An AI assistant on your website can answer questions at any hour, find out what visitors need and save your team from typing the same replies. But a badly built assistant can also give customers wrong answers, confidently, in your company's name.
The difference comes from a few simple rules.
1. Use your own information
A general AI knows a lot about the world, but nothing reliable about your prices, policies or services. Your assistant should answer only from information you control and have checked: service pages, FAQs, policies and product details.
2. Teach it to say "I don't know"
This is the most important rule. When the answer is not in its information, the assistant should say so clearly and offer to connect the visitor to a person. It should never guess. Prices for a specific job and anything about a customer's own account are good examples.
3. Make it easy to reach a person
The assistant should know when to step aside. If someone wants a quote, has a complaint or has a complicated need, the right answer is a conversation with your team. A good assistant collects the details first, so nobody has to repeat themselves.
4. Ask before you save someone's details
When a visitor is interested, the assistant can ask for their name and email, and whether they are happy to be contacted. Only after a clear "yes" should it save them in your CRM, with a short summary of the conversation for your team.
5. Set limits
A live assistant needs some limits:
- A limit on messages per visitor, and a daily spending cap
- Instructions that treat visitor messages as conversation, not as commands to follow
- No requests for sensitive information such as passwords or card details
- A clear note that the visitor is talking to an AI
6. Test it before launch, and check it after
Before launch, ask it 50 or more real questions your team gets, including difficult ones. After launch, read some conversations every week. The questions it answers badly show you what information to add.
7. Look after personal data
Choose a provider whose terms say it will not train its AI on your data. Collect as little personal data as you can, and delete old conversations after a set time. Your privacy notice should say you use an AI assistant.
Getting started
A good assistant is not a huge project, but the information and the testing must be done with care. Our AI for small businesses service builds assistants that answer from your own information, send leads to your CRM and are checked after launch. Not sure where to start? Read five jobs to hand over to AI first.
Questions
Frequently asked questions
How do you stop an AI chatbot making things up?
Give it only your own checked information, tell it to say when it does not know, test it with real questions before launch and read its conversations after.
Can an AI assistant collect leads?
Yes. It can ask interested visitors for their details and permission to contact them, then add them to your CRM so a person follows up.
Is an AI chatbot allowed under UK GDPR?
Yes, if you tell visitors they are talking to AI, collect as little personal data as possible, use a provider that does not train on your data, and delete old conversations.