Insights

What a Good Small-Business Chatbot Should Never Say

A chatbot on your website speaks for your business, and it does it at 11pm with nobody watching. Three habits are worth testing for before you let one talk to a customer: promising things the business never promised, inventing details that sound plausible, and refusing to hand a conversation to a human.

What a chatbot should never do

A good small-business chatbot never states something about your business that it can’t trace back to your own content, and never keeps a conversation going when a person should take over. Everything below is one of those two rules being broken.

It should never overpromise

A chatbot trained to be agreeable says yes to almost anything, because a yes feels helpful. Here’s a made-up example of how that goes:

Visitor: Can you finish this by Friday? Bad bot: Absolutely, Friday works! Better bot: Timelines depend on the project, so I can’t confirm a date. The team can, and the contact page is the fastest way to reach them.

The cost of the first answer shows up later, when a real person has to honor a promise nobody authorized, or tell the customer the bot was wrong. Prices, timelines, and commitments belong to someone who can actually make them.

It should never make up policy details

Ask a chatbot about your return window, your service area, or whether you work weekends. If the answer isn’t in the content it was given, a poorly built one still produces a confident reply.

Visitor: Do you offer refunds? Bad bot: Yes, within 30 days of purchase. Better bot: I don’t see a refund policy in what I’ve been given, so I won’t guess. Please check with the team directly.

Reasonable-sounding is the problem. It’s often close enough to the truth that nobody notices until a customer acts on it.

It should never refuse to let go

The most frustrating chatbot is the one that loops. The customer asks for a person, and the bot offers another menu.

Treat “I need to talk to someone” as a successful outcome, not a failure to prevent. Any complaint, custom request, or question that needs judgment should end with a clear path to a human: a contact form, a phone number, an email.

What this looks like when it’s built right

This is how the chatbot on our own site works. It answers only from what’s published here. It doesn’t quote prices or timelines, because none are published, and it sends those questions to the contact page. It says up front that it’s an AI. And when a visitor describes a specific project, it doesn’t say whether it’s a good fit, because that call belongs to the team, not to software. We build chatbots this way for clients too, trained on their real docs and FAQs.

Not sure a chatbot fits your business at all? We wrote an honest comparison of chatbots, live chat, and no chat that doesn’t assume the answer is always yes.

If you’re evaluating a chatbot of your own, run three tests. Ask it for a price it shouldn’t know. Ask it a policy question your site doesn’t answer. Ask for a human. How it handles those three tells you more than any demo.