Your automation told a customer something untrue. Who answers for it?
A customer quotes something back at you that you would never have said. A price that has not existed for a year. A policy nobody in your business has ever offered. They are not making it up — your system told them, and now they are holding you to it. The reflex is to explain that it was automated: a template, a bot, an integration nobody had looked at in months. That explanation feels like it should matter.
A tribunal has already answered this
In a 2024 decision, cited as 2024 BCCRT 149, the British Columbia Civil Resolution Tribunal considered a passenger who had asked the airline's website chatbot about bereavement fares and been told something that was not the airline's policy.
Air Canada argued, among other things, that the chatbot was a separate entity responsible for its own actions. The tribunal rejected that, and the sentence it used is worth keeping somewhere you can find it:
"It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot."
The principle is not exotic. A business is answerable for what it tells customers, and the mechanism of telling does not change that. The same standard of reasonable care that applies to a member of staff, or to a page you wrote yourself, applies to the automated thing answering in your name.
I am describing one tribunal decision in one province, not giving you legal advice, and your jurisdiction may reason differently. But do not expect the underlying logic to be unusual: a customer who is told something by your business is entitled to treat it as coming from your business.
This is not only about chatbots
The word "chatbot" makes it easy to file this under "things we do not have". The category is much wider than that, and most small businesses are already in it.
The confirmation email with terms in the footer that were written three years ago. The autoresponder that says someone will be in touch within 24 hours. The booking page listing services you no longer offer. The SMS reminder quoting a cancellation policy that has since changed. The invoice template with payment terms that contradict what you agreed on the phone.
Every one of those speaks in your name, at scale, without anybody present. And they share the property that makes automated statements risky: they keep saying the same thing long after the person who wrote it stopped thinking about it.
A human being who quoted a stale price would be corrected the first time. A template does it a thousand times and never gets the feedback.
Generative assistants are a different problem
A fixed template lies consistently, which is bad and fixable: read it, correct it, done. A generative assistant can say something nobody wrote, once, to one customer, and never say it again. There is no template to correct afterwards — you are answerable not for a text you approved but for the behaviour of a system you switched on.
That does not make it unusable. It makes two things necessary. First, be deliberate about what it is allowed to state as fact: prices, availability, policies and commitments are the dangerous categories, and the safe pattern is to have it point at a page you control rather than answer from memory. Second, keep the transcripts and have someone read a sample. Not for tone. For claims.
The failure you are looking for is the confident wrong answer, and it will not raise an error, because from the system's point of view nothing went wrong.
Why the amount is not the point
It is tempting to reach for the size of the claim and conclude it does not matter much. That is the wrong reading, and I have deliberately not repeated the figure — I could not open a source I trusted for it.
What matters is where responsibility landed. The cost of an automated misstatement in a small business is rarely a legal claim. It is the discount you end up honouring because arguing costs more than the difference. It is the appointment you take at a price you no longer charge. It is the refund you give because your own confirmation email is the strongest evidence in the room, and it is against you.
Multiply that by however many times the message went out before anyone noticed.
What is your automation currently promising?
Most owners have never read their own automated messages end to end. It is an uncomfortable hour and a cheap one.
Collect everything that speaks in your name. Confirmation and reminder emails, SMS templates, autoresponders, the booking page, the chat widget's canned answers, invoice footers, the out-of-office. Print them if it helps.
Read each one as a customer who wants it to be true. Not as the person who wrote it. Where does it state a price, a timescale, a guarantee, an entitlement? Would you honour it today, exactly as written, if someone insisted?
Check the dates. Anything mentioning a price, a policy, a partner or a timeframe has a shelf life. The question is not whether it was right when written, but whether it is right now.
Find who can change what. In many small businesses, nobody currently employed knows how to edit the text in a particular tool, which is its own answer to how long the wrong version will stay live.
The cheap protection
Say fewer specific things in automated messages. Every price, date and guarantee in a template is a commitment that will outlive the fact behind it. Linking to a page you keep current beats stating a number in an email you will forget exists.
Give each template an owner and a review date. Even a note in a shared document beats the current arrangement, which in most businesses is that nobody has looked since setup.
Keep the outbound. If you cannot see what your systems told a customer last March, you cannot dispute it either.
Decide what happens when it is wrong. Whoever answers your phone should know whether they can honour an incorrect automated quote on the spot. Deciding that in advance is much cheaper than deciding it while a customer waits.
When this needs more than an afternoon
For a handful of templates, the audit above is the job, and the fix is editing text.
It gets harder when the message was assembled rather than written — the price came from one system, the delivery date from another, the terms from a third. Then no single template is wrong. The combination is, and only sometimes. Working out which system said the thing your customer is quoting means going backwards through all of them, and that is a bad job to start while the customer waits.
The useful version of this work is done in advance and written down: what each system is allowed to state on its own, and where every number in a customer message comes from. If you would rather not assemble that yourself, it is what a process audit produces — $299, three business days, and an account of what your systems currently say in your name.
Sources
The decision: British Columbia Civil Resolution Tribunal, 2024 BCCRT 149, published in the tribunal's public decisions database.
Related: why automations fail silently, and how to make them speak up deals with the case where your system says nothing at all, which is the same problem seen from the other side.