AI Has the Wrong Information About Your Business. There Is No Edit Button.

Somewhere in the next fortnight, a few thousand businesses around the country will quietly swap their summer opening hours for winter ones. Two minutes on the website, job done, never thought about again.

That used to be the end of it. It is not any more, because your opening hours no longer live in one place. They live on your website, in your Google Business Profile, in a Bing listing you have very likely never claimed, in a few directories that scraped your details years ago, and inside four separate AI assistants that will recite whichever version they happened to pick up. Change the website and you have changed exactly one of them.

I want to be precise about the problem, because it usually gets described badly. It is not that AI does not know you exist. That is the visibility question and it is a different argument entirely. It is that AI does know you exist, states something wrong about you with complete confidence, and gives the person asking no reason whatsoever to doubt it.

There is no edit button, and that is the whole difficulty

Every other channel you deal with has an owner and a form. Wrong phone number on your website? Log in, change it, deploy. Wrong hours on your Google listing? There is a field for that. Wrong price in a printed ad? Ring the paper.

There is no equivalent for ChatGPT. You cannot log in to a business dashboard, find the record it holds for you, and correct it. No such record exists in the form you are imagining. What the assistant produces is generated fresh each time from two things: what the model absorbed during training, which was frozen at some point in the past, and whatever it retrieves from the live web at the moment somebody asks.

So the fix is indirect, and that is the part people find genuinely annoying. You do not correct the answer. You correct the sources the answer is assembled from, then wait for the machine to catch up.

Why it says it so confidently

Confidence is the dangerous bit. An assistant that said "I think they close around six, but check" would be almost harmless. They do not talk like that.

The largest study of this so far came from the European Broadcasting Union and the BBC, published in October 2025. Twenty two public service media organisations across eighteen countries, fourteen languages, more than three thousand responses from ChatGPT, Copilot, Gemini and Perplexity, all assessed by working journalists. Roughly 45% of answers carried at least one significant issue. About a third had serious sourcing problems, missing or misleading or simply wrong attribution, and around one in five contained major accuracy problems, which the researchers describe as including hallucinated detail and outdated information.

Two caveats before anyone quotes that at me. It tested news questions, not "what time does the shop shut", and news is a harder problem. It is also a snapshot of models that have since been updated. But the failure mode it documents, confidently stale information presented without workable sourcing, is precisely the one that hurts a local business.

Google says the quiet part out loud in its own help documentation: AI Overviews "can and will make mistakes". That is the vendor talking, not a critic.

Three things go wrong most often. Training data that predates your last change, so the assistant repeats the version of you that was true two years ago. Live retrieval that finds a stale page, usually a directory nobody has touched since it was created. And entity confusion, where the model merges you with a similarly named business two counties over and hands your customer somebody else's phone number.

What that costs, in the only terms that matter

Consider what this looks like in practice. A whale watching operator in West Cork moves the last sailing forward by two hours once the September schedule kicks in, and updates the website the same afternoon. A family planning the trip asks an assistant what time the last boat goes, gets the July timetable read back with total assurance, and drives the guts of three hours to stand on a pier looking at an empty mooring.

They do not blame the assistant. They blame the operator. The error was not yours, the correction was not available to you, and the one star review is going on your listing.

The same shape applies without the drama. A dead phone number lifted from an old directory. A service you dropped eighteen months ago, still being recommended. An address from before you moved. Nobody rings to tell you. They quietly go elsewhere, and the lost enquiry never appears in any dashboard you own.

Ask it yourself, before you assume you are fine

Most owners have never actually checked. That is the honest starting point, and it takes about ten minutes.

Open ChatGPT, Google's AI Mode, Copilot and Perplexity in turn, and ask each the questions a customer would genuinely type. Not clever prompts. Real ones. What time does [business name] in [town] open on a Saturday. What is the phone number for [business name]. Does [business name] do [service you no longer offer].

