Here’s a pattern I see in a lot of small businesses. Someone finds your website at 9:30 on a Tuesday night, likes what they see and sends an enquiry. Nobody’s at the desk, so it sits there. By the time you open it at 7am, they’ve also enquired with two other businesses, and one of them got back to them first.
You can’t be on call 24/7, and you shouldn’t have to be. This is a job an AI website chat handles well: answering the common questions, collecting the details and lining up the next step, so the enquiry is warm and ready when you start work. In this post I’ll show you what a good AI website chat should and shouldn’t do, what to feed it, how the handover works next morning, and how to test it before it goes anywhere near a real customer.
What happens to an enquiry that sits until morning
I’m not going to throw a made-up statistic at you here. Just think about how you shop for a service yourself. When you need a plumber, a bookkeeper or a physio, you rarely contact just one. You send a few enquiries, and the business that replies first with something useful has a head start.
After hours is when a lot of people finally get time to sort out their life admin: kids are in bed, work is done, and they’re browsing on the couch. If your website can only offer a contact form and silence until morning, you’re relying on that person still being keen the next day, and still remembering which website was yours.
I covered the daytime side of this in how to reply to every enquiry fast with AI. After-hours chat is the same idea, for the hours when nobody can reply at all.
What an AI website chat should do (and what it shouldn’t)
The biggest mistake I see is giving a chat bot too much freedom. A good AI website chat has a narrow, clear job. Think of it as a friendly front desk person on their first week: helpful, polite, and very clear about what they can and can’t promise.
Jobs to give it
- Answer common questions using your own information: services, service area, opening hours, how bookings work, what to expect at a first visit.
- Collect the details you’d ask for anyway: name, contact details, suburb, what they need and how urgent it is.
- Offer the next step, such as a booking link for a call, or a clear “we’ll be in touch tomorrow morning”.
Jobs to keep away from it
- Quoting prices it hasn’t been given, or estimating a job it can’t see.
- Making promises about availability, timeframes, refunds or results.
- Advice that needs a qualified person: medical, legal, financial or safety questions.
- Handling complaints beyond acknowledging them and passing them to a human.
Why be this strict? Because what your chat says, you’ve said. There’s a well-known case from Canada, Moffatt v. Air Canada (2024 BCCRT 149), where the airline’s website chatbot told a customer he could apply for a bereavement fare after travelling. That wasn’t the airline’s policy. The British Columbia Civil Resolution Tribunal found the airline responsible for what its chatbot said and ordered it to pay the difference. It’s a Canadian decision, not Australian law, but the lesson travels: if your chat makes it up, customers will hold you to it.
Tell people they’re talking to AI
Don’t dress your chat up as a person. Label it clearly as an AI assistant. The Office of the Australian Information Commissioner’s guidance on privacy and the use of commercially available AI products says businesses should make sure public-facing AI tools, such as chatbots, are clearly identified as such, and that privacy policies explain how AI is used. Being upfront also builds trust. Most people are fine chatting with AI at 10pm as long as they know a real person will follow up.
Feeding it your services, area, hours and FAQs
An AI chat is only as good as the information you give it. If you feed it nothing, it will guess, and guessing is exactly what you don’t want. Before you switch anything on, put together a one-page “knowledge sheet” in plain English.
What goes on the knowledge sheet
- Services: what you do and, just as important, what you don’t do
- Service area: the suburbs or regions you cover, and what happens if someone’s outside it
- Hours: when a human will reply, so the chat can set honest expectations
- Pricing rules: only what you’re happy to say publicly, such as “call-out fee from $X” or “we quote after a site visit”
- Your top 10 to 15 questions, with the answers you’d give on the phone
- Hard rules: “never quote a price not listed here”, “never promise a time”, “if it’s an emergency, give this number”
Not sure what your top questions are? Your inbox and your team already know. My post on building a team FAQ from real customer questions with AI walks through pulling them out of old emails and messages, and that same FAQ becomes the brain of your website chat.
