Written By John B. CEO at Danu Agency AI
Published: 29/09/2026
Quick answer: Usually because it is repeating a source that is wrong or out of date, answering from training memory, or guessing to fill a gap your own website left open. In a test of 600 small businesses, 20% of ChatGPT’s answers included details the business does not publish; where websites stated their details clearly, ChatGPT was accurate 89% of the time.
There is a peculiar indignity in being misdescribed by a machine. A prospect asks ChatGPT about your business, and it answers with perfect composure: the wrong opening hours, a service you discontinued two years ago, an insurance plan you never accepted. The prospect has no reason to doubt it. You have no way of knowing it happened.
In an earlier guide we asked why chatbots fail to recommend a business at all. This is the more insidious problem. Absence costs you a conversation; a confident error can cost you the customer before the conversation begins. The good news is that most errors are neither random nor permanent. They have causes, and the causes leave a trail.
How often AI gets businesses wrong
The research is young, and much of it comes from firms that sell the remedy, which is reason to read it with care rather than to dismiss it. Three recent studies, and one consumer survey, point the same way.
In September 2026, the marketing firm Herringbone put 1,200 questions about 600 small businesses (dental practices, personal injury firms and home-service companies) to ChatGPT and to Google’s AI Overviews, then checked every answer by hand against each business’s own website. 20% of ChatGPT’s answers and 27% of Google’s contained information the business does not publish. Because the sample was limited to those three trades, treat the percentages as directional rather than universal. A detail the business does not publish is not always wrong, but it is always something the business never said.
Searchable, an AI visibility vendor, reported in July 2026 that across more than 72,000 questions about UK high-street retailers, 64% of retailers had at least one false fact in AI answers, and about one answer in 16 was wrong. Perplexity had the highest error rate, at 10%, against 5% for Gemini and 4% for ChatGPT. The most common error was the postcode. In a separate test of 165 London businesses, 93% had at least one basic fact wrong or missing, and smaller firms fared worse than larger ones: 50% had at least one false fact, against 32% of larger companies. These figures come from Searchable as reported by Search Engine Journal and Retail Tech Innovation Hub. We could not access the original report, so treat the exact percentages as indicative.
Customers notice. In SOCi’s 2026 survey of more than 1,000 US consumers, 67% of AI users said they had been given wrong information about a local business at least once, and 30% said it had caused them a real inconvenience. When AI recommends a business, 81% take some kind of verification step before getting in touch.
The problem is not confined to business listings. In October 2025, the European Broadcasting Union and 22 public-service broadcasters evaluated more than 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity about the news. Almost half contained at least one significant issue, and about a fifth contained major accuracy problems, including hallucinated or outdated information.
Why the machine gets you wrong
It guessed
The first cause is built into the technology. In September 2025, OpenAI published research explaining why language models produce confident falsehoods. Models learn by predicting the next word across vast amounts of text, and while regular patterns like grammar become reliable, arbitrary facts that appear rarely cannot be inferred from patterns at all. OpenAI’s own example is a pet’s birthday. By analogy, your opening hours are the same kind of fact to a model: specific, arbitrary and easy to get wrong. OpenAI adds that the way models are evaluated rewards a guess over an honest “I don’t know.”
It remembered instead of looking
ChatGPT does not search the web for every question. When Slate put 1,000 B2B software questions to it in August 2026, ChatGPT showed a source link on 90.7% of buying questions but only 8.1% of research questions; Slate’s reading is that it answers the latter from memory. Whatever a model absorbed in training reflects the web as it was, not as it is. If your business has changed since, the memory has not.
ChatGPT doesn't always look before it answers
Share of ChatGPT answers that showed a source link
Share of ChatGPT answers containing at least one source link, across 1,000 B2B software questions with web search enabled, captured 3 August 2026. Source: Slate
It read someone else’s page about you
When ChatGPT does search, it reads whatever the web offers, and much of that is not yours. Herringbone traced many errors to third-party directories: insurance directories listing plans that dental practices had never claimed to accept. Searchable’s co-founder suggested that smaller businesses suffer because they leave a thinner trail, often just a website and a Google listing, which gives AI less to work from.