Write down every wrong answer and where the assistant says it got it from. Those citations are the map. If three of the four are quoting the same abandoned directory page, you have found your actual problem and it is not your website. Repeat once a quarter, and immediately after anything material changes.

Where your business facts actually live

Before you can fix anything you need to know what you control and how fast a change lands. These four sources cover almost everything an assistant will pick up about a small business.

Source	Who controls it	Realistic time to reflect a change
Google Business Profile	You, once verified	Usually fast, though edits can be held for review
Bing business listing	You, once claimed	Slower, and publishing can queue
Your own website	Entirely you	As fast as your host lets you deploy
Third-party directories	Often nobody	Weeks, months, or never

Each of those needs a different move, which is what the rest of this is about.

Google Business Profile

For anything Google touches, this is the heavyweight. Google's own guidance for site owners on AI features is unusually blunt about it: there are "no additional requirements to appear in AI Overviews or AI Mode", you do not need new files or special markup, and the practical advice it gives includes making sure your Business Profile information is up to date. When the vendor's optimisation advice is "keep your listing accurate", take the hint.

Your Bing listing

This is the one almost nobody has claimed, and it matters more than its search share suggests, because assistants that lean on Microsoft's index inherit whatever it holds. Microsoft rebuilt the product in October 2025 and moved it from bingplaces.com to bing.com/forbusiness. There is an import flow that pulls your verified Google profile across, which Microsoft says is "better at preserving key attributes like business name, hours, and contact details". Twenty minutes, once.

Your own website

Your site is the only source in that table you fully control, and it is the one an assistant is most likely to treat as authoritative about you specifically. Two things matter. The facts have to be in the visible text of the page, not buried in an image of a poster or a PDF menu. And if you use structured data, Google's standing advice is to make sure it matches the visible text, which for hours means the openingHoursSpecification markup and the line customers read must not disagree.

There is a practical dependency underneath all of that. If changing a set of hours on your own site means emailing an agency and waiting three days at an hourly rate, you will not do it in September, and the stale version stays live and keeps getting indexed. Being able to change your own hours and prices in under a minute is not a nice-to-have here, it is the thing that determines whether any of this ever gets done. The same argument runs through what AI-powered WordPress actually means for a business owner: control of the source is the whole game.

The directories

The abandoned listing is the hardest and the most common culprit. Some have a claim process. Some have a contact form. Some are automated husks with nobody home. Work through whichever ones showed up in your citations, claim what you can, and accept that a few of them will outlive all of us.

Correcting a wrong AI answer, in five steps

Audit. Ask the four main assistants the questions a customer would ask, and write down every wrong answer and the source it cites.

Correct the primary record. Verify your Google Business Profile and your Bing listing, and make the name, address, phone and hours identical in both, down to the punctuation.

Fix the page. Put the correct facts in plain visible text on your own site, and verify that any structured data agrees with what a human reads.

Chase the citations. Go after the specific stale directory pages the assistants quoted, in the order they quoted them.

Report and re-test. Use the feedback control in the AI answer itself, then re-run the same questions in two to four weeks and see what moved.

That last one is worth a sentence. Google does provide a way to flag a bad AI Overview: the thumbs-down control, then "Report a problem", with a category. It is a vote, not a support ticket, and you should treat it as such. But it costs nothing.

What none of this can promise you

The limitation, stated plainly, because you will find it out anyway.

Fixing the sources does not fix the answer on a schedule. Retrieval-based answers tend to move once the underlying pages are recrawled, which can be days or can be a couple of months. Answers coming from training data do not move at all until a model is retrained, and nobody outside the labs knows when that is. You can do every step above correctly and still get a wrong answer back six weeks later.

There is also no guarantee of appearance at any point. None of these systems owes you a listing, and there is no appeals process. What you are doing is removing every reasonable excuse the machine has for getting you wrong. That is a genuinely different, and smaller, promise than "fix it".

I will admit I was slow on this one. For a long time I told people that if the website was right, everything downstream would sort itself out eventually. That was true when the downstream was one search engine. It is not true when four systems are each assembling their own version of you from whatever they can reach.

