Your chatbot can make promises you have to keep
An airline once argued in court that its chatbot was a separate legal entity, responsible for its own statements. It lost. In 2026 the courts have settled the question, and a lot of small businesses are being sold the exact thing that lost that case.
In 2022 a man’s grandmother died. He went to an airline’s website and asked its chatbot about bereavement fares. The chatbot told him to buy a full-price ticket and apply for the discount within ninety days.
No such policy existed. The airline refused the refund, and he took it to a tribunal.
The airline’s defence is the part worth remembering. It argued that the chatbot was a separate legal entity, responsible for its own actions.
The tribunal member called this “a remarkable submission” and ordered the airline to pay.
The detail everybody skips
The correct policy was on the airline’s website. It was sitting on another page, accurate and public, the whole time.
It did not help. The tribunal’s reasoning was that a customer has no way to know which part of your website to trust. If one part of it says one thing and another part says something else, that is your problem, not theirs.
Read that twice if you are about to put a chatbot on your site. Having the right answer somewhere on the page does not protect you from the bot’s version of it.
The money was trivial — about eight hundred Canadian dollars, most of it fees. The principle was not, and it has been quietly hardening ever since.
In 2026 this stopped being an open question
Two rulings this year, both in Germany, a fortnight apart.
In May, a higher regional court ruled against a clinic whose website chatbot had listed its doctors with invented specialist titles — qualifications no medical body recognised. The case was brought by a consumer advice centre. The clinic was found liable under unfair competition law.
That one should land for anyone running a small business. Not a multinational. A clinic. And what the bot invented was credentials.
A fortnight later a regional court granted an injunction against a major search provider whose AI-written summaries had falsely associated two businesses with scams and subscription traps. The provider was ordered to stop, and left carrying 80% of the costs.
The principle now stated plainly by the courts: a chatbot acting on behalf of, and in the appearance of, a company is not a third party. Its output is the company’s own statement. The old “we only host it” protection does not apply. And crucially — careful configuration is not a defence, because by deploying it you accept the risk that it will say something nobody anticipated.
“The AI said it, not us” is finished as an argument.
What you are actually being sold
Right now every small business on earth is being pitched a chatbot. Answer customers around the clock. Never miss an enquiry. Twenty pounds a month.
Describe the same thing as a hire and listen to how it sounds. You are taking on someone who:
- works unsupervised, overnight, with nobody checking
- is authorised to state your prices, policies and qualifications
- cannot reliably be trained, and will not tell you when it has gone wrong
- never says “I’m not sure, let me find out”
- makes commitments that are legally binding on you
- leaves no colleague who remembers the conversation
Nobody would hire that person. Plenty of people are installing them.
And the legal risk is only the headline. The faster damage is commercial: a software company’s support bot once invented a licensing restriction that did not exist, and customers began cancelling over a policy the company had never had. Nobody sued. It just quietly cost them, and they found out late.
The question nobody asks first
Here is the thing that gets skipped in every chatbot pitch.
What is the bot for? Answering customers’ questions.
Where do those questions come from? A website that does not answer them.
Go and look at what people actually ask: opening hours, parking, prices, whether you take card, how long it takes, whether you do home visits, what happens if they cancel. Almost none of it is conversational. It is a list of facts that were never written down.
A chatbot, in most small businesses, is an expensive and legally hazardous workaround for a page that does not say enough. Write the answers down and the questions stop arriving — and a sentence you wrote on a page cannot invent a refund policy at two in the morning.
That is not a sophisticated strategy. It is just cheaper, safer, and it works while you are asleep too.
It cuts the other way as well
The Munich ruling is worth holding onto for a different reason.
If an AI system is saying something false and damaging about your business — inventing a policy, attaching you to a scam, getting your credentials wrong — that is not simply bad luck you have to absorb. Two businesses took it to court this year and got an injunction.
Most small businesses assume there is nothing to be done. There is increasingly something to be done.
The fair question
devkoi does not build chatbots. That is a genuine limitation — if you want a conversational widget on your site, this is not the tool.
It is also, on the evidence above, not much of a loss. What devkoi builds is pages that say things you approved, which stay said until you change them.
One thing that is worth knowing given everything above: every version of your site is kept. If someone ever claims your website promised them something, you can go back and see exactly what the page said on the day they read it. That is a dull feature until the first time you need it.
What to do this week
1. If you already have a chatbot, interrogate it. Ask it your five most expensive questions — refunds, cancellations, guarantees, allergens, qualifications — and read the answers as though they were a contract, because a tribunal will. Do it today, not after someone relies on one.
2. Move every answer that costs money onto the page. Anything with a price, a deadline, a guarantee or a legal consequence should be a sentence you wrote, sitting on a page, not generated fresh for each visitor.
“Add a section with our refund policy, cancellation terms and what’s included in the price.”
3. Never let anything automated state your qualifications. That is the clinic case exactly. Registrations, memberships, certifications, insurance — those are facts, and they belong in your words.
“Add our registration number, insurance and qualifications to the about page, exactly as written here.”
4. Write down the questions you actually get asked. Keep a note by the till or in your phone for a week. That list is the most valuable page on your website, and it does not exist yet.
“Add an FAQ with these eleven questions and answers.”
5. If you genuinely need a bot, cage it. Let it answer from a fixed list and hand everything else to a person. “I’ll get someone to call you back” is a complete, safe, perfectly good answer — and it is the one the airline wishes it had given.
6. Check what is being said about you. Ask a few assistants about your business and see what comes back. If something is false and damaging, you now have more standing than you think.
The airline’s argument was absurd, but it was also revealing. Somebody genuinely believed that putting a machine between you and your customer put a layer of distance between you and your promises.
It does the opposite. It gives something the authority to speak for you, at scale, unsupervised, to people who have no way of knowing it is guessing.
A page that says what you meant, and keeps saying it, is a strange thing to have to argue for in 2026. But here we are.