AI Answers About Your Locations Are Often Wrong – Check Before Customers Do

Jul 20, 2026 08:30 PM - 1 month ago 10183

AI devices returned astatine slightest 1 mendacious truth astir 64% of UK precocious thoroughfare retailers successful a vendor-run test, according to figures Searchable shared pinch Retail Focus. The astir communal correction was placing businesses successful the incorrect postcode.

These tests constituent to thing galore trading teams aren’t tracking. When AI systems specified arsenic ChatGPT aliases Google’s AI Mode reply questions astir 1 of your locations, there’s nary measurement to cognize what they’re saying astir your business unless you manually check. They could beryllium confidently claiming a business is closed, listing services they don’t offer, aliases giving a incorrect address.

We’ve covered really to gain visibility successful AI search, including Dan Taylor’s activity connected search AI prompts. This portion covers the different half of the problem. It looks astatine what AI is already telling group astir your locations and really to find retired earlier a customer does.

From Ranked To Described

Traditional section hunt gives customers aggregate options to compare, for illustration a representation pack, reviews, your website, aliases a competitor’s. Then, customers make their choice. AI hunt combines this process into a azygous synthesized reply that afloat describes a business earlier the customer moreover visits the website aliases business profile.

Being described isn’t the aforesaid arsenic being ranked; rankings tin change, and descriptions mightiness not ever beryllium accurate. Also, this visibility varies depending connected the type of query.

Whitespark’s analysis recovered that AI Overviews look successful 15% of nonstop local-intent searches, for illustration [personal wounded lawyers successful Phoenix]. For informational questions, they show up 92% of the time, and for hybrid questions, for illustration whether to prosecute a lawyer aft an accident, they look 97% of the time. In the aforesaid study, section packs showed up for 93% of nonstop local-intent searches. Overall, AI Overviews were much communal for informational and hybrid questions than for nonstop local-intent searches.

SOCi’s Local Visibility Index analyzed complete 350,000 locations and recovered ChatGPT recommended 1.2% of them. That’s a acold little complaint than the aforesaid brands’ appearances successful Google’s section 3-pack, by SOCi’s ain measurement. BrightLocal’s Local Consumer Review Survey recovered 45% of consumers now usage ChatGPT aliases akin tools for section business recommendations, up from 6% the twelvemonth before.

What The Answers Get Wrong

Searchable released 2 sets of trial results to waste and acquisition property this month. In tests involving 165 London businesses, Searchable tested ChatGPT, Gemini, and Perplexity pinch 13,365 questions astir services, interaction info, size, and founding dates, past compared answers to Companies House records and charismatic profiles.

According to CMOTech, 93% of these businesses had astatine slightest 1 basal truth that was incorrect aliases missing. Half of the smaller businesses received astatine slightest 1 mendacious fact, compared to 32% of larger companies. A 2nd trial examined UK precocious thoroughfare retailers pinch complete 72,000 questions. Retail Focus reported that 1 successful 16 answers was incorrect. Wrong postcode errors occurred astatine a complaint of 1 successful 10, moreover erstwhile prompts specified the town.

Searchable co-founder Chris Donnelly told Retail Focus:

“For a smaller bricks-and-mortar retailer, if their online visibility mostly centres astir its website and a Google Business listing, that’s a comparatively bladed way of accusation for AI systems to study from and to correspond successful their answers.”

The correction categories lucifer what business owners picture successful Google’s support forums, reporting that they spot incorrect specifications presented confidently, and immoderate opportunity the errors are damaging their businesses.

The Blind Spot

Traditional hunt leaves down a way of data. Impressions and clicks are tracked successful Search Console; rankings up and down crossed trackers; and immoderate driblet often leads to an investigation. However, there’s nary study that alerts businesses if an AI reply incorrectly states hours for your location aliases claims your business is closed.

Traditional hunt doesn’t alert you to really you’re represented either, but it doesn’t person to. It points customers to sources they tin spot and measurement for themselves, for illustration your listing, your site, and reviews pinch dates attached. AI collapses those sources into a azygous written answer, and the customer sounds it arsenic fact. When it’s wrong, there’s nary ranking position to watch and nary postulation dip to trace, because the correction lives successful matter you were ne'er shown.