It believed your own contradictions
Some errors are self-inflicted. Herringbone found that 46 of 59 businesses whose websites listed no hours had their Google Business Profile set to “open 24 hours”; the AI simply repeated what the businesses had told Google. Others published one set of hours on the contact page, another in the footer and a third in their structured data, leaving the engine to choose. Phrases such as “available 24/7”, usually describing an answering service, were read as opening hours.
Do not let “24/7” mean two things. If your phones are answered around the clock but your office is not open, say so separately, for example: “Phones answered 24/7. Office hours: Monday to Friday, 9am to 5pm.”
It confused you with someone else
Names are less unique than owners assume. When Zarla checked whether ChatGPT and Google’s AI Mode named 1,851 local businesses, 28.31% of them received a name-alike mention that could not be confirmed as theirs: a franchise branch, a near-identical name, a different firm in the same city. A mention that might be you is not reliably you, and the details attached to it may belong to someone else.
What you can correct, and what you cannot
You cannot edit ChatGPT. There is no form through which a business submits its correct details, and OpenAI has retired its web form for model-behavior feedback in favour of in-product channels. What you can do falls into two categories: tell the platforms and fix the sources they read.
Tell the platforms
ChatGPT: click the thumbs-down icon beneath the wrong answer and explain the error.
Google AI Overviews: click the thumbs-down icon, select “Report a problem” and choose the category that fits. Google states plainly that AI Overviews can and will make mistakes.
Perplexity: use the flag icon beneath the answer, or email support@perplexity.ai with the URL of the query, a description of the error and the correct information.
If the error concerns you as a person, as it might for a sole practitioner, OpenAI’s privacy policy lists a right to request correction of personal information, by email to dsar@openai.com.
Feedback is worth giving, but none of these platforms promises that a single report will change future answers. The more durable remedy lies upstream.
Diagnose before you change anything
A wrong answer is not always a persistent problem. Before editing a page or a listing, capture the exact question, the engine, the date, the full answer and every citation. Then ask the same question again in a fresh conversation, using two or three natural phrasings.
If the error appears only once, it may be a one-off. If it repeats, treat it as a real accuracy problem. If the engine shows a citation, open it: the fastest repair is usually to correct the page it actually used. OpenAI says ChatGPT search sometimes partners with other search providers, and may share your general location with trusted third-party providers that help supply local information, such as nearby business listings.
Test with search on and off
Where the platform allows it, ask the question once with web search on and once without. An error that disappears when the engine searches points to a source you can correct: your site, your Google Business Profile, a listing or another page it read. An error that survives without search is more likely to come from the model’s training memory, which is slower to change and cannot be edited directly.
This test does not prove where an error began, but it tells you where to look first, and it stops you rewriting your website to fix an error that came from a stale directory.
Fix the sources, starting with the ones you control
Your own website. State your hours, services, location and contact details plainly on one clear page, and make sure the same facts appear in your footer, your structured data and everywhere else. The gap is striking: where a website stated its details clearly, Herringbone found ChatGPT accurate 89% of the time and Google 84%; where it was vague or silent, accuracy fell to 73% and 63%, and Google filled the gap with invented specifics, without naming a source, 80% of the time. Our guide The Free Five covers how to make your business details consistent.
Say it clearly, and AI gets it right more often
ChatGPT: answers accurate
Google AI Overviews: answers accurate
Share of answers that matched the business's own website, for businesses whose sites stated their details clearly versus vaguely or not at all. 600 small businesses, 1,200 questions, checked by hand; estimates have a margin of a few points. September 2026. Source: Herringbone
Your Google Business Profile. Check your hours, services, categories, website link and any “open 24 hours” setting. Herringbone found that most businesses with no hours on their website had a Google profile set to open 24 hours, which the engines repeated. Our guide How Do I Get Gemini to Recommend My Business? walks through the profile step by step.
Bing Places. ChatGPT search has drawn on Bing, as we set out in What Search Engine Does ChatGPT Use? Also, Bing Places is where you supply Bing with your business details. Keep it identical to your website and your Google profile.
Yelp. Claim and correct your profile. Axios reported in July 2026 that Yelp is licensing its reviews, ratings, photos and business details to OpenAI for use in ChatGPT’s answers.
