When ChatGPT Gets Your Business Wrong and How to Correct It

When ChatGPT gets your business wrong, the error usually traces back to one of two places, which are the data the model learned from or the web pages it found when it searched. Working out which one applies decides what can be fixed and how quickly. It is also the first step in any work on AI search visibility for UK companies. No route guarantees a correction, so the aim is to fix every source you control and report the rest.

The steps below start with diagnosis, then move through your own website, structured data, listings, Wikipedia and Wikidata, OpenAI’s reporting routes and UK GDPR rights. Each one deals with a different source of the error, so working through them in order avoids spending effort on a channel that has nothing to do with the problem.

Work Out Whether the Answer Comes From Training Data or Live Search

A wrong answer comes either from what the model learned during training or from pages ChatGPT retrieved by searching the web, which need different fixes. OpenAI’s guidance on whether ChatGPT tells the truth explains that without search, responses are based on what the model learned during training, while with search it can access and cite web sources for more recent information.

Answers drawn from training data are fixed at a point in time. The same OpenAI page says models are trained on data up to a knowledge cutoff and do not include later events unless tools such as search are used, which is why an old address, a former director or a service you stopped offering can keep appearing long after you changed it. It also warns that the model can produce incorrect facts while sounding confident, a behaviour usually called hallucination.

Answers drawn from live search behave differently. They come with citations, so the wrong claim can usually be traced to a specific page. Changing that page is the most direct fix available. The table below sets out how to tell the two apart.

Most errors about a business fit one of these rows, although some answers mix more than one. Note which row each wrong claim fits before moving on, because that note decides which of the later steps will make a difference.

Record Exactly What ChatGPT Says First

Before changing anything, record the wrong answer in full so you have evidence for any report and a baseline to check against later. OpenAI’s overview of how ChatGPT and its models are developed notes that the same question may produce different answers across different queries, so a single screenshot rarely shows the whole picture.

• The exact prompt used, written down word for word
• A screenshot of the full answer, including any citations
• The date, the account used and whether you were signed in
• A shared link to the conversation for any report you make later
• Each separate wrong claim, listed on its own line

Run the questions buyers are likely to ask, not only your company name. Questions about who provides a service in your area, how your business compares with a competitor and what your company is known for often surface errors that a plain name search never shows.

Repeat the same prompts several times, starting a fresh conversation each time. An error that appears in every attempt deserves more attention than one that appears once and then disappears.

Check the Sources ChatGPT Cites

When an answer includes citations, open every source it lists, because the wrong claim often sits on one of those pages. OpenAI’s guide to searching the web with ChatGPT says responses that use web search may include citations and that selecting Sources shows the cited sources and other relevant links. The same page warns that search results and citations can be incomplete, outdated or incorrect.

If an answer has no citations, open View all tools, select Search and ask the same question again. The guide describes this as the way to start a web search manually. The citations in the new answer then show which pages ChatGPT relies on when it looks for current information.

Sort each cited page into one of three groups. Your own pages can be changed today, pages owned by someone else need a correction request and pages about a different organisation with a similar name point to a confusion problem rather than a factual one.

Check the date on each source as well. An old press release, an archived directory entry or a news story about a previous owner can still be cited years after the facts changed, which explains answers that were true once but are no longer accurate.

Fix the Information on Your Own Website

Your own website is the one source you fully control, so correct it first and make the accurate facts easy to find. Update the about, contact and service pages so the company name, registered address, trading locations, services, leadership and contact details are current and consistent with each other.

Old pages are a common cause of the error, even when the main pages are correct. A news post announcing a service you no longer offer, a team page that still lists a former director or a location page for a closed office can all be cited as current, so update, redirect or remove them rather than leaving them live.

Check that ChatGPT can reach the website at all. OpenAI’s overview of its crawlers explains that OAI-SearchBot is used to surface websites in ChatGPT search results and that websites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links. The same page says changes to robots.txt can take around 24 hours to take effect for search. It also lists GPTBot as a separate setting that only concerns the use of content for training.

Avoid publishing a page that repeats the wrong claim just to deny it. A clear statement of the correct facts gives the same information without adding another page that pairs your name with the error.

Pages that describe the business in clear and consistent terms also help the people who arrive from an AI answer. Our guide to entity SEO and visibility in LLMs covers how AI systems recognise an organisation as a distinct entity.

Use Structured Data to State the Facts Plainly

Structured data states the core facts about your organisation in a format machines can read, which removes guesswork about names, addresses and identifiers. Google’s documentation on organization structured data says adding it to the home page can help Google understand an organisation’s administrative details and disambiguate it in search results.

