WhatsApp FAQ Chatbot: How to Set One Up, Step by Step

A WhatsApp FAQ chatbot answers the questions your team answers every day, such as timings, prices, delivery areas and return rules, the moment a customer asks, and hands everything else to a person. To build one that helps: count the questions customers really send, keep a few topics human, write short answers that fit WhatsApp’s buttons and lists, choose between a menu and AI, and give every customer an easy way to reach your team.

This guide is for businesses whose staff type the same answers dozens of times a day: shops, clinics, coaching institutes, restaurants, salons, travel agents and service companies. It focuses on the FAQ job specifically. Limits, prices and policies are checked against Meta’s documentation as of October 2026.

Using ChatMitra? Welcome messages work on every plan. Keyword rules with buttons, lists and documents, and an AI chatbot that answers from the FAQs you give it, are on Pro, alongside a shared inbox where your team takes over. Jump to building it in ChatMitra.

What an FAQ chatbot does, and what it shouldn’t

An FAQ chatbot has one job: give a correct, approved answer to a common question faster than a person could. It’s not a salesperson, a complaints desk or a general assistant. Keeping it narrow is what makes it reliable.

There are three ways to build one:

1. Menu and keyword bot. Customers tap buttons or a list, or type a keyword, and get a fixed answer. Predictable and cheap, but it can’t understand a question it wasn’t set up for.
2. AI bot answering from your FAQ document. A language model reads the customer’s message and answers using the information you give it. It copes with rephrasing, spelling mistakes and mixed languages, but it needs grounding rules and testing so it doesn’t make things up.
3. Hybrid. A menu for the most common questions, AI for typed questions, and a person behind both. This is where most businesses end up.

Step 1: Find the questions customers really ask

Don’t start from a brainstorm. Teams consistently guess wrong about which questions matter most. Count them instead.

Where to look:

• Your last 30 days of WhatsApp chats. Read the first message of each conversation. That’s where the question usually is.
• The quick replies your team already uses. If someone saved an answer, it’s a frequent question.
• Phone calls. Ask whoever answers the phone to tally questions for a week.
• Google reviews and your Business Profile’s questions, especially complaints about not getting an answer.
• Your website’s contact form and search box.

How to count: make a sheet with one row per conversation and two columns, the question in the customer’s words and a short topic label (“delivery area”, “fees”, “timings”). After 200 to 300 conversations, sort by topic. You’ll usually find that a small number of topics covers most of the volume. For a dental clinic, a month’s tally might look like this (illustrative):

Keep the “example wording” column. Those phrases become your keywords, your test cases and the variations you’ll give an AI bot.

Step 2: Decide what stays with people

A bot that answers the wrong question confidently does more damage than no bot. Before you write a single answer, list the topics the bot should hand over straight away.

Meta also requires this. Its Business Messaging Policy says: “You may use automation when responding during the 24-hour window, but must also have available prompt, clear, and direct escalation paths.” It lists an in-chat transfer to a human, a phone number, email, web support, a store visit or a support form (WhatsApp Business Messaging Policy). The easiest way to comply is a “Talk to our team” option in every menu, and a reply that says how and when someone will respond.

Step 3: Choose how to build it

Your choice depends on how many questions you have, how differently customers phrase them, and how much testing time you can give.

About the free app. Quick replies hold up to 50 saved answers, which your team sends by typing ”/” (WhatsApp Help Center). Greeting and away messages cover first contact and closed hours. Meta’s Business AI in the app, launched in India in May 2026, answers “based on what the business has already added to their profile and catalog” (Meta Newsroom), but WhatsApp says it’s “only available in limited countries to select businesses” (WhatsApp Help Center), and greeting messages are switched off when it’s connected.

About Meta’s own agent on the API. Meta Business Agent is billed at $2 per million tokens, which Meta estimates at “approximately 4–5 cents (USD) per message,” with no free messages (Meta pricing). Meta excludes finance, government, health, alcohol, gambling, over-the-counter drugs and matrimony services (Meta Business Agent overview). For a plain FAQ job, a menu or your own AI model answering from your FAQ document usually costs far less.

