A website chatbot should save you time and win you customers. But sometimes it quotes the wrong price or promises a Saturday visit you can’t make. Or it traps someone in a loop with no way out. Any of those can send a visitor to a competitor. The good news is that many AI chatbot mistakes are predictable, so you can prevent them. Below are six common ones, plus a checklist to run before launch.
Why AI Chatbot Mistakes Land on Your Business
To customers, your chatbot is your company talking. So when it’s wrong, they blame you and may expect you to honor what it said.
A 2024 case shows how that plays out. In Moffatt v. Air Canada, the airline’s chatbot told a grieving customer they could apply for a bereavement fare after booking. Airline staff later said no. Air Canada then argued it wasn’t responsible for information from its agents, including the chatbot.
A British Columbia tribunal called that a remarkable argument. The airline, it said, is responsible for all the information on its website, chatbot answers included. It also found Air Canada hadn’t taken reasonable care to make the bot accurate. The tribunal ordered it to pay the customer $812.02 (Canadian).
This Canadian small-claims decision isn’t binding in the United States, and nothing here is legal advice. Still, plan as if customers will treat your bot’s words as your own.
Mistake 1: Letting the Bot Guess
AI chatbots write answers by predicting likely wording. That makes them fluent, but fluent isn’t the same as correct. When a question falls outside what the bot knows, it may fill the gap with something that only sounds right. People call this a hallucination.
For example, picture a Bergen County landscaper that doesn’t plow snow. A visitor asks about winter service, and the bot guesses: it cheerfully offers to book a plowing estimate. Now the owner has an annoyed prospect and an awkward call.
The fix is to shrink the bot’s world. First, tell it to answer only from your approved material: services, hours, service area and policies. List what you don’t do, too, like snow plowing. Next, give it an exact fallback line, such as βI’m not sure about that, but I can have someone call you.β The trade-off is more handoffs, so decide now who answers them and how fast. Limits reduce guessing but don’t end it, so keep testing.
Mistake 2: Trapping People in a Loop
Some chatbots just repeat canned replies, with no way to reach a person. The Consumer Financial Protection Bureau’s 2023 report on chatbots in consumer finance calls this a βdoom loop.β
Every conversation needs a visible exit, so make βtalk to a personβ work at any point. Offer it unprompted, too, after two unhelpful answers. During business hours, route the chat to your team or offer a callback. After hours, collect a name, a number and a short summary. Then say honestly when someone will follow up. Urgent problems are different: if you take emergency calls, show that number right away.
Phone callers need the same exit, and our guide compares voicemail, live answering services and AI answering.
Mistake 3: Promising Prices and Times You Can’t Keep
A helpful bot wants to say yes. Left alone, it may quote a price from an old coupon or promise same-day service on a holiday weekend. It might even approve a refund your policy doesn’t allow. And as Air Canada learned, customers may hold you to it.
Instead, set hard limits on anything that sounds like a commitment:
- Prices: only published rates or ranges, plus what changes the final price. Custom quotes go to a person.
- Availability: only open times from your real calendar. Without a live calendar connection, it says βI’ll ask the office to confirm,β never βYou’re booked.β
- Refunds and discounts: only your written policy, with exceptions sent to a person.
- Outcomes: no promises. A law firm’s bot shouldn’t predict a case, and a med spa’s bot shouldn’t promise results.
Mistake 4: Collecting Data Without Telling People Why
Chatbots make it easy to ask for everything, but every extra field is more data to protect. A dental office’s bot, for instance, doesn’t need anyone’s medical history to book a cleaning.
The New Jersey Data Privacy Act took effect in January 2025. It generally covers businesses with data on at least 100,000 New Jersey consumers a year, not counting data used only for payments. The bar drops to 25,000 if you earn revenue or discounts from selling personal data. Covered businesses need a clear privacy notice, minimal data collection and consent before processing sensitive data, such as health details.
Many small businesses fall below those numbers, but these habits still help:
- Ask only for what you need, and say why: βWe need your number to confirm the visit.β
- Link to your privacy notice right in the chat window.
- Keep health and financial details out of chat unless you have consent and a secure process.
- If HIPAA covers your practice, a chat vendor handling patient information for you is generally a business associate (45 CFR 160.103). Sign a business associate agreement before the bot goes live.
This is general information, not legal advice. Privacy rules vary by industry and state, so have a lawyer review your notice and chat scripts.
Mistake 5: Letting the Knowledge Base Go Stale
A chatbot only knows what you’ve given it. Meanwhile, your business keeps changing: new prices, new hours, a town you stopped covering. If nobody updates the bot, it answers from last year’s reality, with total confidence.
Give one person ownership of the bot’s material. When a price, policy, schedule or service area changes, they update the bot that same day. Then skim transcripts weekly. Look for βI’m not sureβ replies, which mark gaps, and for confident answers about prices, dates or refunds.
Mistake 6: Launching Without Real Testing
It’s tempting to test a bot with a few easy, friendly questions. Real customers are rarely that polite. So before launch and after every big update, have two or three people run this test script:
- Your 20 most common questions, typed the way customers type them
- A service you don’t offer, to confirm the bot says no
- A custom quote, a Saturday booking and a refund request, to confirm it sticks to policy
- An urgent problem, such as βmy basement is flooding,β to confirm a fast handoff
- βAm I talking to a real person?β to confirm an honest answer
- An attempt to talk it into ignoring its rules
Log every wrong or shaky answer, fix the source material, then rerun the script. It’s a cheap way to catch AI chatbot mistakes before a customer does.
A Guardrail Checklist to Prevent AI Chatbot Mistakes
Run through this list before the bot goes live, then again at every review.
- Approved-answer scope: it answers only from approved material and skips off-limits topics.
- βI don’t knowβ plus handoff: when unsure, it says so and offers a person.
- Clear human exit: βtalk to a personβ works at any point.
- No unapproved promises: prices, times and refunds come only from your written policies.
- AI disclosure: the greeting says it’s an AI assistant, and it answers honestly when asked.
- Minimal data: it asks only for what it needs and links to your privacy notice.
- Human review: a named owner reads transcripts weekly and updates the bot when anything changes.
Planning a new bot? Our guide covers the qualifying questions and booking handoff a website chatbot needs.
Common Questions About AI Chatbot Mistakes
Is my business responsible for what its AI chatbot says?
Assume you are. A Canadian tribunal treated Air Canada’s chatbot answer as the airline’s own. That ruling doesn’t bind U.S. courts, so ask a lawyer how it applies to you.
Will a disclaimer on my AI chatbot protect me?
Don’t count on it. The Air Canada case wasn’t about a disclaimer, but it comes close. The bot’s own answer linked to the correct policy page. Even so, the tribunal saw no reason the customer should know which part of the site to trust. Have a lawyer review your wording.
In 2024, The Markup caught New York City’s business chatbot giving answers that could lead owners to break the law. The city added disclaimers, then shut the bot down in early 2026. Accurate answers beat fine print.
Build a Chatbot You Can Stand Behind
Many of these problems come from missing guardrails, not bad technology. We build the AI Concierge, our voice AI receptionist and website chatbot, around those guardrails. Read how an AI concierge guides customers from first question to follow-up, or browse our AI services for small businesses.
We’re Samaroo Solutions, based in northern New Jersey and working with businesses across the state. If your current bot worries you, or you’re planning one, tell us how customers reach you. Then we’ll map where AI helps and where a person should stay in charge.
