ChatGPT Has Wrong Information About My Business: How to Fix It

Often you hear it from a customer first. They send a screenshot: ChatGPT says you're closed on Sundays, quotes last year's prices, or lists a service you stopped offering. Sometimes it has mixed you up with another business that shares your name.

You can't edit what ChatGPT says about your business, but you can fix what it reads. Wrong answers trace back to a short list of causes: an outdated page, a wrong directory listing, a business with a similar name, or too little information about you. Each cause has its own fix.

Start by saving the exact answer and working out whether it came from a live web search or from training. Then correct the source, report the answer, and re-test under the same conditions. Search-based answers can change after the next crawl. Training-based answers change only when a newer model arrives.

This is the detailed version of the short fix in my guide to getting recommended by ChatGPT.

What should you do first when ChatGPT says something wrong about your business?

When ChatGPT says something wrong about your business, work through six steps in order: save the answer, find which layer it came from, trace the source, fix the source, report the answer, and re-test under the same conditions.

The order matters, because each step depends on the one before it.

1. Save the answer. Screenshot it, copy the exact wording, and note the date, the assistant, the question and whether web search was on. You need this record to prove a change later.
2. Find the layer. Ask the same question with web search on, then ask the assistant to answer without searching. The diagnosis table shows what each result means.
3. Trace the source. Open every citation and find the sentence that states the wrong fact.
4. Fix the source. Correct your own pages first, then listings and other sites.
5. Report the answer. Use each assistant's feedback route once the source is fixed.
6. Re-test and log. Run the same questions under the same conditions two to four weeks later and record how often the error still appears.

Why does ChatGPT say wrong things about businesses?

ChatGPT says wrong things about businesses for five reasons: it read an outdated source, it confused you with a business of the same name, it found too little about you and filled the gap, a third-party listing was wrong, or your own pages contradict each other.

OpenAI says as much in its own help center. Its article on whether ChatGPT tells the truth lists "incorrect definitions, dates, or facts" among the errors it calls hallucinations, and its guide to searching the web with ChatGPT warns that search results and citations "can be incomplete, outdated, or incorrect".

The scale of the problem is easier to see in research on news, where facts are checked by professionals.

The European Broadcasting Union and BBC study had journalists from 22 public service broadcasters review more than 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity. Businesses rarely get that scrutiny. Errors about them can go unnoticed until a customer repeats one.

Why are small businesses described wrongly more often?

Small businesses are described wrongly more often because there is less written about them, and models guess more on facts they have rarely seen. Researchers from OpenAI and Georgia Tech explained the mechanism in the September 2025 paper Why Language Models Hallucinate by Adam Tauman Kalai, Ofir Nachum, Santosh Vempala and Edwin Zhang. They show that training and evaluation "reward guessing over acknowledging uncertainty", and that for arbitrary facts such as birthdays, a model's error rate after training is at least the share of those facts that appear only once in its training data.

A national brand is described on thousands of pages. A two-location clinic might be described on five, and two of those might be out of date. The fix follows directly: more accurate, consistent statements about you on trustworthy pages.

What happened when Google's AI Overview said a bakery was closing?

Google's AI Overview told searchers in July 2026 that Anna Mae's Bakery and Restaurant in Millbank, Ontario would close for good on July 31. It wasn't closing. Google had confused it with a bakery of the same name in Illinois that was closing, as CBC News reported on July 23.

The error appeared over the weekend of July 18 and 19, according to PPC Land's account. Customers emailed, called and visited to ask whether it was true, and the owner, Amanda Herrfort, ran an online campaign pointing people to the bakery's own website and Facebook page. She also typed a correction into Google's AI search box. A Google spokesperson told CBC that errors like this come from gaps in high-quality information on a topic.

The case has every ingredient: two businesses with one name, thin information about one of them, and an owner who first saw the error in a screenshot sent to her phone.

How do you find where the wrong information comes from?

You find where wrong information comes from by asking the same question with web search on and off, in a clean session, and then opening every citation. The pattern of results tells you which layer holds the error.