About AI rules. Since 15 January 2026, Meta bars general-purpose AI assistants from the WhatsApp Business Platform, except where it’s legally required to permit them. An AI bot that answers questions about your own business is “incidental or ancillary” use and is allowed (Meta Terms for WhatsApp Business Platform, §4.7). Keep your bot on your business’s topics.

Step 4: Write answers that work in a chat

A website FAQ answer pasted into WhatsApp reads like a wall. Rewrite every answer for a phone screen:

1. Answer in the first line. “Yes, we’re open on Sundays, 10 am to 2 pm.” Then add one line of detail if it’s needed.
2. Use numbers, not adjectives. “Delivery in 2–3 days in Pune, 4–6 days elsewhere” beats “fast delivery across India.”
3. One action per answer. A single button or link: “Book a slot”, “See the price list”, “Track my order”.
4. Date anything that changes. “Fees for the 2027–28 batch” tells the reader, and your team, when it goes stale.
5. Write the way customers write. If they ask in Hinglish, a short Hinglish answer often lands better than formal English.
6. End with a way out. “Anything else? Tap Main menu or type AGENT.”

Before and after:

Before: “Thank you for reaching out! We are pleased to inform you that our store is operational on all days of the week including Sundays, subject to public holidays, during which timings may vary. Please feel free to contact us for further assistance.”

After: “Yes, we’re open every day, 10 am–9 pm. On public holidays we close at 6 pm. 📍 Tap below for directions.” [Get directions]

Meta’s limits for menus and answers

Buttons, lists and link buttons are interactive messages, and Meta caps them:

Sources: Meta’s pages on reply buttons, list messages and link buttons.

These are free-form “service” messages, so you can send them only during an open customer service window: the 24 hours after a customer messages or calls you, reset each time they do (Meta, service messages). That suits an FAQ bot, which only ever replies. To reach someone after the window closes, for example to follow up on an unanswered question the next day, you need an approved template.

The 24-character row title is the limit that bites most often. “Consultation fee and payment options” won’t fit; “Consultation fee” will, with “Cash, UPI, cards accepted” in the 72-character description. The interactive messages guide shows each format.

Step 5: Design the menu

A menu should get anyone to their answer in two taps or fewer.

• Top level: a list, not buttons, if you have more than three topics. A list holds up to 10 rows, so put your top 8 or 9 questions there and keep the last row for “Talk to our team”.
• Second level: the answer itself, with one button for the next likely step (“Book a slot”) and one back to the main menu.
• Order rows by how often they’re asked, using your count from Step 1, not by your org chart.
• Name rows the way customers ask: “Fees” rather than “Pricing structure”, “Open Sunday?” rather than “Operating hours”.

Two Meta features help customers start. On the WhatsApp Business Platform, you can set up to 4 ice breakers, tappable starter questions of up to 80 characters shown when someone opens a chat with you for the first time, and up to 30 commands that customers see by typing ”/” (Meta, conversational components). Ice breakers are a natural fit for your four most common questions; the conversational components explainer shows how they look. They aren’t shown when a customer arrives through a link with a pre-filled message.

A sample menu for a coaching institute:

Hi! I’m the Vidya Classes assistant. What would you like to know? [View topics]

1. Courses and batches
2. Fees and instalments
3. Batch timings
4. Demo class
5. Admission documents
6. Online or offline?
7. Scholarship test
8. Centre address
9. Talk to our team

Each row leads to a two- or three-line answer with one next step. “Fees and instalments”, for instance, gives the fee for each course and the instalment rule, with a button for “Book a demo class”.

Step 6: If you use AI, write an FAQ source document

An AI bot is only as good as the information it’s given and the rules it follows. Give it a single FAQ source document: short, factual and dated.

Format each entry the same way:

Q: Do you open on Sundays?
A: Yes. Sundays 10 am to 2 pm. Closed on Diwali and Holi.
Also asked as: open sunday, sunday timing, ravivar ko khula hai
Last checked: 1 Oct 2027

Then add grounding rules at the top of the bot’s instructions:

You are the WhatsApp assistant for Sharma Dental Clinic, Pune.
Answer only from the FAQ below. If the answer is not in the FAQ, say:
"I'm not sure about that. Type AGENT and our team will reply between 9 am and 8 pm."
Never guess prices, dates or medical advice.
For pain, swelling or bleeding, give the clinic number +91-XXXXX-XXXXX and say a person will call back.
Reply in the customer's language. Keep replies under 60 words.

