ChatGPT Is Saying the Wrong Thing About Your Business. Here Is How to Actually Fix It.

By Cody Wise, Founder, Wise Media. Published 11 September 2026.

You cannot edit an AI answer directly. You fix the sources it reads, then wait for the crawl. For a local business that means your Google Business Profile first, because it feeds Google’s AI features, then your own site, then the third-party pages that carry most of the citations. Google confirms AI Overviews draw on the regular index, so a recrawl is the lever. Nobody at OpenAI or Google edits facts by hand on request.

Summary

• There is no correction button, but there are correction mechanisms. One competing page states flatly that you cannot report anything to OpenAI. That is wrong, and the exact forms are listed below.
• Only two of these are true fix-the-record tools: Google Business Profile edits and a Wikipedia talk-page request. Everything else feeds a review queue.
• Google states there is no special schema or AI text file required to appear in AI features. Anyone selling you markup as hallucination insurance is selling you something Google says it does not use for that.
• Most citations are earned media, not your website. Muck Rack’s analysis of more than 25 million cited links puts earned media at 84% of AI citations. Fixing only your own site fixes the smallest input.
• Retrieval-based answers can move in days once sources are recrawled. Training-data answers do not move until the model is retrained, on a schedule nobody publishes. Be sceptical of any page quoting a precise number of days.

Table of Contents

• Why AI systems get your business wrong
• Step one: find out which system is wrong, and where it read it
• Fixing Google AI Overviews and AI Mode
• Fixing ChatGPT, Perplexity and Copilot
• When AI confuses you with a same-named business
• The third-party sources that carry most of the weight
• How long a correction actually takes
• When it is defamatory rather than just wrong
• The remediation checklist
• FAQ

Why AI systems get your business wrong

Four causes, and the fix is different for each, which is why generic advice fails.

Cause	What it looks like	Where the fix lives
Stale source	Old hours, a disconnected phone number, a former address, a service you stopped offering	Google Business Profile, your site, directories
Bad third-party source	A directory or aggregator with wrong data that the model trusts	The directory itself
Entity confusion	Your details merged with a same-named business elsewhere	Disambiguation across every profile
Hallucination	Prices, reviews, staff or claims that exist nowhere	Nowhere direct, only crowding it out with correct sources

The fourth is the frustrating one because there is no source to correct. A model generated it. The only lever is to make correct, well-sourced information abundant enough that it wins the retrieval.

Step one: find out which system is wrong, and where it read it

Do not start filing reports. Start with twenty minutes of diagnosis, because the correction path depends entirely on the answer.

1. Reproduce it in each system separately. Google AI Overviews, Google AI Mode, ChatGPT, Perplexity, Copilot and Gemini. Ask the same three questions in each: what does this business do, what are its hours, and how do I contact it. Screenshot everything with the date visible.
2. Check whether the answer carries citations. If it does, the answer is retrieval-based and the cited page is your target. This is the single most valuable thing you can establish, and most owners skip it.
3. Open every cited source and find the wrong fact on it. Nine times out of ten it is sitting in plain sight on a directory listing nobody has looked at since 2019.
4. If there are no citations, ask the system where it got the information. The answer is not reliable, but it often names a directory worth checking.
5. Search your business name alone and look for other companies with the same or a similar name. That tells you whether you are dealing with an error or a merge.
6. Check your own site says the right thing. Owners are frequently surprised here. An outdated hours block in a footer template propagates everywhere.

At the end of this you know whether you are fixing a source, a merge, or a hallucination. Now the work is targeted.

Fixing Google AI Overviews and AI Mode

Google is the most fixable of the lot, because it is the only one that publishes how its AI features source content and gives you an owned surface that feeds them.

Google’s AI features documentation states that to be eligible as a supporting link in AI Overviews or AI Mode a page must be indexed and eligible to appear in Google Search, with no additional technical requirements. It lists keeping Business Profile and Merchant Center information up to date as a best practice. Both facts matter: the regular index is the pipe, and your Business Profile is a direct input for local facts.

1. Fix your Business Profile first

If you have not claimed it, claim it. If you have, audit every field: name exactly as you trade, categories, hours including holiday hours, phone, website, service area, and the services list. Wrong hours in AI answers almost always trace back here, and this is the one surface where you edit the record directly rather than requesting a change.

If someone else has made a bad edit to your listing, or the listing shows a competitor’s phone number, Google Maps carries a suggest an edit flow, and Google operates a business redressal process for listings that have been manipulated. Verify the current redressal form URL in Google’s own help centre before using it, because that URL has moved more than once.

2. Fix the page, then force a recrawl

Correct the fact on your own site, then use URL Inspection in Search Console and request indexing. Google’s documentation on asking Google to recrawl is the canonical reference. This is the step that actually moves AI Overviews, because the AI feature reads the index and the index has to be refreshed.

