Should a Small Business Use an AI Chatbot for Customer Service?

Customers expect quick answers at 9 p.m. on a Sunday. You expect to have a weekend. An AI chatbot promises to close that gap by answering questions around the clock. But a chatbot that gives a wrong price, invents a policy or traps an angry customer in a loop can cost you more than it saves.

So should a small business use an AI chatbot for customer service? This decision guide helps you answer that for your own business, based on the questions you get, how handoff to your team would work and what happens legally when the bot gets something wrong.

The short answer

An AI chatbot is usually worth it when:

• Most customer messages are repetitive and factual
• The answers come from information you control, such as your website, FAQ or order system
• You get messages outside business hours that go unanswered
• You can give customers a clear path to a person

It is usually not worth it, or should be kept very narrow, when your customer conversations are mostly complex, emotional, high-stakes or regulated.

Sort your customer questions first

The best predictor of chatbot success is the mix of questions you receive. Pull 50 to 100 recent messages from email, chat, social media and phone notes, and sort them into 4 groups.

If 60% or more of your messages land in the first 2 rows, a chatbot will likely save real time. If most land in the bottom 2 rows, you may get more value from faster email templates or an AI assistant that helps staff draft replies. Our guide on how small businesses can start using AI covers those staff-facing workflows.

How AI chatbots get their answers

Modern AI chatbots use a large language model to understand questions and write replies. The good ones are grounded: they search your approved content, such as your website, help articles or product catalog, and answer from it. This approach is often called retrieval-augmented generation, explained in our guide to RAG.

Grounding reduces errors but does not remove them. Language models can still produce confident, wrong answers, known as AI hallucinations. That is why the scope of what the bot is allowed to answer matters so much.

If you are unclear on terms, our explainer on AI agents vs chatbots vs copilots sorts out what each one does.

Who is liable when the chatbot is wrong

This is the question many owners skip. In general, a business is responsible for what its customer service tools tell customers, the same as it would be for its website or staff.

A widely cited example comes from Canada. In Moffatt v. Air Canada, decided in February 2024, British Columbia’s Civil Resolution Tribunal found the airline responsible after its website chatbot gave a customer wrong information about bereavement fares. The tribunal rejected the argument that the chatbot was responsible for its own statements and found the airline had not taken reasonable care to make sure the chatbot was accurate. It is a small-claims tribunal decision in Canada, not binding US precedent, but it shows how decision-makers are likely to view these situations.

In the US, the FTC has made clear there is no AI exemption from existing consumer protection law. Misleading claims made by a chatbot can create the same problems as misleading claims anywhere else.

Some states also have disclosure rules:

• California’s bot disclosure law (Business and Professions Code section 17940 and following, effective July 1, 2019) makes it unlawful to use a bot to knowingly mislead people in California about its artificial identity to sell goods or services, unless the bot’s nature is clearly disclosed.
• Utah’s Artificial Intelligence Policy Act, effective May 1, 2024 and amended in 2025, requires businesses to disclose generative AI use in certain consumer interactions, including when a consumer clearly asks, with stricter rules for regulated occupations in high-risk interactions.

Other states are adding rules each year; see our overview of US state AI laws. This is not legal advice. Talk to a lawyer about your specific situation.

Key takeaway: Treat every chatbot answer as if your business said it, because legally and reputationally, it did. Limit the bot to what it can answer accurately, and disclose that it is AI.

Design the handoff to a person

A chatbot without a good handoff frustrates customers. A chatbot with a good handoff makes your team faster. Build these rules in from day 1:

1. Always offer a way out. Customers should be able to reach your team at any point by asking for a person.
2. Trigger handoff automatically for complaints, refunds, cancellations, safety issues, health or legal questions, and any sign of frustration.
3. Pass the context along. The person taking over should see the full conversation, so the customer does not repeat themselves.
4. Be honest about timing. After hours, the bot should say when someone will reply and collect contact details.
5. Escalate when confidence is low. Set the bot to say “I’m not sure, let me connect you with our team” rather than guess.

Protect customer data

Customer conversations contain personal information. Before you launch:

• Read the vendor’s data policy, including whether conversations are used to train models and how long they are kept.
• Tell customers not to share payment card numbers or passwords in chat, and do not ask for them.
• If you handle health information, a general chatbot may not be appropriate under HIPAA. Check whether the vendor will sign a business associate agreement.
• Limit staff access to chat logs to those who need it.

Our guide on whether it is safe to put company data into AI covers the questions to ask vendors.

Pricing models to expect

Chatbot pricing changes often, so focus on the model rather than a specific number:

• Included features: many website builders, help desks and CRMs bundle a basic AI chatbot into existing plans.
• Per seat: a monthly fee per staff user, with the bot included.
• Per conversation or resolution: you pay for each chat the bot handles or resolves.
• Usage-based: costs tied to message volume or AI usage.

Usage-based and per-resolution pricing can rise quickly as traffic grows, so estimate your monthly volume and ask for a cap or alert. For a full vendor checklist, see our guide on how to choose an AI chatbot for your website.

Where to start: a 30-day pilot

1. Week 1: scope it. Pick the top 10 to 20 factual questions from your message review. Write or update the answers on your website.
2. Week 2: set it up. Load only approved content. Turn on AI disclosure, handoff triggers and after-hours messaging.
3. Week 3: test it hard. Have staff ask 50 real questions, including tricky and off-topic ones. Fix wrong or vague answers at the source.
4. Week 4: launch narrowly. Put the bot on 1 or 2 pages, such as the FAQ or contact page, and review every conversation daily.

Measure 4 things: share of chats resolved without handoff, wrong answers found in review, customer satisfaction and staff time saved. If wrong answers stay rare and customers are satisfied, expand. If not, narrow the scope or pause. Our guide to automating customer service with AI shows how to grow from there.

Conclusion

An AI chatbot can be a smart move for a small business with lots of repetitive, factual questions and a clear way to reach a person. It is a poor fit for complex, emotional or regulated conversations. Start with a narrow scope, disclose that it is AI, design a fast handoff and review conversations closely. You are responsible for what the bot says, so keep it limited to what it can get right.

Sources

• Moffatt v. Air Canada, 2024 BCCRT 149 (CanLII) — tribunal decision holding a business responsible for its chatbot’s misinformation
• FTC: Crackdown on deceptive AI claims and schemes — existing consumer protection law applies to AI
• California Business and Professions Code section 17941 — bot disclosure requirement
• Future of Privacy Forum: Overview of Utah’s 2025 enacted AI legislation — Utah AI Policy Act disclosure rules and 2025 amendments
• US Small Business Administration: AI for small business — chatbot use, review and data security guidance