
When the machine answers for you — who is accountable?
Berger+Team · Sep 25, 2026
Topic: Data ownership & security
A customer calls and says a sentence that did not exist three years ago: “The AI said you are open until six.” You close at five. Nobody ever wrote that. And still it was there, in an answer that looked like information from you. And with it the question of who carries accountability for an answer like that.
Who was that?
The question decision-makers actually have.
Between you and your customer there is now a party nobody hired: a machine that gives out information about your business without asking you. It answers in your name, but not on your behalf. It quotes you without quoting you.
The obvious question is: who is liable? It is a fair question, but a dead end — in any single case it is a legal matter, and neither a blog post nor a supplier settles it. The European legal framework is only now taking shape; we described it in a separate piece on data sovereignty towards AI systems, though there with an eye on training, not on the answer.
The question you can answer yourself today is a different one — and in practice the more important: can you trace where a claim came from, and correct it at the source?
Why a machine invents things at all.
Language models do not play back what is stored; they produce the most likely continuation. Where a fact is missing, one still appears — that is not a fault, it is how they work, and the technical term is hallucination.
An uncomfortable consequence follows: a gap is more dangerous than a wrong entry. Where nothing is on record, something plausible gets filled in — the industry's usual hours, the usual price range, the usual services. If you leave no answer, you still get one.
Three cases worth keeping apart.
It pays to separate what can go wrong:
- A false statement about you. The machine claims something untrue. That is a reputation matter — described in When the AI cites you wrong.
- An outdated statement. The fact was right once and no longer is. The most common case, and the mildest — as long as a current source stands against it.
- An invented statement. There never was a source. That is the case at hand.
All three call for the same counter-measure, and it is unspectacular: one place where the truth sits, current, machine-readable and clearly tied to your domain — the idea behind Your domain is the source.
Traceability is an operational quality, not a legal question.
Whether a claim about your business can be traced comes down to three things, all in your hands:
- Is the fact recorded anywhere on your side? Opening hours, price range, area served, services — as data, not only as a sentence on a page that a person finds and a machine skips.
- Is it unambiguous? Two places with two truths are worse than a gap. Then the machine decides which one to believe.
- Is it current? A fact that has stood for two years does not become false because it is old — but it loses authority when something newer sits beside it.
Answer yes three times and, if it ever comes to a dispute, you have more than a good feeling: a datable source with a date on it. That is the difference between “we never said that” and “here is what has been on record since 12 March”.
What that means for the decision.
This is not about pre-answering every conceivable question — nobody manages that. It is about the handful of facts where wrong information does real damage: how to reach you, what you are responsible for, what you do and expressly do not, and anything touching safety or deadlines.
Those facts belong somewhere that can be changed without calling a developer. Because traceability is not a one-off setup, it is a habit: when something changes, it must change in one place — not in seven, four of which get forgotten.
And if you want to know what machines say about you today: asking once yourself costs nothing. How to check that systematically, instead of hearing it by chance from a customer, is in the visibility cockpit.
The one thing to take away.
“Who is liable?” is, case by case, a question for lawyers. “Where did this come from, and where do I fix it?” is a question for you — and you can answer it today. With a maintained, unambiguous source, an incident turns into a correction. Without one, you are arguing without evidence.
Frequently asked questions.
How do I find out whether an AI gives wrong information about my business?
Regularly ask the major AI assistants about your own business — services, opening hours, prices, and how you compare to competitors — and check the answers for accuracy. Without this check, wrong information often only surfaces when customers arrive with false expectations or never get in touch at all.
Can I correct wrong AI-generated information about my business?
You can't directly correct a single AI answer — but you control the sources it's built from. AI systems rely mainly on your website and public directories such as your Google Business Profile or trade directories.
Concretely: enter the same phone number, address, and opening hours everywhere — maintained in one place that all channels draw from. If that source has current, unambiguous, machine-readable details, it can gradually crowd out outdated or wrong information; realistically it takes weeks to months for AI answers to catch up.
Why does an AI recommend me with the wrong number or opening hours?
Because AI assistants mostly pull their facts from map and directory data. If your details are inconsistent or outdated there — an old number, a wrong opening hour — the AI picks up exactly that and states it confidently as an answer.
Directories also often copy from one another, so a single error spreads across several platforms. For your business, that means a small inconsistency in one place can end up as a wrong answer in several AI responses.


