirmed-available column.

Use this graphic on your site

Free to republish with a link back to this page. Copy the embed code:

<a href="https://timpsonmarketing.com/blog/chatgpt-has-wrong-information-about-my-business"><img src="https://timpsonmarketing.com/assets/infographics/chatgpt-has-wrong-information-about-my-business.svg" alt="Process flow tracing wrong business information from its source to the customer, with correction availability marked at each stage" width="1200" style="max-width:100%;height:auto"></a><p>Graphic by <a href="https://timpsonmarketing.com/blog/chatgpt-has-wrong-information-about-my-business">Timpson Marketing</a></p>

Somebody else may be changing your hours

Here is the part owners find hardest to believe. You can fix your Google hours on Monday and find them wrong again on Friday, without anyone in your company touching them.

Google lets any signed in user suggest an edit to a place, and says it reviews those edits before deciding whether to change the listing. Connected third party apps can push data in. Other managers on the profile can change anything at will.

The same Sterling Sky piece notes that Google may update business information based on what it crawls on your website or social profiles. That closes a loop nobody draws in these articles: a stale page on your own site can rewrite your Google hours, which then feed the local pack, which then feed the AI answer that sends someone to your door at 7pm. Fixing the profile without fixing the page just resets the clock on the same failure.

So audit access first. Revoke connected apps you do not recognize, remove owners and managers who left, and check the profile weekly for a month after any correction. If your profile is also losing local pack position or the categories are wrong, fix those in the same sitting, because you are already in there.

Where no correction mechanism exists

This is the section the ranking results will not write, so here it is.

OpenAI has no business correction channel. There is no profile, no dispute form, no support queue that edits what a model says about your company. The in-product thumbs down with a written explanation is a signal into a training and evaluation pipeline. It is not an edit, it changes nothing today, and treating it as a fix wastes weeks.

The privacy route is built for people, not companies. The European privacy group noyb documented a case where OpenAI refused a request to rectify inaccurate information generated about an individual, saying it could filter or block data on certain prompts but not correct it. If that is the answer for a named person under GDPR, a US business owner asking for a phone number change should plan accordingly.

Two products sold as the fix are not the fix. An llms.txt file will not correct anything: Ahrefs checked 137,210 domains and found 97 percent of llms.txt files received zero requests in May 2026, while Google's documentation says you do not need AI text files to appear in Search or its generative features. Schema will not do it either: Ahrefs tracked 1,885 pages that added JSON-LD against 4,000 controls and found AI Overview citations fell 4.6 percent with the other engines statistically flat.

The honest nuance, which is easy to lose in the pile-on: you should still fix the hardcoded openingHours in your markup. Not because it buys an AI citation, but because it is a wrong fact on a page you own, and Google can read it. Fix it as data hygiene, not as a growth lever. We take the same line on whether schema markup is worth paying for.

What this actually costs you in time

The strategic move for the memory case is quiet and effective: make the retrieved answer so unambiguous that search-enabled responses stop deferring to stale memory. That is the same work as the retrieval fix, which is why we sequence it identically and why it is worth doing even when the diagnosis looks hopeless.

One more thing worth saying out loud. If AI answers are wrong about you, the more expensive problem is usually that they are not mentioning you at all, and being wrongly described is at least evidence that you exist in the data. Fix the facts first, then work on presence. Our full method for that is in GEO marketing, and the mechanics of where AI chatbots get their data explain why the listing feeds carry so much weight. If you would rather have someone trace the wrong fact for you, that is exactly what the AI visibility audit does first.

If you want the rest of the series, start with how many customers are genuinely asking ChatGPT for local recommendations, then the differences between AI Overviews, AI Mode and ChatGPT, and how local businesses rank in AI Overviews. The whole blog library is organised by the question an owner actually asks.