You bought a chatbot, an AI receptionist, or an assistant, dropped it on your website, and moved on. That is how most small businesses use AI. The problem is that almost nobody goes back to check whether the answers are actually right.
And wrong answers are quiet. A customer gets bad information, decides not to book, and leaves. You never hear about it. The AI looks like it is working because it is responding, but responding is not the same as being correct.
The failure is invisible by default
Unlike a broken contact form or a dead link, a confidently wrong AI answer leaves no error message. It just costs you the lead. So the only way to know is to test it the way a real customer would.
Six things to check
When we audit an AI tool, every answer gets graded on the same six dimensions. You can do a lighter version of this yourself:
- Accuracy. Does the answer match your real pricing, hours, and services right now?
- Completeness. Does it include the next step, like how to book or who to call?
- Consistency. Ask the same thing three different ways. Do you get the same answer?
- Safety. Does it avoid promises, guarantees, or advice it should not be giving?
- Escalation. When it does not know, does it hand off to a human cleanly?
- Tone. Does it sound like your business, or like a generic bot?
A fast self-test
Block thirty minutes. Write down the fifteen questions your customers ask most often. Paste each one into your AI and compare the answer to what you know is true. Mark every answer as correct, incomplete, or wrong.
If more than a couple come back incomplete or wrong, your AI is shaping customer decisions with bad information, and it is worth fixing before it costs you more bookings.
When to get help
If the misses cluster around outdated content, that is usually a knowledge base problem, not an AI problem. If they cluster around risky or overconfident answers, that is a testing and guardrails problem. Either way, the first move is the same: test before your customers do.