Industry and local directories. Search for your business name alongside old addresses, former phone numbers and discontinued services. Herringbone found third-party directories to be a recurring source of false details, including insurance plans that dental practices had never claimed to accept. Perplexity is often the easiest place to trace a claim, because it shows the pages behind each answer, even though it gave the most false answers in Searchable’s test.
Correct the source where the error actually appears, then check again.
What no one can tell you yet
How long a correction takes to reach an AI answer is not established. We found no published study that measures it, and the answer almost certainly differs between engines, and between answers that come from search and answers that come from memory. Treat any promise of a fixed timescale with suspicion. Check again after a few weeks and keep checking.
The Danu Agency AI Accuracy Check
Ask what it knows: in ChatGPT, Claude, Gemini and Perplexity, ask "What do you know about [your business name] in [your city]?"
Repeat the question: ask it again in a new conversation, in slightly different words, and note whether the same wrong fact returns.
Ask what customers ask, your hours, prices, services and location, and compare every answer with your own website, word for word.
Trace each error to its source: open the cited pages. Is the wrong detail on your site, your Google Business Profile, Bing Places, Yelp, an industry directory, or nowhere at all?
Fix the source, then report the answer through each platform's feedback channel.
Repeat monthly. Answers shift from one day to the next, so a single check is a snapshot, not a verdict.
Is AI getting your details right? Our free AI Accuracy Check asks ChatGPT and Gemini about your services and prices, your location and contact details, and who runs your business, then checks every answer against the facts you give us. Results land in your inbox.
Frequently asked questions
Why does ChatGPT have wrong information about my business?
Usually because it is repeating an outdated or inaccurate source, such as a directory or an old profile, or answering from memory instead of searching. Where your own website is vague, it may fill the gap with a guess.
Can I contact OpenAI to correct ChatGPT?
Not as a business. OpenAI's channel for feedback on answers is the thumbs-down icon in ChatGPT. Individuals can request correction of personal information under OpenAI's privacy policy by emailing dsar@openai.com.
How do I report a wrong AI Overview on Google?
Click the thumbs-down icon under the AI Overview, select "Report a problem", choose a category and describe the error.
Will fixing my website fix what ChatGPT says?
It helps most when ChatGPT searches the web, because a clear, consistent website gives it less to guess. Answers drawn from training memory may take longer to change, and no published study yet measures how long.
Why does AI mix my business up with another one?
Similar names, franchise branches and businesses in the same area are easily confused. In Zarla's study of 1,851 local businesses, 28.31% received a name-alike mention that could not be confirmed as theirs.
Sources
OpenAI, Why language models hallucinate (September 2025): https://openai.com/index/why-language-models-hallucinate/
Herringbone, Do ChatGPT & Google AI Get Your Business Details Right? (September 2026): https://herringbone.com/does-ai-get-your-business-details-right/
Search Engine Journal, AI Answers About Your Locations Are Often Wrong (July 2026): https://www.searchenginejournal.com/ai-answers-about-your-locations-are-often-wrong-check-before-customers-do/582205/
European Broadcasting Union, News Integrity in AI Assistants (October 2025): https://www.ebu.ch/research/open/report/news-integrity-in-ai-assistants
Slate, How to Get Cited by ChatGPT, Perplexity, Claude and Gemini (August 2026): https://slatehq.com/data-studies/how-to-get-cited-by-chatgpt-perplexity-claude-and-gemini
Zarla, AI Visibility Index (August 2026): https://www.zarla.com/guides/ai-visibility-index
OpenAI, Model behavior feedback: https://openai.com/form/model-behavior-feedback
OpenAI, Privacy policy: https://openai.com/policies/privacy-policy/
Google Search Help, AI Overviews: https://support.google.com/websearch/answer/14901683
Perplexity Help Center, How can I report incorrect or inaccurate answers?: https://intercom.help/perplexity-ai/en/articles/10354902-how-can-i-report-incorrect-or-inaccurate-answers
Metavert, State of AI Search Citations: Q4 2026: https://metavert.io/state-of-ai-search-citations
Retail Tech Innovation Hub, Many UK high street retailers suffering from AI chatbot inaccuracies, Searchable research finds (July 2026): https://retailtechinnovationhub.com/home/2026/7/15/many-uk-high-street-retailers-are-suffering-from-ai-chatbot-inaccuracies-searchable-research-finds
