Build a Chatbot From the Questions You Already Get: Turn Old WhatsApp Messages and Emails Into Your Bot’s First Answers (Free Question Grouper)
Build a chatbot the way most guides suggest and you start with a blank screen, guessing what customers might ask. There’s a better starting point sitting in your phone and inbox: the questions people have already asked you. This guide shows how to collect a month of real questions, group them by topic, write one answer per group, and decide which topics to keep for people. A free tool groups your pasted questions and tells you how much a first set of answers would cover.
⚡ In short
- Don’t guess what customers ask. Copy a month of real questions from WhatsApp, email, phone notes and comments.
- Group them by topic. A handful of groups usually covers most questions.
- Write one short answer per group that works for every wording in it.
- Questions about a specific order or payment stay with people. Teach the bot to hand those over.

Why start here
Your customers have already written the brief
Most first-time bots are built from imagination. The owner sits down and writes answers to the questions they think customers ask: “What is your company’s mission?”, “What products do you offer?” Then the bot goes live and real visitors ask “cod hai?”, “kab tak aayega?” and “does XL fit a 40 chest?”, none of which it was taught.
The fix is to start from evidence. If your business has been running for even a few months, you already have hundreds of real customer questions scattered across WhatsApp chats, email threads, Instagram comments, phone call notes and maybe an existing chat widget. Those questions tell you three things no brainstorm can:
- What people actually ask, in the order of how often they ask it.
- The words they use, including typos, Hinglish and local terms, which is exactly what your bot needs to recognise.
- What you already answer well, because your replies to those messages are draft bot answers.
When you build a chatbot from that material, the first version already covers the questions that matter most. You aren’t hoping visitors will ask what you prepared for; you’ve prepared for what they ask.
It also keeps the project small. Instead of writing fifty speculative answers, you’ll usually find that five to eight topics cover most of the questions. Write those well, and you have a useful bot in a day. Everything else can wait until real chats show it’s needed.
If you haven’t decided yet whether to use a ready-made tool or code your own, our make-vs-buy guide helps. The method below works either way.
Five steps
From scattered messages to a working bot
Copy 30 to 100 real questions from the last month or two, from every channel you use.
Remove names, numbers and order IDs. Keep the customer’s exact wording, typos and all.
Sort questions by topic. Use the free tool below or do it in a spreadsheet.
Write one short answer per big group. Decide which groups stay with people.
Ask the bot every original question. Fix misses, then go live and review weekly.
Step 1
To build a chatbot, collect a month of questions from every channel
Set aside an hour. Open every place customers contact you and copy the questions into one plain text file, one question per line. Don’t tidy them yet.
WhatsApp Business. Scroll back through the last month. Copy the first message of each chat; that’s usually the real question. If a customer asked a second, different question later, copy that too.
Email. Search your inbox for the last month of customer emails. Copy the question part, not the whole email.
Instagram, Facebook and Google reviews. Comments like “price?” or “do you ship to Kerala?” count. So do questions inside reviews.
Phone calls. If you or your staff take calls, ask them to jot down each caller’s question for a week. Even rough notes help.
Your existing website chat or contact form. If you have one, it’s the best source, because those visitors were already on your website.
Aim for at least 30 questions. A hundred is better. You’ll notice repeats straight away; keep them all, because how often a question repeats is exactly what you’re measuring.
Don’t filter as you go. The odd, one-off questions are useful too: they show you what the bot should hand over rather than try to answer.
Step 2
Clean the list without losing the customer’s voice
Before you do anything else with the list, remove personal details: names, phone numbers, email addresses, order numbers, addresses. Replace them with a placeholder like [name] or [order] if the question doesn’t make sense without them. You don’t need personal data to build a bot, and keeping it out is good practice under India’s DPDP Act 2023.
Then stop cleaning. It’s tempting to correct spelling and grammar, but don’t. “shiping charges?” and “kab milega” are exactly the wording your bot will see. If you tidy them into “What are your shipping charges?” you’ll write answers that only match tidy questions.
Do split messages that contain two questions. “Do you deliver to Pune and is COD available?” becomes two lines: one about delivery, one about COD. Each will go into a different group.
Finally, strip out messages that aren’t questions at all: thank-yous, greetings on their own, and spam. They don’t need answers.
You should end up with a plain list of real questions, in real words, one per line. That list is the raw material for everything else.
Free tool
Group your questions in one click
Paste your cleaned questions, one per line. The tool groups them by topic, ranks the groups, separates questions that need a person, and estimates how much your first answers would cover.
Paste questions customers have asked you, one per line, from WhatsApp, email, phone notes or chat. 30 or more works best.
Runs only in your browser; nothing is sent or saved. Remove names, numbers and order IDs before pasting.
Step 3
Group by what the customer wants, not by the words
Grouping is where the list turns into a plan. Each group is one thing customers want to know, however they phrase it.