One thing worth doing while you are in there: make sure the corrected fact appears as plain readable text on the page, not only inside an image, a tab that loads on click, or a PDF. If a crawler cannot read it as text, it does not exist for this purpose.

3. Knowledge Panel, if you have one

Claiming still works. Search for your entity, use the claim option, verify through a linked YouTube, Search Console, X or Facebook account, then use the feedback option to flag an incorrect fact with supporting links.

Important distinction most articles blur: Google’s own guidance separates Knowledge Panels from Business Profiles, and directs businesses serving customers at a location or within a service area to a Business Profile instead. If you are a local service business, the Business Profile is your surface and chasing a Knowledge Panel is usually a detour.

4. Report the specific answer

AI Overviews carry a feedback control. Use it, but understand what it is. It is a signal into a quality review process, not an edit. Do it once, with a clear description, and do not treat it as the fix.

What you do not need

Google’s AI features documentation says this directly: you do not need to create machine readable files, AI text files, or markup to appear in these features, and there is no special schema.org structured data required. That contradicts a large amount of marketing currently sold as anti-hallucination infrastructure.

To be clear about what structured data is still worth doing, because this gets overcorrected in the other direction: Organization and LocalBusiness markup with accurate hours, phone, address and sameAs links remains sound practice. It makes your facts unambiguous to parse and it makes you eligible for rich results. What it does not do is what vendors claim, which is act as a verification layer that stops a model hallucinating. Google says it is not used that way. Do it because it is cheap and correct, not because someone sold it to you as a fix.

Similarly, llms.txt is not a lever. Google’s John Mueller described it in June 2026 as purely speculative, noting the file has existed for years and no AI system uses it. It is free, so leaving it in place costs nothing. Expecting it to do anything is a mistake.

Fixing ChatGPT, Perplexity and Copilot

Weaker levers, and you should know that going in so you spend your time where it pays.

ChatGPT

OpenAI runs OAI-SearchBot for ChatGPT search indexing, separate from GPTBot for training and ChatGPT-User for live fetches. Make sure you are not blocking OAI-SearchBot, because blocking it removes you from ChatGPT search entirely, which is a much bigger problem than a wrong fact. We covered which crawlers to allow in our guide to AI crawlers in robots.txt.

On reporting, one widely-shared page on this topic states that there is no way to report an error and no way to contact OpenAI about what the model says about your business. That is not accurate. Three routes exist:

• In-app feedback. Thumbs down on the response, with the option to flag a safety or legal concern.
• The content report form at openai.com/form/report-content. This is a policy and harm reporting channel rather than a factual correction desk, and it is worth using for genuinely damaging output.
• The privacy portal at privacy.openai.com, which handles requests to remove personal data from responses.

Read the privacy portal limitation carefully, because it is the thing people misunderstand. OpenAI’s own documentation makes clear the process is built around identifiable individuals and weighs factors such as public interest and a person’s role in public life, and that it does not remove information from external sites or search engines. It is not a mechanism for correcting your opening hours. If a false statement is about you personally and is damaging, that route is available. For ordinary business facts, it is not the right door.

What actually moves ChatGPT is the source layer. Correct the pages it is citing, make sure your own site is crawlable by OAI-SearchBot, and get the correct information onto the third-party pages discussed below.

Perplexity

PerplexityBot crawls and indexes, Perplexity-User fetches at answer time, and both respect robots.txt. There is a flag control under each answer. Perplexity’s own help documentation is refreshingly direct about what that does: it creates a record, it is not an override, and it does not guarantee any answer changes. Use it, then go fix the sources.

Copilot and Bing

Claim and correct your Bing Places listing, which is the Bing-side equivalent of your Business Profile and is neglected by almost every Canadian small business. If an outdated cached page is the problem, Bing Webmaster Tools has a content removal tool that lets a verified site owner request removal or refresh of outdated cached content. Verify site ownership first.

When AI confuses you with a same-named business

This is the hardest category and I will be straight about why: no platform offers a “this is not the same company” control. There is no merge-dispute form. What you are doing is making the distinction so consistent and so well-evidenced across every source that retrieval stops blending you.

What works, in rough order of effect:

1. Pick one canonical name and use it identically everywhere. Not “Wise Media”, “Wise Media Inc.” and “WiseMedia” across three profiles. One string, character for character.
2. Always pair the name with a distinguishing qualifier in titles, descriptions and bios. A city is the strongest one available to a local business.
3. Organization schema with sameAs linking every profile you control back to your canonical domain. Worth doing as entity hygiene. Be aware that no Google statement confirms sameAs as a disambiguation mechanism for AI answers specifically, so treat it as sound practice rather than a guaranteed fix.
4. A Wikidata item, which has a lower bar than Wikipedia and gives machines a structured identifier.
5. Explicit disambiguating text on your about page. A plain sentence noting you are not affiliated with the similarly named firm in another city gives retrieval something unambiguous to find.
6. Consistent founder and staff attribution, with Person schema. Person-level identity is frequently what pulls two entities apart.

