Since 15 January 2026, Meta's terms for the WhatsApp Business Platform have barred AI providers from offering general-purpose assistants on WhatsApp in most countries, while businesses can still use AI to serve their own customers. The line Meta drew is a useful one for any small business: a bot that knows your shop is fine, a bot that will talk about anything is not.
WhatsApp's Business Messaging Policy adds a second rule for automated replies. If you use automation in the 24-hour customer service window, customers must have a prompt, clear and direct way to reach a person, such as a transfer to a staff member, a phone number or a support form.
In short:
- Let a chatbot answer from facts you've approved: prices, hours, policies and free slots.
- Keep complaints, refunds, health questions and exceptions with a person.
- Make the way to reach a human obvious in every conversation.
- Ask a vendor what the bot does when it doesn't know.
The safe jobs are the ones with a right answer you wrote down
Most front-desk questions have one correct answer that already exists somewhere: the price list, the timetable, the cancellation policy, the address and parking note. An AI chatbot is good at finding that answer in a messy question ("how much for colour + cut, long hair, this sat?") and replying in plain English or 中文.
Bookings fit too, as long as the bot reads a real calendar. "Saturday 2pm is free with Aisha, shall I hold it?" is safe when the slot comes from your booking system and the bot can't invent one.
The risky jobs are the ones that need judgement
A chatbot shouldn't decide whether a customer gets a refund, promise a discount you never offered, or tell a parent their child should skip a level. It shouldn't answer health questions either, from "is this rash normal after a facial" to "can I do pilates with a slipped disc". Those need a person who can see the context and own the answer.
The test we use is simple. If a staff member would need to ask the owner before answering, the bot should hand over instead.
A bot that admits what it doesn't know is the safe kind
The failure owners fear is a bot that makes something up: a price that doesn't exist, a promotion that ended, a policy nobody wrote. General-purpose AI models will happily fill gaps. A business chatbot shouldn't.
So the most important behaviour to check is what happens at the edge. A safe bot says "I'm not sure, let me get the team to confirm" and passes the chat on with the question attached, rather than guessing. Ask it something you know isn't in your price list and watch.
Six questions to ask before you buy one
- Where do its answers come from, and can I see and edit that source?
- What does it say when the answer isn't there?
- How does a customer reach a person, and how fast do we hear about it?
- Which topics can I mark as person-only?
- Where are conversations stored, and can I delete a customer's data?
- Can I read every conversation it has had?
A vendor who can't answer the second question clearly is selling a demo, not a front desk.
Test it the way your customers actually type
Many Singapore customers switch between English and Chinese in the same chat, or write "tmr" and "can or not". Before you trust a bot, send it twenty real questions copied from last month's chats, typed exactly as customers sent them. Check every answer against your price list and policies.
Then try to break it. Ask for a discount, a service you don't offer and a time you're closed. A safe bot says it doesn't know, or hands over, rather than agreeing to something you'd have to honour or refuse at the counter. Run the same test again whenever your prices change.
Start with your ten most common questions
Before you talk to any vendor, scroll back through last month's chats and write down the ten questions you answer most. For most salons, studios and tuition centres, that list covers the bulk of the work: price, availability, location, what to bring, how to cancel.
Write the answer you'd want sent for each one. That list is your test. A bot that answers all ten correctly, in your tone, and hands over the eleventh question is ready. Our guide to human handoff covers how to design the hand-over itself.
Frequently asked questions
Can I use an AI chatbot on WhatsApp for my business?
Yes, for serving your own customers. Since 15 January 2026, Meta's WhatsApp Business Platform terms bar AI providers from offering general-purpose assistants, but businesses can still use AI for customer service. Automated replies must come with a clear way to reach a person.
What should a small business chatbot never do?
It should not decide refunds, invent discounts, give health or medical advice, or answer anything a staff member would need to check with the owner. Those chats should go to a person.
How do I stop a chatbot making things up?
Use one that answers only from sources you control, such as your price list and booking calendar, and that says when it doesn't know. Test it with a question that isn't covered before you go live.
How Furoki can help
Furoki Concierge answers only from the prices, policies and slots you approve, in English and 中文. It books and reschedules real appointments, and when a question isn't covered, or is on your person-only list, it says so and hands the chat to your team with the conversation attached. Website chat is live today, and WhatsApp on your own number is rolling out with our first pilot businesses.
Ask the demo salon a question it can't answer and see what it does.
