Can a company be held to what its chatbot promised a customer?

The question usually shows up after the fact

A customer asks a support chatbot whether they can get a refund, or whether a feature is included, or whether a deadline can be extended. The bot says yes. Later, your team says no. The customer points at the chat and asks why the answer changed.

Whether a company can be held to what its chatbot said is partly a legal question, and the answer depends on where you operate and on the specifics. We are not lawyers, and nothing here is legal advice. What we can do is lay out why the position "the bot said it, not us" is a weak one to plan around, and what you can settle ahead of time.

Why "the bot said it" is hard to lean on

From the customer's side, the chat window sits on your website, carries your name, and answers in your voice. They have no way to tell which statements were reviewed by a person and which were generated a second ago. It is reasonable for them to treat the answer as yours, and a company that tries to disown it tends to look evasive even if it wins the argument on a technicality.

There is also a practical side. Even when you are confident you are in the right, the dispute costs time, goodwill, and sometimes a concession you did not plan to make. So the useful question is less "could we win?" and more "how do we make this situation rare?"

The chatbot is not the only place this happens

A human agent can also promise something they should not, and companies already deal with that through training, written policy, and a way to correct mistakes. A chatbot differs in one important way: it can make the same mistake with every customer who asks the same question, all day, without anyone noticing. That is why the controls matter more, not less.

Nobody in that exchange acted in bad faith. The bot produced a confident, friendly answer that did not match the policy. Now you have a customer holding a screenshot.

What you can settle ahead of time

• Decide which topics the bot may answer at all. Refunds, legal terms, medical or financial advice, and anything with a dollar amount attached deserve a deliberate decision, not a default.
• Make the source of an answer visible. If an answer points to the page it came from, you and the customer can both check it against the real policy. If it points nowhere, nobody can.
• Check the bot's answers against your own policies. Ask it the questions customers actually ask about refunds, cancellations, and deadlines, and compare the answers to your policy text. Do this again whenever the policy changes.
• Give the bot a way to say "I'm not sure." A bot that hands the question to your team on a sensitive topic is easier to defend than one that always produces something.
• Read the vendor's contract on responsibility. Ask who is responsible when the bot is wrong. Expect the answer to be you, and find that out before you sign.

What not to rely on

A small disclaimer under the chat box saying answers may be inaccurate is better than nothing, but it is not a substitute for the steps above. A disclaimer tells the customer not to trust the answer, which also undercuts the reason you put the bot there. Treat it as a backstop, not a plan.

If this matters for your business, for example if you sell something regulated or high-value, talk to a lawyer who knows your jurisdiction. They can tell you what a chatbot's statement counts for where you are, which no general article can.

How we handle this

Our agent is built to answer from your documentation, and on a multi-page help center it names or links the page the answer came from, so a statement can be checked against your own text. When it cannot answer from your documentation, it is instructed to say so plainly and point the customer to the contact path set up for your deployment, rather than guess. We do not promise the agent will never be wrong, and we do not offer a guarantee that it will always match your policies. Because it reads your documentation, an out-of-date policy page produces an out-of-date answer, so keeping that page current, and telling us when it changes so we can update what the agent reads, is on your side.

Related reading

• See all 26 guides in the Actually Helpful library, grouped by what you are actually trying to figure out.
• Why every chatbot answer should link to a real source, on making answers checkable.
• What should happen when a chatbot can't answer, on the moment a bot should step aside.
• What to check before you sign a chatbot contract, on who carries the risk.