This is slow work and it compounds. There is no fast version, and anyone offering you one is guessing.

The third-party sources that carry most of the weight

Here is the finding that should reorder your priorities. Muck Rack’s Generative Pulse analysis of more than 25 million links cited by ChatGPT, Claude and Gemini across 17 industries found earned media accounted for 84% of AI citations, with paid and advertorial content at 0.3%. That proportion has held in a consistent band across successive editions since mid-2025.

The same analysis found citation behaviour differs sharply by platform: ChatGPT cites in 96% of responses averaging around five citations with Wikipedia as its top domain, Gemini cites in 82% averaging eight with Reddit on top, and Claude cites in 55% averaging thirteen.

A related figure circulates as “91% of AI answers cite third-party content.” We could not attribute that specific number to a real study. It appears to be a rounded synthesis of several differently-defined measurements. Use the 84% earned media figure, which has a methodology attached.

The practical consequence: if you fix only your own website, you have corrected a small minority of the inputs.

Where to go, in priority order

1. The directories your diagnosis actually surfaced. Not a generic list of ninety citation sites. The two or three that were cited.
2. Wikipedia, carefully. If your business has an article carrying a wrong fact, do not just edit it. Wikipedia’s conflict of interest policy strongly discourages editing articles about yourself or your employer, and paid editors are very strongly discouraged from direct article editing, with disclosure of your affiliation required. The correct route is a talk-page edit request with sources. Direct editing is contemplated for clear defamation and serious errors, and even then you should disclose and escalate through the appropriate noticeboard.
3. Wikidata, which is more accessible and directly machine-readable.
4. Industry associations and trade bodies whose member listings tend to be trusted and are almost always outdated.
5. Local news and genuine earned coverage. This is the 84% category. It is also the slowest and the most durable.

How long a correction actually takes

You will find pages quoting confident numbers here: a median of six days for AI Overviews, specific percentile figures for ChatGPT, thirty to ninety days for full propagation. We traced those numbers and they come from vendor blogs with no disclosed methodology, sample size or replication. Do not plan against them.

What can be said honestly splits in two:

Answer type	What governs it	Realistic expectation
Retrieval-based, with citations	Crawl and index refresh of the cited sources	Days to weeks after the source is recrawled
Training-data-based, no citations	The model’s own retraining cycle	Unknown. No provider publishes a schedule

Google’s own wording on crawl frequency is that it can take anywhere from several days to several months depending on how often its systems determine a page needs refreshing. That is the honest ceiling, and it is why requesting indexing in Search Console is worth the two minutes.

How wrong are these systems, in general?

Worth knowing for calibration, with the caveat attached. The Columbia Journalism Review’s Tow Center tested eight AI search tools in March 2025 and found they collectively returned incorrect answers to more than 60% of queries about news article attribution, with wide variation between tools. That study measured citation and attribution accuracy for news content specifically, so do not stretch it into a general claim about business facts. It does establish that confident wrong answers are a structural property of these systems rather than a glitch affecting you alone.

When it is defamatory rather than just wrong

Wrong hours is an operations problem. An AI stating that your business is under investigation, has been sued, or has a criminal record is a different category.

Where the law stands, stated carefully. No Canadian court has ruled on whether an AI provider is liable for defamatory statements it generates about a business. That gap is the honest answer, and any page telling you otherwise is overstating it.

One Canadian case gets cited constantly here and it is the wrong case. Moffatt v. Air Canada, decided by the British Columbia Civil Resolution Tribunal in 2024, held Air Canada liable for negligent misrepresentation when its own customer-facing chatbot gave a passenger wrong information about bereavement fares, and rejected the argument that the chatbot was a separate legal entity. That is a business being liable for its own chatbot. It is close to the opposite of your situation, which is a third party’s AI describing you incorrectly. Do not let anyone cite it at you as though it settles your question.

Internationally, the pattern so far favours the AI providers. In Walters v. OpenAI in Georgia, the first US defamation suit against a generative AI company, OpenAI won summary judgment in May 2025, with the court finding the plaintiff was a limited-purpose public figure, that OpenAI had not acted with actual malice, and that the product’s own hallucination warnings undercut a recklessness finding. A suit against Google over Gemini output was filed in October 2025 and remains unresolved.

If output about your business is seriously damaging: document it thoroughly with dated screenshots and the exact prompts, use the provider’s legal or safety reporting channel rather than the general feedback control, and speak to a lawyer. Canadian defamation law, the defences available, and the anti-SLAPP regimes in several provinces all matter to how such a claim would run, and none of that is something to work out from a blog post. This article is not legal advice.

