AI tools returned at least one false fact on 64% of major UK retailers in a vendor-led test, according to viewable figures shared with Retail Focus. The most common mistake was placing businesses in the wrong zip code.
These tests reveal something that many marketing teams don’t track. When AI systems like ChatGPT or Google’s AI Mode answer questions about one of your locations, there’s no way to know what they say about your business unless you check manually. They might confidently state that a business is closed, list services they don’t offer, or give a wrong address.
We covered how to gain visibility in AI search, including Dan Taylor’s work on AI prompt tracking. This piece covers the other half of the problem. It looks at what AI is already telling people about your locations and how to find out before a customer does.
From classified to described
Traditional local search gives customers several options to compare, such as a map pack, reviews, your website or a competitor’s website. Then, customers make their choice. AI search combines this process into a single synthesized response that fully describes a business before the customer even visits the company’s website or profile.
Being described is not the same as being classified; Ratings may change and descriptions may not always be accurate. Also, this visibility varies depending on the type of query.
Whitespark Analysis found that AI previews appear in 15% of direct searches with local intent, such as (Phoenix personal injury lawyers). For informational questions, they appear 92% of the time, and for hybrid questions, like whether or not to hire a lawyer after an accident, they appear 97% of the time. In the same study, local packs appeared in 93% of direct searches with local intent. Overall, AI insights were more common for informative and hybrid questions than for direct searches with local intent.
SOCi Local Visibility Index analyzed over 350,000 locations and found that ChatGPT recommended 1.2%. That’s a much lower rate than the same brands’ appearances in Google’s local 3-pack, according to SOCi’s own measurements. BrightLocal Local Consumer Rating Survey Found 45% of consumers now use ChatGPT or similar tools for local business recommendations, up from 6% the previous year.
What are the wrong answers?
Searchable released two sets of test results to the trade press this month. In tests involving 165 London businesses, Searchable tested ChatGPT, Gemini and Perplexity with 13,365 questions about services, contact details, size and founding dates, then compared answers to Companies House records and official profiles.
According to CMOTech93% of these companies had at least one fundamental fact wrong or missing. Half of small businesses received at least one false fact, compared to 32% of large businesses. A second test looked at UK retailers with more than 72,000 questions. Focus on retail indicated that 1 in 16 responses were incorrect. ZIP code errors occurred at a rate of one in ten errors, even when prompts specified the city.
Searchable co-founder Chris Donnelly told Retail Focus:
“For a small brick-and-mortar retailer, if its online visibility is primarily focused on its website and a Google Business listing, this represents a relatively thin trail of information from which AI systems can learn and represent in their responses.”
The error categories match what business owners describe in Google Support Forumsreporting that they see incorrect details presented with confidence, and some say the errors hurt their business.
The blind spot
Traditional search leaves a trail of data behind. Impressions and clicks are tracked in Search Console; rankings fluctuate between trackers; and any drop often gives rise to an investigation. However, there is no reporting to alert businesses if an AI response incorrectly lists your location’s hours of operation or claims your business is closed.
Traditional search also doesn’t alert you to how you are represented, but it doesn’t have to. It directs customers to sources they can view and evaluate for themselves, like your listing, your site, and reviews with dates attached. The AI consolidates these sources into a single written response, and the client reads it as fact. When it’s wrong, there’s no ranking position to monitor or traffic drop to trace, because the error lies in the text that was never shown to you.
Major consumer AI tools generally don’t offer location-level alerts to businesses when their information is wrong. Google’s documentation states that AI responses may include errors, and individual AI previews have a comments link. This feedback is used to correct errors but does not constitute a proactive monitoring system. This only works if someone notices the error first.
5 systems, not 1
The testing process can get complicated on different AI research platforms because they don’t work the same way. AI previews now appear in regular Google search results. AI Mode is a conversational experience right in Google Search. Gemini acts as Google’s separate AI assistant. Meanwhile, ChatGPT and Perplexity are separate products offered by OpenAI and Perplexity AI.
Each of these tools processes, selects and combines information differently, so only one question can receive different answers depending on the platform you use. This variation is also found in Searchable retail dataPerplexity was found to give inaccurate answers 10% of the time, compared to 5% for Gemini and 4% for ChatGPT.
Testing one system doesn’t tell you much about others. A location that is verified in AI previews may still be incorrectly described in ChatGPT.
How to test what customers see
Start by thinking about questions your customers might have, like hours, services, or if a location is good. Write these questions down in a standard list so that each location is checked the same way. Run questions through AI Previews and AI Mode in Google, as well as ChatGPT, Gemini, and Perplexity. Where possible, test without saved conversation context or personalization, and keep track of the prompts and responses you get.
Since the answers may vary each time, ask the same questions several times. Organize what you find into categories such as factual errors, missing information, or perception issues. Treat factual errors and omissions separately. The sentiment and order of recommendations are reputational issues, not simple fixes. Pay special attention to correcting errors that might prevent someone from visiting you, such as incorrect hours or a branch being marked as closed.
When sources are cited, verify that their information is accurate. Correct the information you control or request corrections if necessary. Continue to check back regularly, as correcting a source today does not automatically update the AI responses. Use a consistent list of questions and a journal to make the process easier, especially if you have multiple locations. Change the check frequency based on the number of locations and how often the details change.
What to do when the AI makes a mistake
Finding the error is the first step. Fixing this requires carefully adjusting the inputs these systems depend on, and then monitoring whether responses improve. Remember, every adjustment brings you closer to better, more reliable results.
If an answer is wrong about your hours, confirm that your Google Business Profile, website, and your own location pages are consistent with each other and with reality. The conflicting details between these are one of the clearest reasons why an answer is wrong. The same name, address and telephone number digital consistency which Local SEO has always required is the foundation, and it becomes more complex once you run multiple locations.
After fixing the sources you control, look at the sources you don’t own. When a response cites your company, open it and check its date and what it actually says. This might involve pulling information from an old directory page, a review listing, or a page about a business with the same name. Addressing these issues will likely require direct outreach.
This is also where the size gap Donnelly points out starts to make sense. Smaller companies generated more errors in Searchable’s tests, and he read that a thin trace of third-party information made these systems less effective. Our own coverage what correlates with ChatGPT quotes comes to the same conclusion. A fuller, more consistent presence in the places these systems read gives them better things to describe you.
If you find that accuracy isn’t the issue and you’re more interested in getting recommended, it’s the visibility side. Dan Taylor wrote for Search Engine Journal about track how AI represents your brand over time And how discovery changes as search becomes personalized to each user. This is where we need to start to answer the question “How can we win here”.
Looking to the future
Monitoring tools for AI responses are still in their early stages. Most can tell you if you are mentioned and how often, which is not the same as telling you if the mention is correct.
Checking what an answer actually says about a place always comes down to reading it yourself. As these tools develop, this may change, but for now auditing is manual, and when you discover something wrong, the first step is to optimize the sources you control.
More resources:
Featured Image: Natalia Kosarevich/Shutterstock