“do you deliver to pune”, “how many days for delivery”, “when will my parcel come” and “shipping charges?” use different words, but they’re all about delivery. “is cod available” and “can i pay cash on delivery” are both about cash on delivery. The tool above groups common shop topics automatically and puts other questions together when they share most of their words. Check its groups and move anything that looks wrong; you know your business better than any tool.
If you prefer a spreadsheet, put each question in column A and type a short topic name in column B. Sort by column B and count.
Then rank the groups by size. In most small businesses a pattern appears quickly: a few groups hold most of the questions, and a long tail of groups have one or two each. In the tool’s example, delivery, returns, cash on delivery, hours, sizing and prices account for most of the list.
That ranking is your build order. Write answers for the biggest groups first, because each answer pays off every time that topic comes up.
Watch for groups that are really two topics. “Delivery” might split into “delivery time” and “delivery charges” if both come up a lot. A separate answer for each is clearer than one long answer covering both.
Step 4
One answer per group, and a decision for each
Now go through the groups in order and make one decision for each: does the bot answer it, or does a person?
The bot answers
General questions with a stable, published answer: delivery times and areas, return rules, COD limits, opening hours, sizing, prices that don’t change daily. For each, write one short answer that works for every question in the group. Read the group’s questions aloud and check the answer fits all of them.
Good bot answers are short (one to three sentences), put the key fact first, use the customers’ own words, and include the one detail people usually ask next. Our guide to writing plain bot messages shows how.
Look at how you already replied in WhatsApp or email. Your best past reply is often 90% of the answer; tidy it and remove anything specific to that one customer.
A person answers
Questions about a specific order, payment, account or complaint need someone to look something up or make a judgement. The tool puts these in “Keep these for your team”. The bot’s job for them is to hand over quickly, or collect a contact when nobody is online. Our handover guide covers the wording.
Not yet
Groups with one question can wait. Revisit them after a month of real chats.
In LiveChatBot, general answers go into the AI Agent’s knowledge, and the same answers can be saved as saved replies so the free keyword bot and your team can use them too. See training your AI chatbot and saved replies.
Example
One group, one answer
Step 5
Test with the original questions, then launch
Your question list now has a second job: it’s your test set. Before going live, ask the bot every question on the list, exactly as the customer wrote it, typos included.
For each one, note whether the bot gave the right answer, the wrong answer, or no answer. Wrong answers are the most important to fix, because a confident wrong reply does more harm than “I’m not sure, let me get someone.” No answer usually means the group’s answer needs more of the customers’ words in it, or extra keywords if you use a keyword bot.
Then test the handover groups. Ask “where is my order” and “my payment failed”. The bot should hand over or collect a contact, not try to answer.
When most questions get the right answer and the rest hand over cleanly, launch. Our launch checklist has 50 checks if you want to be thorough.
After launch, keep the habit going. Every week, look at the chats the bot couldn’t handle, add them to your question list and regroup. Our weekly review guide shows a 30-minute routine and a tool that sorts chats by outcome. Over a few weeks you’ll build a chatbot that covers more and more of what your customers really ask, without ever writing an answer nobody needs.
Checklist
Question-first build
Getting started
Collect your first 30 questions today
Open WhatsApp and your inbox, copy 30 real questions into a text file, remove personal details and paste them into the grouper. Write answers for the top three groups tonight. Tomorrow, add them to your bot, test with the original questions and switch it on.
LiveChatBot’s free plan includes an AI Agent with a small free monthly allowance, a free keyword bot that answers from saved replies, the team inbox, the Android app, Flows, opening hours with an away form and tickets, for two websites with 100 visitors per website each month. See pricing and start at livechatbot.in.
Frequently asked questions
How many questions do I need to build a chatbot this way?
Thirty is enough to see the main topics; a hundred gives a clearer ranking. Collect from every channel customers use, over the last month or two.
Should I correct spelling in the questions?
No. Keep the exact wording, typos and Hinglish included. That’s what your bot will see, and answers written for real wording match better.
What if I don’t have any past questions?
Ask your staff what customers ask most, note every question for two weeks, or start with the common topics for your type of business and refine after launch.
Which questions should the bot not answer?
Anything about a specific order, payment, account or complaint. The bot should hand those over to a person or collect a contact.
Does the grouping tool send my questions anywhere?
No. It runs entirely in your browser and nothing is sent or saved. Still remove personal details before pasting.
Keep reading
Related guides
Bots Chat in Plain Words: A Readability Check for Every Message Your Website Bot Sends (Free Message Checker)
Bot Technologies Compared on One Question: Keyword Rules, Fuzzy Matching, Search and AI Answers (Free Matching Simulator)
Bot Live Chat Weekly Review: Read Last Week’s Chats and Decide What the Bot Learns Next (Free Outcome Tagger)
Start from what customers ask
Turn your top questions into answers the AI Agent can use, and hand the rest to your team. Start free on livechatbot.in.