Somewhere today, an AI has described your business to somebody and got part of it wrong.
Maybe it was minor. An old phone number, a service you dropped years ago, hours that changed in 2023. Maybe it was worse: the wrong prices, work you do not do, or a competitor’s business confidently described as yours.
Even if it is right ninety five percent of the time, that remaining five percent is being said to real people who are deciding whether to ring you. And for some businesses it is a great deal worse than five percent.
The good news is that this is fixable, and the reason it is fixable is the thing most people get wrong about it.
Hallucinations are not random
That is the key idea, so it is worth stating plainly.
When an AI says something incorrect about your business, it has almost never invented it from nothing. It has read something and drawn a reasonable conclusion from bad or confusing information.
Which means every persistent error has a source. Find the source, deal with it, and the error usually goes with it.
That reframes the whole problem. You are not arguing with a machine. You are tracking down a bad piece of information and correcting it.
Step one: find out what it is actually saying
If you have not already, go and ask. I have written a full session for interrogating an AI about your own business, so I will not repeat it here, but the short version is: ask about your business, ask how you compare to competitors, ask what somebody should know before hiring you, and write down anything that is wrong.
Run it more than once and on more than one platform. You are looking for errors that repeat, not one odd answer.
Sort what you find into two piles. Things that are factually wrong, and things that are simply missing. This post is about the first pile. The second pile is a content plan, which is a different job.
Step two: find where it came from
This is the bit almost nobody does, and it is the whole game.
Ask the AI directly:
“Where did you get that information?”
Most tools will show you their sources. Then open every single one and read it. Not skim, read. You are looking for the specific sentence that produced the error.
Sometimes it is obvious. An old directory listing with a number you stopped using. A review site profile you did not know existed. A page on your own site you forgot about.
Sometimes it is subtler: two sources disagreeing, and the machine picked the wrong one. Or a page about a similar business that got tangled up with yours.
Either way, you are now looking at the actual cause rather than guessing.
The three situations you will find
Nearly everything falls into one of three, and each has a different fix.
One: it is on something you control
The best outcome, and more common than you would expect.
An old page on your own website. A Google profile you have not updated. A social bio still describing what you did in 2021. A listing you set up years ago and forgot.
Fix: update it, and update it everywhere, not just in the one place you found it. Then be patient, because it takes time for that to filter through. Weeks rather than days.
While you are there, look for anything that disagrees with your other sources. Contradiction is what causes most of this. If your website says one thing and a directory says another, the machine has to choose, and it may not choose yours.
Two: it is on somebody else’s site
More awkward, and there are two moves.
First, just ask. Find whoever runs the page and politely tell them the information is out of date, with the correct version. A surprising number of people will fix it, particularly directories and local listing sites, who generally would rather be accurate.
Be pleasant about it. You are asking a favour.
Second, if they will not, out-compete it.
Here you need to understand why that page got cited at all. When somebody asks an AI a question, it does not run one search. It generates a spread of related sub-questions and searches those, then reads what comes back. The page containing your bad information ranked for one of those sub-questions.
Some tools will show you those sub-queries directly. If not, look at the title of the cited page and you can usually make a decent guess at what it was answering.
Then write something better for that question, on your own site. More specific, more current, more genuinely useful. You are not trying to get the bad page removed. You are trying to be the more obvious answer to the question it was answering.
Honest expectation: this works better the more established your site is, and it is not instant. But it is the only lever you have when somebody will not correct their own page, and it produces a useful page either way.
Three: nothing is wrong, the AI is just confused
The strangest one, and more common than it sounds.
The usual version is mistaken identity. Another business with a similar name, often in a different town or a different trade, and the machine has merged the two of you. Sometimes it is a former trading name, a similar domain, or a company that shares a director’s surname.
Nothing out there is factually incorrect. The confusion is being created by the overlap.
Fix: write the page that clears it up.
If the AI is already citing a page of yours for these questions, add a section to it addressing the confusion directly. If not, or if the muddle affects lots of different queries, make a dedicated page for it.
Something plainly titled, along the lines of “[Your business] and [similar name]: two different companies.” Then explain, factually and briefly, who you are, where you operate, what you do, and that you are unconnected.
A word of caution here, because this is where people get themselves in trouble.
Keep it neutral. Clarify the confusion, and stop. Do not use the page to explain why you are better than the other business, and do not make claims about them at all. You are publishing something about an identifiable company, and factual claims about other businesses are a genuinely risky thing to put your name to. A disambiguation page is a public service. A comparison page dressed as one is asking for a letter.
Neutral works better anyway. The machine needs to learn you are separate entities. It does not need your opinion of them.
What to expect
Three honest things.
It is slower than you want. These systems do not update the moment you change a page. Expect weeks, sometimes longer, and check again rather than assuming it did not work.
You will not fix everything. Some errors come from sources you cannot reach or identify. Aim to correct the ones that actually affect whether somebody rings you, and let the trivia go.
And it is worth doing once properly. Most of what causes this is old, forgotten, and correctable in an afternoon. The businesses with the worst problems are almost always the ones with a decade of stale listings nobody has looked at since they were created.
That is not a technology problem. It is a tidying problem that has suddenly started mattering.
Want somebody to track down what is being said about you and why? Get your free audit, or email hello@inkon.co.uk. I read every message myself.