A simple instruction block
Whatever chat tool you use, it will have somewhere to put instructions. Here’s an example you can adapt:
You are the AI assistant on [Business]’s website. You are an AI, and you say so if asked. Answer only using the information below. If you don’t know, say a team member will reply during business hours. Never quote a price, timeframe or availability that isn’t listed. Your goal is to help, then collect the person’s name, phone or email, suburb and what they need. Keep replies short, friendly and in Australian English. If someone mentions an emergency, give them [emergency number] straight away.
The handover: how a person picks it up next morning
An after-hours chat is only half the job. The other half is what happens at 7am. If the details land in an inbox nobody checks until lunchtime, you’ve lost most of the benefit.
Make the summary easy to act on
Set the chat up so every conversation ends with a short summary sent to wherever your team actually looks: email, your CRM or a shared inbox. A good summary has:
- Name and contact details
- What they need, in one or two lines
- Urgency and suburb
- Any questions the chat couldn’t answer
- What the customer was told would happen next
Give someone the morning job
Decide who works through overnight enquiries first thing, and by when. Even a simple rule like “every overnight chat gets a personal reply before 9am” makes a huge difference. That reply should come from a human and pick up where the chat left off, so the customer never has to repeat themselves.
If you want the follow-up to happen even when that first reply gets no response, that’s where an Instant Lead Reply and a follow-up routine come in. Both are in my free Steal My AI Agents pack, written as plain-English instructions that work with ChatGPT, Claude, Grokbot or any AI assistant.
Test it like a customer before it goes live
Never switch a chat on without trying to break it first. Spend 30 minutes being your own trickiest customer. Get a team member to have a go too.
Questions to throw at it
- “How much does it cost?” (Does it stick to your pricing rules?)
- “Can you come tomorrow at 8am?” (Does it avoid promising a time?)
- “Do you service [suburb outside your area]?”
- “Is this safe to fix myself?” (Does it avoid advice it shouldn’t give?)
- “Am I talking to a real person?”
- A grumpy complaint about a past job
Every time it gets something wrong, fix the knowledge sheet or the rules, then test again. Keep testing for the first couple of weeks after launch too: read every conversation, note anything odd and tighten it up. It’s also worth checking what customer details the chat tool stores and who can access them. My post on whether AI is safe for small business covers what to look for.
What to do this week
- Check your after-hours gap. Look at when your last 20 enquiries arrived and how long each waited for a reply.
- Write your knowledge sheet. Services, area, hours, pricing rules, top questions and hard rules, on one page.
- Set up the handover. Decide where summaries land and who replies first thing each morning.
- Test it hard. Run through the question list above and fix anything it gets wrong.
- Review weekly. Read the conversations, add new questions to your FAQ and adjust the rules.
Done well, an AI website chat doesn’t replace you. It keeps the conversation going while you’re off the clock, so you start each morning with warm enquiries instead of cold ones. If you’d rather have one built and looked after for you, that’s exactly what we do at WebsiteChat.
FAQ
Can an AI website chat answer enquiries after hours?
Yes. A well set-up AI website chat can answer common questions, collect the customer’s details and offer the next step at any hour. A person then picks up the conversation during business hours.
Will an AI chat give customers the wrong information?
It can if you leave it to guess. Give it a clear knowledge sheet, tell it to answer only from that information, and set hard rules about prices, timeframes and advice. Then test it and read the conversations regularly.
Should I tell customers they’re chatting with AI?
Yes. Label it clearly as an AI assistant. The OAIC’s guidance on commercially available AI products says public-facing AI tools such as chatbots should be clearly identified, and being upfront builds trust.
Can the chat quote prices or book jobs?
Only within rules you set. It can share pricing you’re happy to publish and offer a booking link, but leave custom quotes, availability promises and anything needing judgement to a person.
Do I need a developer to add AI chat to my website?
Not always. Many chat tools can be set up without code, but the knowledge sheet, rules, handover and testing are what make it work. If you’d rather not do it yourself, a free AI audit is a good place to start.
Stop losing the enquiries that arrive after hours
Grab the free Steal My AI Agents pack: ready-to-use agents for lead replies, follow-ups, reviews, onboarding, inbox triage and a weekly business review. Works with ChatGPT, Claude, Grokbot or any AI assistant. Want help working out where your enquiries are slipping through? Book a free AI audit.