Rules that matter most:

• “Answer only from the FAQ” is the single most important line. Without it, models fill gaps with plausible-sounding invention.
• Give an exact out-of-scope reply, so the bot hands over the same way every time.
• Keep the document short. Every message the AI processes includes your instructions, so a bloated document costs more tokens per reply and makes answers vaguer. A few hundred words per topic is plenty.
• Include the “also asked as” phrasings from your Step 1 count, including Hinglish and common misspellings.

Test before launch with at least 30 real messages from your chat history: the top questions in several phrasings, a few in Hindi or Hinglish, and some deliberately out of scope (“can you write my essay?”, “what’s the weather?”). The bot should answer the first group correctly and hand over or decline on the rest. Our AI prompts for WhatsApp chatbots include support and FAQ prompts you can adapt.

Step 7: Plan the fallback and the handover

Every FAQ bot meets questions it can’t answer. What happens next decides whether customers trust it.

A good fallback reply:

• admits the bot doesn’t know;
• says what happens next, and when (“Our team replies between 9 am and 8 pm, usually within an hour”);
• keeps the customer’s question so they don’t have to type it again.

Two misses, then a person. If a customer hits the fallback twice, or types “agent”, “human” or “talk to someone”, stop the bot for that chat and alert your team.

Out of hours, say so plainly: “We’re closed now. We’ll reply at 9 am. Meanwhile, here’s our menu for common questions.” Don’t pretend a person is coming at 2 am.

When the person takes over, they should see the whole conversation, including the menu taps. The customer should never have to repeat their question. The WhatsApp customer service guide covers what good human follow-up looks like.

Step 8: Test before you switch it on

Run through this checklist with someone who didn’t build the bot:

1. Every menu row opens the right answer, and every answer has a way back.
2. Every button and row label fits Meta’s character limits and reads well on a small phone.
3. Each of your top 20 questions, typed in three different ways, gets the right answer or a clean handover.
4. “Agent”, “human”, “talk to someone” and the local-language equivalents reach a person.
5. Out-of-scope questions get the fallback, not an invented answer.
6. Prices, timings and dates match your website and your shop’s signboard.
7. Links open, and any documents, such as a price list or brochure, are the current versions.
8. Out-of-hours messages get the closed-hours reply.
9. No message arrives twice. If you run keyword rules and an AI bot together, test the overlap carefully.
10. A team member can see the conversation and take over in the inbox.

Step 9: Measure, then improve every week

Track a few numbers you can actually collect:

Spend 20 minutes a week reading the chats where the bot fell back or handed over. Add an answer only when the same new question keeps coming back. Rewrite answers that cause repeat questions.

Keep your answers current

The fastest way to lose trust is an outdated answer: last year’s fees, a holiday schedule that’s passed, a delivery promise you’ve stopped making.

• Give each answer an owner, the person who knows when it changes: the accountant for fees, the store manager for timings.
• Add a “last checked” date to each entry, and review anything older than three months.
• Set calendar reminders for known changes: new academic year, festival hours, price revisions, monsoon delivery delays.
• Change the bot the same day you change your website or signboard. Make it one step in the same checklist.
• Keep seasonal answers separate, such as Diwali hours or a sale’s terms, and remove them when the season ends.

Ready-made FAQ sets by business type

Use these as a starting point, then replace them with your own counts from Step 1.

What a WhatsApp FAQ chatbot costs

Since 1 October 2026, Meta has charged for service messages, which are free-form replies inside the 24-hour window, whether sent by a person or a bot. Every business number gets 1,000 free service messages a month, and Meta charges from the 1,001st (Meta pricing). In India, each one after that costs ₹0.115 on Meta’s rate card effective 1 October 2026. Your platform and, for an AI bot, your AI provider charge separately.

A worked month. A coaching institute’s FAQ bot handles 3,000 conversations a month in India. Each conversation averages four bot messages: the menu, an answer, a follow-up answer and a closing message.

