What happens when AI tells customers the wrong price or opening hours?

Where does an assistant get your opening hours?

From wherever it can, and not from you. No assistant rings up to check. It reads your Google listing, your website, and whatever directories and review sites carry your name, then produces one confident sentence out of the lot.

Google is the only one of these companies that writes down how this works. Its documentation for site owners, updated on 10 December 2025, says a page has to be indexed and eligible to appear in ordinary Google search with a snippet, which means it already shows up in normal results and Google is allowed to display a line or two of its text.3 Its newer guide for generative answers, updated on 10 July 2026, tells local businesses to keep a Google Business Profile, and says plainly that no special hidden code is required.4

The part most owners have never been told is that your Google listing is not only yours. Google Maps lets people suggest edits to a business's name, address, opening hours, phone number, website and whether it has closed down. Google says it reviews all the edits people make, and that changes may take time to appear on the map.2 A passer-by who finds your door shut on a Tuesday can report you as closed, and the first you hear of it is the phone going quiet.

How the error gets in

The route from a change you made to an answer a customer reads

01

You change something

New winter hours, a price rise, a second number for the workshop.

02

You update one place

Usually the website, or the Google listing. Rarely both, almost never the directories.

03

Everything else keeps saying the old thing

Including the hidden code on your own site, which nobody remembers is there.

04

An assistant picks one

It answers in one sentence, with no sign that anything was in dispute.

How often do these systems get things wrong?

Often enough that you should assume it will happen to you. Nobody has published a study of how accurately AI assistants describe local businesses, so anyone quoting you a figure for that is making it up. What does exist is a large, careful study of how accurately they describe the news, which is the closest measurable thing.

The European Broadcasting Union and the BBC had journalists at 22 broadcasters in 18 countries, working in 14 languages, grade 3,113 answers from ChatGPT, Copilot, Gemini and Perplexity. The answers were generated in May and June 2025 and the report came out in October 2025. 45% of the answers had at least one significant problem. Sourcing was the biggest cause at 31%, and 20% had accuracy problems, which the authors describe as including invented details and out-of-date information.1

The assistants were not equally bad. Gemini had a significant problem in 76% of its answers, more than double the next one, driven mostly by how it handled sources.1

EBU and BBC, October 2025

Share of answers with at least one significant problem

Gemini 76%

Copilot 37%

ChatGPT 36%

Perplexity 30%

News is harder to summarise than opening hours, so the honest read is not that 45% of what an assistant says about your salon is wrong. It is that these systems produce a confident single answer whether or not the underlying sources agree, and that out-of-date information is one of the recognised failure modes rather than a freak event.

Does a wrong detail actually cost you a customer?

Yes, and this part has been measured, though not recently and not here. BrightLocal surveyed 1,138 consumers in the United States in September 2023, before AI answers were a normal part of searching. 62% said they would avoid using a business if they found incorrect information about it online, and 26% said they had turned up at a business too early or too late because the opening hours were wrong, at least once a month.5

Two things about that survey are worth saying out loud. It is American, and there is no UK equivalent we could find. It is also three years old, which in this context cuts the wrong way: it measured a world where a wrong opening time sat in a listing the customer could see and cross-check against three other results on the page. An assistant removes the other results.

Do your own facts even agree with each other?

For a lot of businesses, no. Since nobody had checked this for UK service businesses, we did it ourselves on 26 August 2026. We took 800 dentists, hairdressers, beauty salons, GP practices, vets and garages that record a website in OpenStreetMap across London, Manchester, Birmingham, Glasgow and Cardiff, and read each site the way an automated visitor would: the homepage, plus up to two pages linked from it that looked like contact, hours or pricing pages.7

194 of the 800 sites did not load for us at all. Of the 606 that did, 322 stated opening hours in ordinary page text, and 259, or 42.7%, gave us no opening hours anywhere we looked. A quarter of the sites showed more than one phone number across the pages we read, which leaves an assistant to guess which one to give out.

Frontpage AI research

What 606 working UK service business websites told an automated reader

Counts and percentages for five checks run against 606 reachable UK service business websites on 26 August 2026.

What we looked for	Sites	Share
Opening hours written in ordinary page text	322	53.1%
No opening hours anywhere we looked	259	42.7%
More than one phone number across the pages read	153	25.2%
No phone number we could find at all	120	19.8%
Opening hours also written in hidden code for machines	71	11.7%

The interesting part was the small group that stated their hours twice: once in plain text and once in the hidden code some websites carry to spell out facts for machines. Only 71 sites had that code at all. On 46 of them we could compare both versions for the same day of the week, and our automated check flagged 23 as contradicting themselves.