When this is genuinely not worth your afternoon

Proportion matters, and I would rather say this than have you spend a Sunday on it for nothing.

The Reuters Institute's Digital News Report for 2026 found weekly use of AI chatbots for news rose from around 7% to about 10% globally, with only about 1% calling it their main source. Different question to local business lookups, so treat it as a direction of travel rather than a measurement of your market. But it is not yet the front door for most people.

If you are a walk-in trade with a fixed shopfront on a busy street, most of your custom arrives on foot or by phone, and your details have not changed in a decade, this is genuinely low on your list. Keep the Google listing accurate, which you should be doing regardless, and get on with your week. The businesses that need to act are the ones where something has recently changed, where customers travel to reach you, or where getting the detail wrong costs somebody a journey.

For a fuller picture of how these systems actually read a site before they answer, how AI agents read a business website covers the structural side of it.

Conclusion

The uncomfortable shift is not that machines answer questions about your business. It is that they answer without asking you, and there is nobody to ring when they get it wrong.

What you can do is make the correct version of your business the easiest one to find and the hardest one to contradict. Accurate listings on both major platforms. Facts in readable text on a site you can edit the day something changes. A ten minute check every quarter, and again the moment anything moves.

Ask the four assistants what they currently say about you. Whatever comes back, you will know more than you did this morning.

Frequently Asked Questions

Can I contact OpenAI or Google to correct information about my business?

Not in the way you can correct a listing. There is no business dashboard where you edit what an AI assistant says about you, and no support queue for factual corrections about a small business. Google offers a feedback control on AI Overviews, including a "Report a problem" option, but it is a signal rather than a ticket. The reliable route is to correct the underlying sources, principally your Google Business Profile, your Bing listing and your own website, and let the systems pick the change up.

How long does it take for an AI assistant to stop repeating old information?

It depends on where the answer came from. If the assistant is retrieving live pages, changes typically appear once those pages are recrawled, which can be anywhere from a few days to a couple of months. If the wrong detail is baked into the model's training data, it may persist until that model is retrained, and no provider publishes a timetable for this. Re-test every few weeks rather than assuming it is done.

Why does AI give different answers about my business than Google Search does?

Because they are not reading the same thing. Google Search returns pages; an AI assistant assembles a statement from several sources at once, which may include a search index, its training data, and directory or review sites. Google has described using a "query fan-out" technique that issues multiple related searches across subtopics and data sources. If one of those sources is stale, it can contaminate an answer even when your own website is entirely correct.

Do I need special code or an AI file on my website to be read correctly?

No. Google states directly that there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary", and that you do not need new machine readable files or special schema.org structured data. What helps is ordinary good practice: the correct facts in visible text, structured data that matches what a human reads on the page, and a site that is crawlable.

Should I claim a Bing listing if almost nobody uses Bing?

Judge it on where the data goes rather than on search share. Assistants that draw on Microsoft's index inherit what that index holds about you, so an unclaimed or wrong listing there can surface in answers well away from Bing itself. Microsoft moved the product to bing.com/forbusiness in late 2025 and added an import flow from a verified Google Business Profile, so claiming it is closer to twenty minutes than an afternoon.

What if an AI assistant confuses my business with a different one?

Entity confusion is usually a consistency problem. Where two businesses have similar names, models merge records unless something separates them clearly. Make your name, address and phone number byte-identical everywhere you control, use your full trading name rather than a shortened version, and make sure your website states plainly who you are and where you operate. It will not always be enough, but inconsistent details are what makes the merge likely in the first place.

Sources

European Broadcasting Union and BBC, News Integrity in AI Assistants, October 2025

Google Search Central, AI features and your website

Google Search Help, AI Overviews and how to give feedback

Microsoft Bing Blogs, Introducing the New Bing Places for Business, 3 October 2025

OpenAI Help Center, Searching the web with ChatGPT

Reuters Institute, Digital News Report 2026, executive summary