Building a WhatsApp FAQ Bot Without Any AI

People hear 'chatbot' and imagine something that needs a data scientist and a monthly bill in dollars. For a tutoring centre, a car workshop or a catering business, you usually need much less. Ten questions come up again and again. A bot that answers those ten, and hands off the rest, is worth building. It does not need any AI.

A keyword-based bot is also predictable. It says what you wrote, every time. That matters when the answer is a fee, a timing or a policy.

Step 1: Mine your real questions

Do not guess. Export or scroll through the last two months of chats and tally what people ask. A workshop will find things like 'service cost', 'pickup available?', 'do you work on Sundays', 'my car is ready?'. Write each question the way customers actually phrase it, including the lazy versions: 'timings', 'timing', 'open?'.

Group these into topics. Aim for eight to twelve. If you have thirty, you are probably splitting hairs or including things that need a human.

Step 2: Write one answer per topic

1. 1Lead with the answer, not a greeting. If the question is about timings, the first line is the timings.
2. 2Keep each reply under about 60 words. Long replies get skimmed, and WhatsApp is not an email client.
3. 3End with a next step: a booking link, a phone number, or 'reply AGENT to talk to a person'.
4. 4Avoid screenshots of price lists. Text is searchable and loads on bad networks.

Step 3: Map keywords carefully

This is where most hand-built bots go wrong. Each answer should respond to several words that mean the same thing: price, rate, cost, charges, fees. Keep a short list per topic. Then check for overlap. If 'service' appears in both 'car service cost' and 'service centre address', you will fire the wrong reply.

Whether your tool matches exact words, partial words or whole phrases changes how you write rules, so test with real customer phrasing. WA Getway supports auto-replies, but wording rules differ from tool to tool, so confirm in whichever one you use how matching actually behaves.

Step 4: Design the dead end

Every bot meets a question it cannot answer. What happens then is the real test. A bad bot repeats 'Sorry, I did not understand' on a loop. A good one says once what it can help with, offers a person, and then goes quiet so your team can step in.

We go through the handover in more detail in bot-to-human handoff on WhatsApp.

Step 5: Test it badly

Nobody types the way you do. Test with typos, short forms, mixed languages, voice notes and a single 'hi'. Ask three family members to message the bot with no instructions and see where they get stuck. Their confusion is free feedback.

Do it from a spare number, not a customer's. Our guide to testing a WhatsApp integration without spamming customers lists safe ways to do that.

What a rule-based bot cannot do

It cannot understand an unusual sentence. It cannot read a photo of a broken part and diagnose it. It cannot negotiate. That is fine. Those conversations belong with your staff, and the bot's job is to clear the pile of repetitive questions so they get to those conversations sooner.

If your questions need live data, like order status or appointment slots, a fixed-text bot will not be enough. That is where a webhook or the REST API comes in: your own system looks up the answer and your gateway sends it. It is a bigger build, and we would only suggest it once the simple version has proven that people use it.

Keep it alive

Once a month, check which questions fell through to a human and add the common ones as new topics. Delete answers that went stale. Compare against the broader rules in our auto-reply setup guide so you do not automate things better left to people.

A tip on tone: write the bot's replies in the same voice as your staff. If your team says 'Hi, happy to help' at the counter, the bot should sound like that too. The goal is that a customer cannot tell where the automated part ended and the human part began, and does not feel cheated either way. When in doubt, shorter and warmer wins over long and formal.

A small FAQ bot is a weekend project, not a transformation. If you want it wired into your own order or booking system, that is something we do on the services side, so ask.