AI is telling people the wrong things about your business
If a customer asks ChatGPT where your business is, there is a reasonable chance it sends them somewhere else.
That is not a hypothetical. The AI visibility platform Searchable put more than 72,000 questions about UK high street retailers to ChatGPT, Google’s Gemini and Perplexity, then graded every answer against the businesses’ own verified information. The findings, published in Retail Focus, showed the chatbots gave an answer 98% of the time. One answer in sixteen was false, and roughly two in three businesses had at least one false fact returned about them.
The most common error was the postcode, at a rate of about one in ten, and that held even when the prompt clearly stated which town the business was in. In 15% of those cases the address given was more than twenty miles from the real one.
It gets worse for websites. Around one in fifteen website recommendations pointed to a dead link, a lookalike site, or a completely different business. About one in forty answers attributed a shop to the wrong brand entirely.
Worth noting that Searchable is a commercial AI visibility platform publishing its own research, so read the numbers with that in mind. But the pattern it describes matches what we see when we test client businesses by hand, and the scale of the sample makes it hard to dismiss.
Small businesses come off worst
The findings were not spread evenly. Smaller businesses were considerably more likely to have false information returned about them than larger brands.
The explanation is straightforward, and it is the single most useful thing in the whole study. Large companies have a big pool of third-party sources online that all say the same accurate things about them. Directories, press coverage, review sites, trade bodies, Companies House filings, Wikipedia. When an AI model assembles an answer, it has plenty of consistent material to work from.
A kitchen fitter in Blaydon or a care provider in Sunderland has a much thinner footprint. Maybe a Google Business Profile, a website, two or three directory listings of varying accuracy, one of which still has the phone number from before they moved. The model fills the gaps by inference, and inference is where the wrong postcodes come from.
This is the same underlying problem we described in our piece on how AI is changing search rankings, seen from the other end. It is not only about whether AI recommends you, which we covered in can AI find your business. It is about whether the thing it says when it does recommend you is actually true.
The platforms are not equally reliable
Searchable’s testing found meaningful differences between the three tools. Perplexity returned inaccurate answers in around 10% of cases, compared with about 5% for Gemini and 4% for ChatGPT.
That matters for how you prioritise. If you only have time to check one platform, check the one your customers are most likely to use, but do not assume a clean result on ChatGPT means you are clean everywhere. Regional cuts of the same research tell a similar story, and the Leeds breakdown found the same postcode problem showing up at a rate of one in fourteen answers.
Why this costs you money
There is a temptation to file this under interesting but not urgent. It is not.
A wrong postcode is a customer who drives to the wrong place and gives up. A dead link is an enquiry that never lands. A phone number from two offices ago is a call that rings out.
None of these show up in your analytics. There is no bounce to measure, no form abandonment, no drop in sessions. The enquiry simply never happens, and nothing in your reporting will ever tell you why. It is the same invisible failure mode that makes a website quietly stopping generating enquiries so hard to catch, and the same reason we keep arguing that your website is either your best or your worst salesperson.
There is a trust cost on top of that. Consumer research quoted alongside the Searchable findings suggests a majority of shoppers lose confidence in a brand when AI gives them incorrect information about it. The customer does not know the model made it up. They assume you cannot keep your own details straight.
How to test your own business
You can do this yourself in about twenty minutes. The method matters, so be systematic rather than asking a few casual questions. Search Engine Journal has a good write-up of why manual checking is currently the only option, since none of the platforms notify you when they get something wrong.
Pick six questions and ask each of them on ChatGPT, Gemini and Perplexity. Use the same wording every time and include your town or city in the prompt.
- Where is [business name] in [town] based? Give the full address and postcode.
- What are [business name]’s opening hours?
- What services does [business name] offer?
- What is the phone number for [business name]?
- What is [business name]’s website?
- Is [business name] still trading?
Then grade every answer against your own verified information. Not what you think is out there, what is actually correct. Start with your Companies House record, then your current Google Business Profile, then your own site.
Score three things for each answer. Did it answer at all, was it factually correct, and how did it frame you. That third one is easy to skip and often the most revealing. Being described as a general builder when you are a specialist heritage restorer is not a false fact, but it is still costing you the right enquiries.
Do this quarterly and keep the results in a spreadsheet. What you want is not a single snapshot but a trend, because these answers move as the underlying sources move.
Fixing what you find
Where the answers are wrong, the fix is almost never on your own website alone. It is in the consistency of everything that describes you.
Start with the basics that feed everything else. Your Google Business Profile, your Companies House record, your major directory listings, your social profiles. Make the name, address and phone number identical across all of them, character for character. Old listings from a previous address are a common culprit and they are worth hunting down and correcting rather than ignoring.
Then look at your own site. Make the facts explicit and machine readable rather than implied by design. A trading address sitting inside an image or tucked into a footer graphic is invisible. Proper structured data, a clear contact page, service pages that state plainly what you do and where you do it, and an about page that says when you were founded and who runs the business. This is one of the reasons we build websites and digital platforms around structure rather than decoration.
Then work on the third-party footprint, because that is what large brands have and you do not. Trade body listings, local press, supplier and partner pages, case studies published somewhere other than your own domain. Each one is another consistent source for a model to draw on, and it is a good part of what our AI, automation and insight work actually involves.
None of this is exotic. It is mostly the discipline of saying the same true thing in enough places that a model has no room to guess. If you want us to run the test for you and tell you what is coming back, get in touch and we will send you the results.
