Will an AI chatbot make up prices?

It can, if nothing stops it. A model with no connection to your records generates a plausible answer, and a plausible price is indistinguishable from a real one until a customer turns up expecting it. The protection is a tool that reads your actual records before replying.

Do I need technical skills to set one up?

For most tools, no — connecting a channel is an authorisation flow you click through. The work is in the content: your products or services, your prices, your opening hours. A tool with nothing to read cannot answer anything.

Can I take over a conversation myself?

You should be able to, immediately, and it is worth testing before you commit. Any tool you cannot switch off mid-conversation is one you will regret during your first complaint.

What happens when the AI gets something wrong?

Ask specifically. A tool that escalates to you, and shows you which conversations are waiting, is a different proposition from one that fails silently and leaves you to find out from the customer.

Guides

What to check before you put an AI chatbot in front of your customers

A short, practical checklist: where its answers come from, what it does when it does not know, what happens to your data, and the questions that separate a useful tool from a liability.

Ask where the answer comes from

This is the question that matters most and the one least often asked. When a customer asks a price, does the tool look it up, or does it generate a plausible number?

A chatbot built on a language model with no connection to your records will answer confidently either way. You will not notice until a customer arrives expecting a price you never charged.

Ask to see what happens when a product is out of stock. The answer tells you almost everything.

Ask what it does when it does not know

Every tool handles the easy questions. The difference shows up at the edges: a complaint, an unusual request, a question about an order that has gone wrong.

A good tool stops and fetches a person. A bad one keeps talking. The second is more impressive in a demo and worse in your inbox.

Ask whether it can promise things it cannot do

A subtle failure: the bot says "let me check and get back to you" and then nothing happens, because most systems can only reply when a customer writes first. The customer waits for a message that will never come.

Ask what happens after that sentence. If the answer is "nothing", that is a customer lost politely.

Ask about your customers' data

Where are the conversations stored, and who can read them?
Does connecting it require your social account password? It should not — Meta's official authorisation flow never asks for one.
Can you turn it off instantly and answer manually?

Ask what it costs when it works

Pricing usually scales with messages. Work out what a busy month looks like for you, not an average one, and check what happens when you exceed the plan — whether it stops replying or silently charges more.

On WhatsApp specifically, remember Meta charges separately for conversations you start outside the 24-hour window. That cost belongs to Meta, not to the tool, and no tool can waive it.

Start small

Connect one channel. Put in a handful of real products or services rather than your whole catalogue. Message it yourself from another phone and try to break it — ask for something out of stock, complain, change your mind halfway through an order.

Ten minutes of that tells you more than any feature list, including this one.

Common questions

Will an AI chatbot make up prices?: It can, if nothing stops it. A model with no connection to your records generates a plausible answer, and a plausible price is indistinguishable from a real one until a customer turns up expecting it. The protection is a tool that reads your actual records before replying.
Do I need technical skills to set one up?: For most tools, no — connecting a channel is an authorisation flow you click through. The work is in the content: your products or services, your prices, your opening hours. A tool with nothing to read cannot answer anything.
Can I take over a conversation myself?: You should be able to, immediately, and it is worth testing before you commit. Any tool you cannot switch off mid-conversation is one you will regret during your first complaint.
What happens when the AI gets something wrong?: Ask specifically. A tool that escalates to you, and shows you which conversations are waiting, is a different proposition from one that fails silently and leaves you to find out from the customer.