If the bot uses AI, add the token cost from your own AI provider’s account, which depends on the model and the length of your instructions. Two ways to bring the Meta line down:

• Send fewer, better messages. If each conversation drops from four messages to three, the Meta line falls to (9,000 − 1,000) × ₹0.115 = ₹920. That’s another reason to put the answer in the first message and skip the “Thanks for your question!” filler.
• Answer in one message instead of a chain. One message with the answer and a button beats three short bubbles.

ChatMitra’s fee is ₹0.20 per conversation, meaning once per customer per 24 hours however many messages are exchanged, not per message. Enterprise, at ₹2,499 a month, has no per-conversation fee (ChatMitra pricing).

Building an FAQ chatbot in ChatMitra

ChatMitra connects your number to the WhatsApp Business Platform and gives you three building blocks for an FAQ bot.

1. Welcome message (every plan). A welcome message can send up to 20 messages in reply to a customer’s first message, such as a greeting and your FAQ list, and re-welcome returning customers after a number of days you choose (0–365). It can also tag new contacts.

2. Auto Reply rules (Pro). These are the menu and keyword bot.

• Each rule matches up to 10 keywords as an exact match, starts with, or a regex pattern. Regex is case-insensitive, so an unanchored pattern such as refund|return|exchange catches those words anywhere in a message.
• A rule can reply with text, images, documents, video, location, contacts, buttons (up to 3) or a list (up to 10 rows), products or your catalog, within Meta’s limits above.
• Button and list taps are matched by their title, so make each rule’s keyword the exact row or button title. That’s how you chain a menu: the main list’s “Fees” row triggers the “Fees” rule, which replies with the answer and a button.
• A rule can add a tag to the contact. Tag each FAQ rule, and you can count which questions are asked most, which ChatMitra doesn’t otherwise report.
• An Away message covers one daily closed-hours window with your time zone; see these away message examples.

3. AI chatbot (Pro). It answers typed questions using the AI provider and API key you choose, from 18 supported providers. Paste your FAQ source document and grounding rules into the bot’s instructions (or its company description). You can set how many recent messages it remembers (1 to 100, 20 by default) and for how long a conversation stays in memory (30 minutes to 24 hours). It reads typed text only, not button taps or images.

Then add your team. Everyone answers from one shared inbox on the web or Android app. To take over a chat, pause automation for that customer from the inbox, for 30 minutes to 24 hours or until you turn it back on. A pause stops both the rules and the AI for that customer.

A setup that works:

1. Welcome message: one greeting line plus your FAQ list of up to 10 rows.
2. One Auto Reply rule per row, keyed to the exact row title, each answer ending with a button to the main menu or “Talk to our team”.
3. Either AI for typed questions, or keyword rules for the most common typed words, but think carefully before running both (see below).
4. A “Talk to our team” rule that adds a “Needs agent” tag and tells the customer when someone will reply. Your team filters the inbox by that tag and pauses automation when they step in.

ChatMitra’s advantages for FAQ bots:

Good to know:

• Auto Reply rules have no “didn’t understand” reply. A typed message that matches no rule gets no automatic answer unless the AI chatbot is on, so make sure your team watches unanswered chats.
• If keyword rules and the AI chatbot are both on, a typed message that matches a keyword gets two replies. Key rules to exact menu titles when AI handles typed questions.
• A Fixed Reply answers every message and switches off the rules, so it isn’t an FAQ tool.
• Handover is manual: there’s no automatic pause after repeated misses and no chat assignment, so agree who watches the “Needs agent” tag.
• Bot analytics, such as top questions or unanswered questions, aren’t shown in the app; use rule tags and a weekly read of the chats.

Common mistakes

• Building from a brainstorm instead of a count. You’ll automate the questions you think matter, not the ones customers send.
• Answering everything. Complaints, refunds and anything personal belong with people.
• Website-length answers. If an answer needs scrolling on a phone, rewrite it.
• Menus three levels deep. Two taps to any answer, or customers give up and type.
• No way out. Every menu needs “Talk to our team”, and Meta’s policy requires a clear escalation path.
• An ungrounded AI bot. Without “answer only from the FAQ”, it will invent prices and policies.
• Two systems answering the same message. Decide which one owns typed questions.
• Stale answers. Old fees and holiday hours do more harm than no answer.
• Never reading the fallbacks. The questions your bot can’t answer are your to-do list.