We then read all 23 by hand, and 15 of them turned out to be our own reader being fooled rather than the business being wrong. A Christmas opening times notice, a table with the days in one column and the times in another, a page using a long dash instead of a hyphen: each one produced a mismatch that was not really there. Eight of the 46 genuinely contradicted themselves.

What survived is unglamorous and completely real. A Manchester dental practice whose page says it closes at seven on a Tuesday while its code says eight. A London dentist advertising eight in the evening on a Wednesday, with five in the afternoon written underneath in the code. A Cardiff barber opening at nine on the page and ten in the code. A London hair salon whose own page says Tuesday, 11 pm to 5 pm. One Fulham barbershop lists its phone number, both on the page and in the code, as +44 20 1234 5678.

Our own 15 false alarms are the more useful finding. We wrote that reader specifically to pull opening hours off a website, and column layouts and holiday notices still defeated it. An assistant reading the same page has the same problem, with less patience and no way to ask.

Who is responsible when a machine gets it wrong?

Nobody has settled this for the case that affects you. There is one well-known decision about a machine giving out wrong information, and it is worth understanding precisely because it does not cover your situation.

In February 2024, the British Columbia Civil Resolution Tribunal decided Moffatt v Air Canada. A customer had been told by Air Canada's own chatbot that he could apply for a bereavement fare after flying, which was not the airline's policy. Air Canada argued that the chatbot was responsible for its own actions. The tribunal disagreed, saying that while a chatbot has an interactive component, it is still just part of Air Canada's website, and that it should be obvious a company is responsible for all the information on its own website.6

That was a Canadian tribunal, and more importantly it was a bot the company ran itself. If ChatGPT tells someone your salon opens at nine when you open at ten, no equivalent case has tested who carries that, in the UK or anywhere we could find. Practically, the answer today is that you do: you lose the customer, and the correction is yours to make.

What to actually do about it

Make one version of your facts, put it where machines look, and check it on a schedule. In order of how much difference it makes:

Write your opening hours in plain text on your website, on a page anybody can reach in one click from the homepage, with the days spelled out. Not in an image of a sign, not only in a booking widget that loads after the page does. If you only do one thing, do this one, because 42.7% of the sites we checked have not.

Check your Google Business Profile monthly, and after any change. You are looking for edits you did not make, since other people can suggest them. Whoever set the profile up can show you where the notifications land, and it takes two minutes.

Ask the assistants directly, the way a customer would. Open ChatGPT, Google's AI Mode, Gemini and Copilot, and ask what time you open on a Saturday and what you charge. It costs nothing and it is the only way you will find out. Our guide to checking what AI says about your business walks through it.

If your site carries hidden code with your hours in it, make somebody own it. Two versions that disagree are worse than one version, because nothing on your site settles the question. Whoever built the site can tell you in a minute whether it is there, and it is usually a setting in the same place your address lives.

All of that is a morning's work and none of it needs an agency. What people pay us for is that it stays true after the morning ends. Frontpage AI is a UK studio that builds and manages websites for independent service businesses, structured so that Google and AI answer engines can read, trust and name them. We work only with UK service businesses, such as trades, clinics, salons and instructors, and charge from £88 per month on a 30-day rolling contract with no setup fee. Part of that is keeping one set of facts about you in every place a machine reads, which is a maintenance job rather than a clever one. The way we approach AI search page sets out the rest.

What this evidence does not tell you

It does not tell you how often assistants are wrong about UK businesses. The 45% figure is about news, graded by journalists, in May and June 2025. We have used it because it is the most rigorous measurement of confident AI error that exists, not because it transfers to your opening hours.

The consumer figures are American and from September 2023. A British survey might find people more forgiving or less, and nobody has run one that we could find.

Our own check has real limits too. It is one morning, five cities and six trades, drawn from businesses that happen to be recorded in OpenStreetMap, which skews towards places somebody thought worth mapping. It measures what an automated reader could see in a few seconds, so a site that assembles its pages in the visitor's browser may have hours we never saw. That is a genuine limitation of our method, and also the exact problem being described.

And the 8 out of 46 figure describes only the small group of businesses that had bothered to state their hours twice. That is not a random sample of anything. If it means something, it is that doing the technical half without keeping it in step with the ordinary half creates a new way to be wrong, which is worth knowing before somebody sells you the technical half.