Test in a temporary chat, so your saved memories and past chats don't shape the answer. OpenAI's search guide says that when memory is on, ChatGPT may use saved memories when it rewrites your question into search queries, and it uses an approximate location from your IP address for local results. Ask from the area your customers are in, or name the city in the question.

A fact that is wrong only without search is the slowest to change, because it sits in the model's training. While you wait for newer models, make the correct version easy for search to find. Questions that trigger a search then get the right answer.

Which prompts reveal the source of a wrong answer?

These four prompts reveal the source of a wrong answer. Run them in a temporary chat and replace the brackets.

Search the web: what do you know about [Business name] in [city]? List each fact on its own line with the source you used for it.

You said [the wrong fact]. Which page did that come from? Give me the exact URL and quote the sentence.

Without searching the web, tell me what you already know about [Business name].

Are there other businesses or people called [Business name]? How would you tell them apart from the one in [city]?

Treat the assistant's account of its sources as a lead, not proof, and open the links yourself. If a cited page doesn't contain the fact, the error came from somewhere else. The third prompt shows whether that place is training data.

How do you fix wrong information on your own website?

You fix wrong information on your own website by stating every key fact once, plainly, in text on the page that owns it, and by making every other page and profile agree with it.

1. Write a facts list. Your legal and trading name, what you do and for whom, locations and service areas, hours, price ranges, founding year, owners, and services you don't offer. Keep it on your About page, which works as your business's home base for entity information.
2. State changes out loud. Deleting an old fact leaves the old pages elsewhere as the only source. Write the change instead: "We stopped offering same-day delivery in March 2026" or "We moved from Leeds to Manchester in January 2026."
3. Retire old pages. Find old price lists, closed location pages and expired offers, then update them or redirect them to the current page. Search your own site for the old figure.
4. Add structured data. Mark up your home or About page with Organization or LocalBusiness structured data: name, URL, logo, address, phone and sameAs links to your official profiles. Google's Organization structured data guide says this markup can help Google understand your organization.
5. Keep facts in HTML text. Prices inside images and hours inside scripts may never be read. Check that AI crawlers can reach the page: OAI-SearchBot for ChatGPT, PerplexityBot and Claude-SearchBot. My guide to AI crawlers and robots.txt has the rules. OpenAI's crawler documentation says a robots.txt change takes about 24 hours to reach its systems.
6. Ask for a recrawl. Use URL Inspection in Google Search Console and submit the URL in Bing Webmaster Tools, or turn on IndexNow. Google's recrawl guide says crawling can take anywhere from a few days to a few weeks.

The markup below adds a disambiguatingDescription, which Schema.org defines as a short description "used to disambiguate from other, similar items". Google doesn't list it among its recommended Organization properties, so treat it as a clarity aid, not a ranking signal.

{
  "@context": "https://schema.org",
  "@type": "Bakery",
  "name": "[Business name]",
  "url": "https://www.example.com/",
  "logo": "https://www.example.com/logo.png",
  "description": "[What you do, for whom, in one sentence]",
  "disambiguatingDescription": "[Business name] in [town, region], founded in [year]. Not affiliated with [the other business], [its location].",
  "address": { "@type": "PostalAddress", "streetAddress": "[Street]", "addressLocality": "[Town]", "addressRegion": "[Region]", "addressCountry": "[Country code]" },
  "telephone": "[Phone]",
  "sameAs": ["[Google Business Profile URL]", "[Facebook URL]", "[LinkedIn URL]"]
}

Most of this is basic SEO housekeeping, a point Peter Rota made when he joined me on the podcast.

On the same episode, Peter described the gap his team found between what customers think about a brand and what language models say about it, and how a GEO audit surfaced it. The full conversation covers what that audit turned up.

How do you fix wrong information on other websites?

You fix wrong information on other websites by correcting the profiles you control first, then asking publishers to update the rest, starting with the sources AI answers cite most for businesses like yours.

For local businesses, that list starts with listings. BrightLocal's August 2026 study of AI citations, which covered 1,355 locations in the US, UK and Australia across ChatGPT, Google AI Overviews and AI Mode, found AI answers cited the business's Google Business Profile for 94% of them (1,276 of 1,355). ChatGPT relied far more on Yelp, its most-cited local source, followed by TripAdvisor and OpenTable. Yext's October 2025 analysis of 6.8 million AI citations found websites made up 44% and listings 42%. Both are vendor studies. Both point the same way: your profiles carry nearly as much weight as your site.

Work through the profiles in this order, starting with the ones you control.

• Google Business Profile. Edit it while signed in with the Google Account that manages it. Google's Business Profile help says changes are reviewed before they go live, and that Google can also update a profile itself, for example through automated outreach. Anyone can suggest an edit on Google Maps, so check the profile monthly.
• Bing Places. Microsoft moved Bing Places to bing.com/forbusiness in October 2025, according to the Bing blog. Claim the listing and match it to your Google profile.
• Yelp, TripAdvisor, OpenTable and industry directories. Claim each listing that applies to you and correct names, hours, categories and descriptions. Add the directories that matter in your sector, such as healthcare or legal directories.
• Articles, blogs and roundups. Send the publisher a short note with the exact wrong sentence, the correct fact, and a link to the page that proves it.
• Wikipedia. Wikipedia's conflict of interest guideline strongly discourages editing an article about your own business. Post an edit request on the article's Talk page using the Edit COI template, state the exact change, and cite an independent source.
• Wikidata. Wikidata's sources guide says most statements should show where the data comes from, so add a reference with any correction.

Subject: Correction request: [Business name] on [page title]

Hi [Name],

Your page [URL] says "[exact wrong sentence]". This is out of date: [correct fact], as shown on [link to the page that confirms it].

Could you update it? I'm happy to send anything else that helps.

Thank you,
[Your name], [role], [Business name]

Keep the request short and specific: one exact sentence, the correct fact, and proof. Wikipedia's own edit-request guidance makes the same point, noting that volunteers are less likely to respond to unclear or open-ended requests.

How do you stop ChatGPT confusing you with another business?

You stop ChatGPT confusing you with another business by giving it details that only fit you, such as your location, category, founding year and founder, and by repeating those details in the same words on every page and profile.

Five habits separate your business from a namesake.

1. Use a qualified name in key places. Pair your name with your town or category in your title tag, About page heading and listings, for example "Anna Mae's Bakery & Restaurant, Millbank, Ontario".
2. Publish facts the namesake can't share. Founding year, founder, address and the services you offer separate two businesses faster than any description.
3. Link only to your own profiles. Your sameAs links and profile links connect your pages to the right entity in Google's Knowledge Graph and in AI answers.
4. Add a short "not affiliated" line only when the confusion is real. One factual sentence on your About page, such as "We are not affiliated with [the other business] in [location]", gives assistants a clear statement to quote.
5. Correct the mixed-up sources. Directory entries and articles that blend the two businesses keep feeding the error until they're fixed.

This is the practical side of entity SEO: making sure every source describes the same business in the same way.

How do you report a wrong answer to ChatGPT, Google, Copilot, Perplexity and Claude?

You report a wrong answer with the feedback control under it in each assistant. None of them promises that a report will change the answer, so report it after the source is fixed and include the correct fact with a link.

When the wrong information is personal data about a person, such as an owner's name or history, OpenAI offers a separate route. Its personal data removal page explains how to submit a "Remove personal data from ChatGPT responses" request through the Privacy Portal, with evidence and a government ID or other proof of identity. OpenAI says it "cannot independently investigate disputed facts", so the evidence has to do the work. It covers people, not company facts.

How long does it take for ChatGPT to show the right information?

No AI assistant publishes a timeline for corrections. Answers built on web search can change once the corrected page is crawled again, while answers built on training data change only when a newer model is trained on newer data.

The pieces you can time are the crawling steps:

• Robots.txt changes: about 24 hours for OpenAI's search systems, and up to 24 hours for Perplexity, according to their crawler documentation.
• Google recrawls: from a few days to a few weeks after you request one, per Google's recrawl guide.
• Training data: no published timeline. Anthropic's models overview, for example, lists a reliable knowledge cutoff of June 2026 for its current Claude models, which are in use in October 2026, so anything that changed after June is invisible to those models without search.

Re-test two to four weeks after a fix. Treat a training-data error as a longer project: keep the web consistent, and the next generation of models learns the corrected version.

How do you check that the fix worked?

You check that a fix worked by re-running the same questions under the same conditions and counting how many answers still contain the error, because a single answer proves nothing.

The re-test follows five rules.

1. Use the same questions you saved in step 1, word for word.
2. Keep the conditions the same: temporary chat, signed out where the assistant allows it, the same location, search on.
3. Run each question 3 to 5 times per assistant. AI answers change from run to run: SparkToro's January 2026 study found less than a 1 in 100 chance of the same list of brands appearing twice.
4. Record the error rate: "wrong in 4 of 5 runs" before, "wrong in 1 of 5" after.
5. Repeat monthly with the same set, as part of the routine in my guide to measuring AI search visibility.

The free correction log below keeps the record for you, from the first screenshot to the final re-test.

What should healthcare practices do when AI gets their details wrong?

Healthcare practices should treat a wrong AI answer about their services, insurance, hours or clinicians as urgent, because health falls under what Google calls Your Money or Your Life topics, where wrong information can affect someone's wellbeing. Fix those facts at every source before any other marketing work.

Four checks matter most for clinics and practices.

1. Publish a facts page for each location. List the services you offer and don't offer, the insurance and payment options you accept, your hours, and the clinicians who work there with their credentials. Assistants repeat what they can find stated plainly.
2. Match your health directories to it. Correct the listings patients use, such as Healthgrades or Zocdoc where you appear, alongside your Google Business Profile.
3. Use a specific Schema.org type. Mark up location pages as MedicalClinic and individual doctor pages as IndividualPhysician, with sameAs links to official profiles only.
4. Re-check after every change. A new insurance contract, a clinician leaving or new hours are the moments facts go out of date across the web. Re-run your questions two to four weeks later.

My E-E-A-T checklist for healthcare websites covers the trust signals health pages need, and my healthcare SEO work includes keeping practice facts consistent across every source AI assistants read.

What belongs on a checklist for fixing wrong AI answers about your business?

A checklist for fixing wrong AI answers covers recording the error, diagnosing the layer, fixing your site, fixing other sources, reporting, and re-testing.

What else should you know about wrong AI answers about your business?

Five more questions matter once the main fixes are done.

Does the thumbs-down button fix a wrong answer?

No, the thumbs-down button doesn't fix a wrong answer on its own. It sends feedback to the company that runs the assistant, and none of the help pages promises a correction. The lasting fix is the source the answer was built on.

Should you block ChatGPT's crawlers to stop wrong answers?

No, blocking ChatGPT's crawlers makes wrong answers harder to fix. Blocking OAI-SearchBot keeps your corrected pages out of ChatGPT search, which is where corrections show up first, and blocking GPTBot only affects future training. Neither removes what a model already learned.

Do you need a Wikipedia page to fix wrong AI answers?

No, most businesses fix wrong AI answers without a Wikipedia page. Your website, listings and the articles that mention you carry the weight for most local and small businesses. If a Wikipedia article about you exists and is wrong, request the edit on its Talk page rather than editing it yourself.

What if an AI answer about your business is negative but true?

If an AI answer about your business is negative but true, reporting it won't remove it, because the feedback routes exist for inaccurate answers. The answer reflects real reviews or coverage, so the fix is to change what new sources say: resolve the complaint, reply to the reviews it comes from, publish what you changed, and ask satisfied customers for honest reviews.

Reviews now shape the answer directly. BrightLocal's 2026 Local Consumer Review Survey found 82% of consumers read AI-generated review summaries. Never post or buy fake reviews to tip the balance: the US Federal Trade Commission's 2024 rule bans them.

Can you take legal action over a wrong AI answer?

Yes, legal action is possible in some cases, depending on the harm and the law where you operate, so speak to a lawyer if an answer damages your business. Gemini has a "Report legal issue" option, and OpenAI's privacy route covers personal data about individuals. This guide isn't legal advice.