What an AI Answer Gets Wrong About Your Business, and Who Wears It
When a chatbot gives a customer the wrong price, the wrong hours or the wrong service area, the machine is not the one who loses the job. Here is where the errors come from and what you can actually control.
A plumber I know lost a Saturday job because a chatbot told someone he did not work weekends.
A plumber I know lost a Saturday job because a chatbot told someone he did not work weekends.
He does. He has worked Saturdays for eleven years. Somewhere on the internet there is a profile that says otherwise, put there by whoever built his site in 2016, and a machine read it, believed it, and passed it on with total confidence to a person who then rang somebody else.
He never found out. His wife found out, months later, from the customer, at a barbecue.
I have written before about how wide the gap is between what a small business looks like and what it measures. This is the same gap, moved somewhere you cannot watch.
That is the shape of this problem. It is not dramatic. Nobody hacks you. You just quietly stop being recommended correctly, and the feedback loop that would normally tell you something is wrong has been cut.
The machine does not carry the cost
I want to be precise about this, because it is where people get comfortable.
When a search result was wrong, you could see it. You typed the query, you saw the stale listing, you rang the directory and got it fixed. The error had an address.
An AI answer has no address. It is generated fresh, it is different for the next person who asks, and there is no page you can point at and say "that one is wrong". The error is real and the consequence is yours, but there is nothing to correct, because there is nothing that persists.
The plumber did not lose a job because of a technology failure. He lost it because of a decade-old line on a page he had forgotten about, which nothing had ever punished him for until something started reading the whole internet and repeating it out loud.
Where the wrong answer comes from
After a fair bit of digging, I think there are only three sources, and it helps to know which one you are fighting.
Sometimes it is a live search at the moment the question is asked. This is the one you can influence. Fix the page, and the next answer drawn from it improves.
Sometimes it is training data from an old crawl. A model can be confidently describing the version of your business that existed three years ago. You cannot reach into that and change it. You can only make the current version easy enough to find that it wins.
And sometimes it is somebody else's page about you. A directory, a supplier's site, an industry list, a profile you set up once and never logged into again. For a lot of small businesses this is the richest source of information about them anywhere on the web, which means it is the easiest thing to quote. It is also why showing up in an AI Overview is not simply a matter of fixing your own site.
That third one is usually the culprit, and it is the one nobody checks.
Contradiction is worse than silence
Here is the thing I got wrong for years.
I assumed the risk was having too little information out there. It is not. The risk is having several versions of the same fact, all plausible, none marked as current.
Give a model one phone number and it will use that number. Give it three and it will pick one. It has no way of knowing which one you would have chosen, and it will not tell you it had to choose.
So the first job is not publishing more. It is going and finding every place your hours, your number, your address and your service area appear, and making them agree. That is the same discipline behind keeping your name, address and phone number consistent for ordinary local search, and it is unglamorous work. It is also the only part of this you fully control.
What structured data does and does not do
There is a lot of advice going around that says the fix is schema markup. It helps, but not for the reason it is usually sold.
Google is unusually blunt about it: structured data "isn't required for generative AI search, and there's no special schema.org markup you need to add" (Google Search Central). That second clause is the one that travels. Schema.org is a shared vocabulary, not a Google product, and it contains no AI type. Anybody selling you AI schema is selling you something that does not exist in the vocabulary at all.
What markup does do is state a fact in a form that cannot be misread off a page layout. Your address in a LocalBusiness block is unambiguous in a way that an address in a footer is not.
Most Australian small businesses have not done it. In our September research, 74% published some JSON-LD, and 57% published an Organization block, but only 9% published opening hours*. So the business name is usually machine-readable and the trading hours almost never are. Ninety-one per cent of Australian small business sites cannot answer "are they open right now" to anything that reads the page. The markup is not a ranking trick. It is you writing the facts down in a form that leaves less room for a guess.
You will not be told
The hardest part of this is not technical. It is that the failure is silent.
Google now reports a little of it. Search Console's generative AI report covers AI Overviews and AI Mode, and carries exactly one metric: impressions, "how many times links to your site were shown to a user in a generative AI feature on Google Search" (Google Search Console Help). It groups them by page, country, device and date. No clicks, no queries, and no indication of what was actually said about you.
For every other tool, you get nothing. No alert, no error, no report. Just a slightly quieter phone, and no way to attribute it.
So you have to go and look. Ask the four big tools the questions your customers ask, write down what comes back, and do it on a schedule rather than when you get suspicious. An afternoon a quarter is not much to pay for the only feedback loop available.
The plumber has his Saturdays back on three profiles now. He still does not know how many calls he lost in the eleven months before anyone noticed, and neither do I.
What would you find, if you went and asked?
Sources
- Google Search Central, Optimizing your website for generative AI features on Google Search - retrieved 19 September 2026, https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- Google Search Console Help, Generative AI performance report (Search) - retrieved 19 September 2026, https://support.google.com/webmasters/answer/16984139?hl=en
- Red Bridge Cyber, Australian SMB Posture Baseline, September 2026, https://redbridgecyber.com.au/research/smb-posture-baseline