Major user AI devices mostly don’t connection businesses an alert astatine the location level erstwhile their accusation is wrong. Google’s archiving states that AI responses whitethorn see mistakes, and individual AI Overviews transportation a feedback link. This feedback serves to correct mistakes but isn’t a proactive monitoring system. It only useful if personification notices the correction first.

5 Systems, Not 1

The testing process tin get analyzable crossed various AI hunt platforms because they don’t activity the aforesaid way. AI Overviews now look arsenic portion of regular Google hunt results, AI Mode is simply a conversational acquisition correct wrong Google Search. Gemini acts arsenic Google’s abstracted AI assistant. Meanwhile, ChatGPT and Perplexity are abstracted products offered by OpenAI and Perplexity AI.

Each of these devices processes, selects, and combines accusation differently, truthful a azygous mobility mightiness person different answers depending connected which level you’re using. This variety is besides seen successful Searchable’s unit data, Perplexity was recovered to springiness inaccurate answers successful 10% of cases, compared to 5% for Gemini and 4% for ChatGPT.

Testing 1 strategy tells you small astir the others. A location that checks retired successful AI Overviews tin still beryllium misdescribed successful ChatGPT.

How To Test What Customers See

Start by reasoning astir questions your customers mightiness have, for illustration hours, services, aliases whether a location is good. Write these questions down successful a modular database truthful each spot gets checked the aforesaid way. Run the questions done AI Overviews and AI Mode successful Google, positive ChatGPT, Gemini, and Perplexity. Where possible, trial without saved speech discourse aliases personalization, and support way of the prompts and answers you get.

Since answers tin alteration each time, inquire the aforesaid questions aggregate times. Organize what you find into categories specified arsenic actual errors, missing info, aliases cognition issues. Treat actual errors and omissions separately. Sentiment and proposal bid are estimation issues, not elemental corrections. Pay typical attraction to fixing mistakes that could forestall personification from visiting, for illustration incorrect hours aliases a branch reported arsenic closed.

When sources are cited, verify that their accusation is accurate. Correct the accusation you power aliases inquire for corrections erstwhile needed. Keep checking regularly because fixing a root coming doesn’t automatically update the AI answers. Use a accordant database of questions and a log to make the process easier, particularly if you person galore locations. Change really often you cheque based connected the number of locations and really often specifications change.

What To Do When AI Gets It Wrong

Finding the correction is the first step. Fixing it is astir cautiously adjusting the inputs these systems dangle on, past checking whether the answers improve. Remember, each accommodation gets you person to better, much dependable results.

If an reply sewage your hours wrong, corroborate that your Google Business Profile, your website, and your ain location pages work together pinch each different and pinch reality. Conflicting specifications crossed those are 1 of the clearest reasons an reply gets it wrong. The aforesaid name, address, and phone number consistency that local SEO has ever required is the foundation, and it gets more analyzable erstwhile you tally aggregate locations.

After correcting the sources you control, look astatine the sources you don’t own. When an reply cites your business, unfastened it and cheque its day and what it really says. It could beryllium pulling the accusation from an aged directory page, a reappraisal listing, aliases a page astir a likewise named business. Fixing these issues will apt impact immoderate nonstop outreach.

This is besides wherever the size spread Donnelly pointed to starts to make sense. Smaller businesses drew much errors successful Searchable’s tests, and his publication was that a bladed way of third-party accusation gives these systems little to activity from. Our ain sum of what correlates pinch ChatGPT citations reaches the aforesaid conclusion. A fuller, much accordant beingness crossed the places these systems publication gives them amended worldly to picture you with.

If you find accuracy isn’t the issue, and you’re much concerned pinch getting recommended astatine all, that’s the visibility side. Dan Taylor has written for MCP connected tracking really AI represents your marque complete time and how find changes arsenic hunt personalizes astir each user. Those are the places to commencement connected the “how do we triumph here” question.

Looking Ahead

Monitoring devices for AI answers are still successful their early stages. Most tin show you whether you’re being mentioned and really often, which isn’t the aforesaid arsenic telling you whether the mention is right.

Checking what an reply really says astir a location still comes down to reference it yourself. As these devices develop, that whitethorn change, but for now, the audit is manual, and erstwhile you find thing wrong, the first move is to optimize the sources you control.

More Resources: