AI 101, Lesson 1: Customer-Service Chatbots
What happens after you query a chatbot—and why the answers can be surprisingly helpful or so confidently wrong.
The chat window is only the front counter
Should I bother using this thing?
That may be the most useful question to ask about a customer-service chatbot.
You go to an airline, bank, utility or retailer’s website because you need something. Before you can find a human being, a little box appears:
How can I help you today?
Sometimes it really can help. It finds the right form, tells you where your package is or resets your password in seconds.
Other times it confidently gives you an answer that turns out to be completely wrong.
So what is actually happening after you type a question? And how can you tell when the answer is worth trusting?
Use a chatbot as a fast guide. Do not automatically treat it as an authority.
To understand why, it helps to know what may be hiding behind that little chat window.
When you type a question into a customer-service chatbot, it feels as though you are talking to one thing.
You aren't necessarily.
The chat window is an interface—the digital equivalent of the front counter at a store. Behind it may be several different systems performing different jobs.
A chatbot might:
Follow responses written in advance.
Search the company’s help pages or internal documents.
Look up information in your account.
Apply company rules.
Generate a new answer using a large language model.
Combine several of those methods.
Two chatbots can look almost identical while working completely differently underneath.
For practical purposes, it helps to think about three broad types.
1. The scripted chatbot
This is basically an automated decision tree.
A company anticipates common questions and writes approved responses.
If you ask about changing a flight, the system sends you down one branch. If you ask about a refund, it sends you down another.