That documentation covers Google’s use of the markup, so treat it as a way of making the facts unambiguous rather than a guaranteed fix for ChatGPT. Include the legal name, the trading name, the registered address, the company number where relevant and links to your official profiles.

The sameAs property on schema.org exists for exactly this job. It points to reference pages that unambiguously indicate an organisation’s identity, such as its Wikipedia page, Wikidata entry or official website, which is useful where another business has a similar name.

Markup only helps when it is valid and matches what the page says. Our guide to fixing structured data errors covers the common faults and how to test for them.

Correct Directory Listings and Official Records

Directory listings and official records are the next place to correct, since they describe your business on websites you do not control and can be cited alongside or instead of your own pages. Start with the records that carry legal or official weight, then work outwards to industry directories and review platforms.

Use one version of the business name, address and phone number everywhere. Small differences between listings, such as an old suite number or a former trading name, give any system reading them two versions of the truth to choose between.

Approach Wikipedia and Wikidata With Care

Wikipedia and Wikidata can be corrected, but only through their community processes and never by rewriting an entry about your own business. Wikipedia’s guideline on conflict of interest editing strongly discourages editing articles about yourself, your employer or your clients and requires paid editors to disclose who is paying them. It asks editors with a conflict of interest to propose changes on the article’s talk page instead.

A good request on the talk page names the error, gives the correct fact and links to an independent source that supports it. The same guideline lists heavy use of primary sources and sources that are not independent as a common problem with conflict of interest edits, so a correction backed by a Companies House record or a trade publication stands a better chance than one backed only by your own website.

Wikidata holds structured facts such as founding dates, headquarters and official websites in a form other services can reuse. Its notability policy accepts an item when it has a valid link to a Wikimedia page, describes a clearly identifiable entity that serious and publicly available references can describe or meets a structural need, so a small business without that coverage may not qualify for an entry.

Where an entry exists and holds an outdated fact, correct it with a reference and an edit summary that explains the change. Where the facts are disputed, raise them on the item’s discussion page rather than editing repeatedly.

Report the Wrong Answer to OpenAI

OpenAI offers reporting routes for responses, though none of them is an edit button for business facts and none promises a change. Its guide to reporting content in ChatGPT says conversations, shared links and responses that may break its Terms of Use or applicable laws can be reported through a content reporting webform or from the message itself.

The same guide says reported domains and other content may be reviewed by OpenAI’s Model Quality team, which can apply filters or other measures that stop ChatGPT relying on unreliable sources in future responses. That makes a report most useful when the wrong answer cites a specific inaccurate page, because the report can name the source as well as the response.

Where the answer amounts to a damaging false statement about your business, take legal advice before relying on any reporting route. A report records the problem with OpenAI, but it does not decide whether the statement is unlawful.

Use UK GDPR Rights When the Error Concerns a Person

UK GDPR rights apply when the wrong information is about an identifiable person, not when it is only about the company. The ICO’s guide to what personal information is says information about companies is not personal data, but information about sole traders, employees, partners and company directors can be when it relates to them as individuals.

That distinction matters for a small business. A wrong claim about a limited company’s services falls outside data protection law, while a false statement about a named director or a wrong home address for a sole trader may fall inside it.

The ICO explains the right to get your data corrected, which lets a person challenge the accuracy of personal data an organisation holds and ask for it to be corrected or deleted. OpenAI’s help article on personal data removal from ChatGPT describes a request through its Privacy Portal to stop personal information appearing in ChatGPT responses where it is inaccurate, excessive, irrelevant or no longer appropriate.

The request needs specific examples, such as links to chats and the URLs cited in them, along with proof of identity. If the response does not resolve the problem, OpenAI’s article confirms the right to complain to a supervisory authority, which in the UK is the ICO.

Wrong personal data on another organisation’s website can be raised with that website’s owner under the same right. Correcting it there removes a source that ChatGPT search could otherwise keep citing.

Keep Checking Without Expecting a Quick Fix

Corrections reach ChatGPT unevenly, so check the same prompts on a regular schedule rather than waiting for one decisive change. Answers that rely on live search can change once the cited pages are updated, while answers drawn from training data may persist until later models are trained on newer information.

Keep the log from the first step and add each correction with its date. Comparing new answers against that log shows which fixes made a difference and which sources still need work.

Wrong facts are one part of a wider question about how AI tools describe your organisation. Our guide to tracking and improving brand visibility in AI search covers monitoring across platforms, while the guide to AI brand mentions in generative AI looks at how brands earn mentions in the first place.

None of these steps can force ChatGPT to change an answer. Anyone who promises a guaranteed correction is overstating what is possible, so judge progress by the dated record rather than by a single response. Fixing every source you control and reporting the rest gives the accurate version of your business the strongest chance of being the one ChatGPT repeats.