The remediation checklist

1. Document it. Dated screenshots, exact prompt, every system tested.
2. Establish whether the answer is cited or uncited. This determines everything that follows.
3. Open every cited source and locate the wrong fact.
4. Fix your Google Business Profile completely. Every field, including holiday hours and the services list.
5. Claim and fix Bing Places. Ten minutes, almost universally neglected.
6. Correct the fact on your own site as plain crawlable text, then request indexing in Search Console.
7. Correct every third-party source your diagnosis named. Directories first, then associations.
8. For Wikipedia, use a talk-page edit request and disclose your affiliation. Do not edit the article directly.
9. Add or correct Organization and LocalBusiness schema with sameAs. Entity hygiene, not a magic fix.
10. Add explicit disambiguating text if a same-named business is in play.
11. File one feedback report per platform. Once, clearly. Then stop and let the sources do the work.
12. Re-test every two weeks and keep the dated record. This is also your evidence if it ever becomes a legal matter.
13. Confirm you are not blocking OAI-SearchBot or PerplexityBot in robots.txt.

Frequently asked questions

Can I contact OpenAI to correct what ChatGPT says about my business?

You can report it, but not edit it. In-app feedback with a safety or legal flag, a content report form, and a privacy portal for personal data removal requests all exist. None of them is a factual correction desk for business information, and OpenAI does not hand-edit facts on request. The reliable route is correcting the sources ChatGPT retrieves from.

Why does Google’s AI say my business is permanently closed?

Almost always a Business Profile marked closed, a directory listing carrying a closed status, or a stale cached page. Check and correct your Business Profile first, since Google names keeping it up to date as a best practice for its AI features, then find and fix any directory repeating it, then request a recrawl of your own site in Search Console.

Does schema markup stop AI from getting my business wrong?

Not according to Google. Its AI features documentation states you do not need machine readable files, AI text files or special schema.org structured data to appear in these features. Accurate Organization and LocalBusiness markup remains worth having for clarity and rich-result eligibility, but treat any claim that schema functions as hallucination protection with scepticism.

How long does it take for an AI answer to update after I fix the source?

If the answer cites sources, it can update within days to weeks once those sources are recrawled. If the answer comes from training data with no citations, it will not change until the model is retrained, and no provider publishes that schedule. Google notes that crawling can take anywhere from several days to several months. Be wary of any page quoting a precise median.

What do I do if AI confuses my business with another one that has the same name?

No platform has a control for this, so it is remediated through consistency. Use one canonical name string everywhere, pair it with a geographic qualifier, add Organization schema with sameAs links to every profile you control, create a Wikidata item, and put explicit disambiguating text on your about page. It is slow and compounding rather than a single fix.

Is it worth fixing my website if most AI citations are third-party?

Yes, but do not stop there. Muck Rack’s analysis of over 25 million cited links found earned media makes up roughly 84% of AI citations. Your own site is necessary because it is the reference other sources check, and because Google’s AI features read it through the regular index. It is not sufficient on its own.

Can I sue an AI company in Canada for saying something false about my business?

No Canadian court has ruled on that specific question. The Air Canada tribunal decision often cited here concerned a company’s liability for its own chatbot, which is a different situation. The first US defamation case against an AI provider was decided in the provider’s favour in 2025. If output about your business is seriously damaging, document it, use the provider’s legal reporting channel, and get advice from a lawyer.

The bottom line

Two mechanisms in this entire article let you change the record directly: your Google Business Profile, and a Wikipedia talk-page request. Everything else is a signal into somebody’s queue. That sounds bleak and it is actually clarifying, because it tells you where to spend the afternoon.

Fix the Business Profile completely. Fix the two or three third-party pages the answer was actually citing. Correct your own site and request a recrawl. Then stop refreshing the answer and let the crawl happen. The owners who get this wrong are the ones who file fifteen feedback reports and never open the directory listing that caused it.

Get an AI visibility and accuracy audit

We run the diagnosis described above across Google, ChatGPT, Perplexity, Copilot and Gemini, identify the exact sources producing the wrong answer, and remediate them. If an AI is telling your customers something untrue, tell us about your project and we will find out where it is reading it from.

Related reading: which AI crawlers to allow in robots.txt, measuring AI search visibility in Search Console and Bing, and our website growth packages.

Sources

• AI features and your website, Google Search Central
• Ask Google to recrawl your URLs, Google Search Central
• Claim a knowledge panel, Google Help
• Wikipedia: Conflict of interest, Wikipedia
• We compared eight AI search engines, Columbia Journalism Review Tow Center, March 2025
• What is AI reading, Muck Rack Generative Pulse, May 2026
• Google says llms.txt is purely speculative for now, Search Engine Journal, June 2026

This article is general information and not legal advice. It does not assess any particular business’s situation. Consult a qualified lawyer before acting on anything in the defamation section.