Every website builder and half the emails in your inbox are now offering the same thing: an AI assistant for your site. Paste in a line of code, it reads your pages, and it answers customers day and night.

Sometimes that is useful. But there is one thing the sales pages leave out, and a court has already spelled it out: whatever the bot says, you said.

This is the honest version — what goes wrong, what customers actually think of these bots, and the kind of chat that helps without the risk.

The short version

  • A chatbot on your site speaks for your business. If it promises a refund, a price or a policy, a customer can hold you to it.
  • AI chatbots make things up with total confidence, and a small print warning does not fix that.
  • Most customers would rather you didn't use AI for customer service. Their top worry is not being able to reach a person.
  • A scripted helper — your real answers to your most common questions, plus a clear way to reach you — gets most of the benefit with none of the risk.
  • If you do add a live AI bot, limit what it may talk about and never let it quote prices, refunds or anything legal.

"The chatbot said so" — and the airline paid

In November 2022, Jake Moffatt's grandmother died. He went to Air Canada's website to book a flight to the funeral and asked its chatbot about bereavement fares. The chatbot told him he could book at the normal price and claim the discount back afterwards, within 90 days.

That was not Air Canada's policy. Its real bereavement page said the discount could not be claimed after travel. When he asked for the refund, the airline refused.

He took it to British Columbia's Civil Resolution Tribunal, and in February 2024 he won. Air Canada's defence is the part worth remembering: it argued the chatbot was, in effect, a separate legal entity responsible for its own actions. The tribunal was blunt:

It should be obvious to Air Canada that it is responsible for all the information on its website.

The airline also argued he should have checked the chatbot's answer against the policy page. The tribunal rejected that too — there was

no reason why Mr. Moffatt should know that one section of Air Canada's webpage is accurate, and another is not.

The bill was small — C$650.88 in damages, plus interest and fees, about C$812 in all. The principle is not. That was a Canadian small-claims tribunal, not an American court, but it is the case every lawyer now cites on this, and the reasoning is ordinary: a business is responsible for what it tells customers, whoever or whatever does the telling.

Chatbots make things up — confidently

The Air Canada bot was not broken. Making up a plausible-sounding answer is simply what these systems do when they don't know.

In 2024, The Markup tested the chatbot New York City had built for small business owners, running on Microsoft's AI. It told them, among other things:

  • "Yes, you can take a cut of your worker's tips" — illegal in New York.
  • "No, landlords do not need to accept these tenants" — about housing vouchers, which landlords there cannot refuse.
  • "There are no restrictions on the amount of rent that you can charge" — not true for rent-stabilised apartments.

It carried a disclaimer that it "may occasionally produce incorrect, harmful or biased content." That did not make the answers right, and nobody reading one could tell which answers were the wrong ones. After two years of reporting, the city's new administration moved to shut it down in early 2026.

That was a city government with Microsoft behind it. A plug-in bot on a five-page site reading your About page will not do better.

What customers actually think

In December 2023 Gartner surveyed 5,728 customers. The headline:

  • 64% would prefer companies didn't use AI in customer service.
  • 53% would consider switching to a competitor if they found out a company was going to.
  • The top concern: it becomes harder to reach a person. Others on the list: that it takes away jobs, and that it gives wrong answers.

For a local business that is the whole point. People come to a small business partly because there is a person on the other end. A bot that stands between them and you can cost you the very thing that sets you apart.

Don't oversell it either

If you do use AI, describe it honestly. In 2024 the Federal Trade Commission started a crackdown it called Operation AI Comply. "Using AI tools to trick, mislead, or defraud people is illegal," the FTC's chair said at the time. One of the first cases was DoNotPay, which sold an "AI lawyer". According to the FTC it never tested whether its AI performed like a real lawyer, and in February 2025 it was ordered to pay $193,000 and stop claiming it could without evidence.

The lesson for a small business is simple: don't let your site claim the bot is "an expert", "always accurate" or a substitute for talking to you.

What works instead

Most questions a local business gets are the same ten. Are you open Sunday? Do you cover Battle Ground? How much is a standard clean? Can I bring my dog?

For those, a scripted helper is better than a live AI:

  • It shows a handful of suggested questions, each with your own answer, written once and checked by you.
  • Anything it has no answer for goes straight to a person — your phone number, the contact form or a booking link.
  • It can never invent a refund policy, because it only says what you wrote.

That is exactly how the chat assistant on my demos page works. Try it: every answer is scripted, and it never pretends to know what it doesn't.

And often the better answer is not a chat window at all, but a clear FAQ section on the page — which Google and the AI search tools can read too.

If you still want a live AI bot

Sometimes it genuinely earns its place — a large product catalogue, or lots of questions after hours. Then:

  • Fence it in. It should only answer from your own pages, and say "I don't know — here's how to reach us" for anything else.
  • Keep it off money and law. No prices it can make up, no refunds, no medical, legal or financial advice.
  • Say it's a bot. Label it clearly, every time.
  • Always offer a human. A visible phone number or contact link in the window.
  • Read the transcripts. Weekly, at least at first. You will find answers you would never have given.
  • Check what it keeps. Customers type names, addresses and phone numbers into chat windows. Know where the vendor stores them and for how long.

What I do

  • I build the scripted kind by default: your real questions, your real answers, and a clear route to you.
  • I recommend a proper FAQ before any chat window — it helps customers and search at the same time.
  • If a client wants a live AI bot, I set it up fenced in, labelled, with a human handoff — and I tell them plainly that what it says is theirs.