Test Your WhatsApp Chatbot Before Customers Do
- 08 Oct, 2026
Test Your WhatsApp Chatbot Before Customers Do
Chatbots rarely fail on the greeting. They fail three messages in, when a real person types something the flow did not expect — and the bot repeats its menu for the fourth time while the customer leaves.
Run this list before launch. It takes an hour and prevents the complaints that make a business switch the bot off a week later.
The basics (5)
- Send "hi" from a number that has never messaged you. Does the greeting arrive within seconds?
- Send "hi" again from the same number an hour later. Does it repeat the full greeting like you are a stranger?
- Does the first message say what the bot can do, in one line, without a wall of options?
- Is there a visible way to reach a human in the first reply?
- Does your business name appear, so people know who they are talking to?
The ways real people type (6)
- Hinglish: "price kya hai". Does it understand, or fall back to the menu?
- Devanagari: "कीमत क्या है". Many bots silently fail here — see the Hindi and regional languages guide.
- Typos: "pricee", "delevery". People type fast on phones.
- Two questions at once: "price and delivery time?" Does it answer one and drop the other?
- Just an emoji or a single "?". What happens?
- A voice note. Does it respond sensibly instead of ignoring it? Indian customers send voice notes constantly.
The dead ends (5)
- Type nonsense three times. Does the bot offer a human by the second failure, or loop forever?
- Ask something genuinely outside its scope. Does it admit that plainly?
- Ask for a human. Does the handover actually happen, with the conversation visible to the agent?
- Ask for a human outside business hours. Is the answer honest about when someone will reply?
- Abandon the chat mid-flow, return two days later. Does it resume sensibly or restart from zero?
The ones that cost money (4)
- Send a message after 24 hours of silence. Confirm your system uses a template, not a free-form reply that will fail.
- Check every link the bot sends. Broken payment or catalogue links in an automated flow break silently, at scale.
- Trigger the same event twice and confirm the customer does not get two identical messages.
- Test "STOP". Does it actually stop, immediately, with no further messages?
Two tests people skip
- Test on a slow connection. Media-heavy replies that work on office wifi time out on 4G in a smaller town.
- Have someone outside the project test it without instructions. You cannot un-know how your own flow works; they will find the dead end in four minutes.
After launch, watch three numbers
- Handover rate — what share reach a human. Rising fast means the bot is failing, not working.
- Drop-off point — the step where people stop replying. That is your next fix.
- Repeat questions — what people ask that the bot cannot answer. That is your next flow.
Review those weekly for the first month. A bot launched and left alone gets worse, because your business changes and the script does not.
Frequently Asked Questions
What is the most common WhatsApp chatbot failure?
Looping. The customer types something unexpected, the bot repeats its menu, and the customer leaves. Any bot should offer a human by the second failed attempt rather than cycling through the same options.
Does a WhatsApp chatbot need to handle Hinglish?
In India, yes. A large share of customers type in transliterated Hindi or a mix of English and their own language. Test with phrases like "price kya hai" and with Devanagari text before launching, because many bots fall back to a menu for both.
What should we check in the first month after launch?
Three things weekly — the share of chats handed to a human, the step where people stop replying, and the questions the bot cannot answer. Each one points directly at what to fix next.
ZupiChat is developed by the team at Codelith Lab, a Pune software company specialising in websites, apps, SEO and AI automation.