How to design WhatsApp bot conversations that don't feel robotic
Design WhatsApp bot conversations that feel natural and human with five principles covering personality, formatting, follow-ups, and graceful fallbacks.
Rebecca Pearson6 min read
Most WhatsApp bots feel robotic not because of bad AI, but because of bad design. Rigid menus, generic responses, and zero personality make conversations feel like form-filling rather than genuine interaction. This guide covers five design principles that make bots feel natural — and one test at the end that tells you whether you've succeeded.
TL;DR
- Personality lives in your system prompt, not your platform. A few specific instructions create dramatically more natural conversations.
- WhatsApp is a mobile medium. Short replies, smart formatting, and natural follow-ups respect the medium.
- "I don't understand" moments are design opportunities, not failures. How you handle confusion determines whether users stay or leave.
Why most bots feel robotic
The traditional chatbot was essentially a decision tree. Press 1 for hours. Press 2 for returns. Press 3 to speak to an agent. This worked when that was the only option. Now it feels deeply outdated.
Modern WhatsApp bots powered by large language models don't need decision trees. They understand natural language. But many builders still design them like decision trees — offering numbered menus, refusing to handle unexpected input, and responding in stiff, corporate language that sounds nothing like how a real person talks.
The irony is that the AI can do so much more. It's the design that's holding it back.
Five design principles for natural conversations
Principle 1: give your bot a clear personality via the system prompt
Personality isn't a luxury feature. It's the primary mechanism that determines whether your bot feels like talking to a person or talking to a form.
Your system prompt is where personality lives. Be specific about it. Instead of "be helpful and friendly," try:
"You're Maya, the customer service assistant for Oakwood Interiors. You're warm but efficient — you get to the point without being abrupt. You're interested in helping people find the right piece of furniture for their home, not just processing queries. You use natural language, avoid corporate-speak, and occasionally use light humour when it fits naturally."
The difference in output is substantial. "How can I assist you today?" becomes "Hi! What can I help you find?"
Give your bot:
- A name (it helps users relate to it)
- A specific tone (warm, efficient, playful, professional — pick two or three)
- A genuine purpose (not just "help customers" but why the business cares about their customers)
- Things to avoid (jargon, questions that are too personal too early, corporate filler phrases)
Principle 2: keep replies short and mobile-friendly
WhatsApp is a phone app. Most people use it with one thumb, glancing at their screen between other things. A reply that requires scrolling is already a UX problem.
Target a maximum of three to four sentences per reply for most responses. If you need to convey more information, break it into follow-up messages or use formatted lists.
Here's the before and after:
Before: "Thank you for getting in touch with Oakwood Interiors. We would be delighted to assist you with your enquiry today. Our store located at 14 High Street, Clifton is open Monday to Saturday between the hours of 9am and 6pm, and we also offer Sunday opening from 11am to 4pm. If you would like to speak to a member of our team, you are also welcome to call us on 0117 XXX XXXX during business hours."
After: "We're open Mon–Sat 9am–6pm and Sunday 11am–4pm. Store's at 14 High Street, Clifton. Want our number too?"
The second version is easier to read on a phone screen, more conversational, and ends with a natural follow-up rather than a full stop.
Principle 3: use WhatsApp formatting intelligently
WhatsApp supports a small set of formatting: bold (with asterisks), italic (with underscores), and strikethrough (with tildes). You can also use line breaks, bullet points with dashes, and numbered lists.
These are powerful for readability, particularly for information-dense replies. A list of opening hours, a set of options, or a summary of a booking should be formatted, not delivered as a paragraph.
Formatting tips:
- Use bold for the most important information in a message (a price, a time, a name)
- Use dashes for lists of three or more items
- Leave a blank line between distinct thoughts or sections
- Keep formatted lists to five items maximum — more than that, break into multiple messages
Don't over-format. A simple "We open at 9am" doesn't need bold and bullets. Save formatting for when it genuinely improves clarity.
Principle 4: include natural-feeling follow-ups
A conversation that ends abruptly after answering a question misses an opportunity. Real customer service includes a natural next step.
After answering a question, your bot should move the conversation forward with a relevant follow-up — not a generic "is there anything else I can help you with?" (which feels like a call centre script) but something specific to the context.
Examples:
- After answering a price question: "Would you like to book a slot, or do you want to see what's available first?"
- After sharing opening hours: "We're pretty busy on Saturdays — worth booking in advance. Want me to check availability for you?"
- After handling a complaint: "I've flagged that for the team. They'll be in touch within 24 hours. Is there anything else I can sort for you in the meantime?"
These follow-ups feel natural because they're contextually relevant. They also drive conversions — a customer who asked about your price is often ready to book. Don't make them ask.
Principle 5: handle "I don't understand" gracefully
Every bot has limits. A user will eventually ask something outside its scope, phrase a question in an unexpected way, or send something ambiguous. How the bot handles this moment determines whether they stay in the conversation or give up.
The worst response is an error message or a hard failure ("I'm sorry, I didn't understand your request. Please rephrase."). It's abrupt, unhelpful, and removes the human warmth that makes WhatsApp feel natural.
The best response acknowledges the limitation, offers an alternative, and keeps the tone warm:
"I'm not sure I quite got that one — I'm better at [specific things the bot does]. Could you try rephrasing, or would you rather I get a real person to help?"
Build two to three graceful fallback responses into your system prompt and instruct the bot to use them when it hits the edge of its knowledge. Include a human escalation path in the fallback so users always have a next step.
Testing for robotic-ness: the five-person test
Here's the most effective way to know if your bot passes the naturalness test: give it to five people who didn't help build it.
Don't give them a script. Just tell them to message the bot as they naturally would — the kinds of questions they'd actually ask a real business.
Watch for:
- Moments where they hesitate or seem confused
- Questions the bot answered badly
- Places where the conversation felt stilted
- Responses that prompted an eye-roll
Outsider perspectives catch things you'll always miss. When you've spent hours building a bot, you know exactly how to phrase things to make it work. Real users don't.
Fix the three biggest friction points from the five-person test, then run it again. Repeat until people stop hesitating.
Example system prompt: before and after
Weak system prompt: "You are a customer service chatbot for a restaurant. Answer questions about the menu, opening hours, and reservations. Be polite."
Strong system prompt: "You're Remi, the AI automation assistant for The Grove Kitchen, a relaxed neighbourhood restaurant in Edinburgh. You help with menu questions, table bookings, and general queries. You're friendly and casual — like a knowledgeable friend who works there, not a corporate helpdesk. Keep replies short (two to three sentences) and use the customer's name when you know it. If someone wants to book, check what day and time works for them and how many people, then confirm. If they ask something you can't help with, be honest and offer to pass them to the team."
The second version produces a completely different conversation quality — and takes about five minutes to write.
CodeWords lets you write and iterate your system prompt through Cody, the AI automation assistant. Describe the personality you want and Cody will draft a structured prompt you can refine.