Can a chatbot understand Manglish and bahasa rojak?
Test it with the messages your customers really send, not a perfect sentence.
Read inBahasa Melayu中文
Yes, a good AI chatbot can understand Manglish, bahasa rojak and messages that mix Chinese with English, but plenty still can't. The only way to know is to test it with the messages your own customers send: short, mixed, misspelt and full of "lah", "boleh tak" and "can or not". This guide gives you a 10-message test.
Why do Malaysian customers write in mixed language?
Because that is how Malaysians talk. Switching between English, Malay, Chinese and Tamil inside one sentence is normal, and on WhatsApp people write the way they speak. A customer may open in English, ask the price in Malay and end with a Chinese phrase, all in one chat.
The mix is real for businesses too. In a sample of Mampu AI conversations where the customer's language could be detected, 41.7% were in English, 32.5% in Malay and 25.8% in Chinese. A chatbot that only works well in formal English misses more than half of a typical Malaysian SME's customers, before you even count the messages that mix all three.
Customers also expect quick answers on messaging. In HubSpot's research, 82% of people said an immediate reply was important or very important when they had a sales or marketing question. A chatbot that cannot read "got stock or not?" either replies with something useless or hands the chat to a person, and the speed advantage is gone.
Why do many chatbots get Manglish wrong?
Most problems come from how the bot was built, not from the customer's spelling.
- Keyword bots look for exact words. If the menu expects "price", a customer who writes "how much ah" or "berapa harga" falls through to "Sorry, I don't understand".
- Short forms and spelling vary. "tmr", "tmrw" and "esok" all mean tomorrow; "tq", "anot", "boleh tak", "can or not" are everyday questions a script never lists.
- Particles change the meaning. "Can lah" agrees, "can meh?" doubts, "cannot lah" refuses. A bot that ignores them can answer the opposite of what was asked.
- Languages switch mid-sentence. "Boleh book for tomorrow petang tak?" carries a booking request across two languages. Translating word by word breaks it.
- Two scripts in one line. "可以包 parking 吗？" mixes Chinese characters and an English word. Some systems treat it as broken text.
- Replies come out too formal. Even when the bot understands, a stiff textbook answer to "hi boss" feels like talking to a form.
What does good Manglish handling look like?
A chatbot handles mixed language well when it does five things:
- Reads the meaning, not the words. "Parking got anot?" is a question about parking, and the answer should be about parking.
- Answers the question that was asked. Not a menu, not a generic greeting.
- Keeps a natural tone. Friendly and clear, in a language the customer is comfortable with.
- Handles voice notes. Many customers would rather speak than type, especially when mixing languages.
- Knows when to ask or hand over. If a message is unclear, it asks one short question; if the customer wants something only a person can decide, it passes the chat on.
How can you test a chatbot before you buy?
Ask the vendor for a trial or a demo, and send these messages yourself. Use your own products and prices so the answers can be checked.
| Message to send | What it tests | A good answer |
|---|---|---|
| hi boss, got stock or not ah? | Manglish greeting and question | Answers about stock, not a menu |
| how much one? | Very short price question | Gives a price or asks which item |
| Bos, harga untuk 2 kotak berapa? | Malay with quantity | Price for two boxes, in Malay |
| Boleh book for tomorrow petang tak? like 3pm | Malay-English rojak booking | Offers or confirms 3 pm tomorrow |
| 这个还有货吗？ | Chinese | Answers in Chinese |
| 可以包 parking 吗？ | Chinese with an English word | Answers the parking question |
| tmrw can deliver to Penang? | Short forms | Understands tomorrow and the place |
| can meh? | Doubt particle | Explains, does not just repeat "yes" |
| (a voice note in Malay) | Spoken message | Understands what was said |
| want to talk to real person | Hand-over | Passes the chat to your team |
Score each answer simply: did it understand, did it answer, did it sound natural. If a vendor's bot fails three or more, expect your customers to notice too.
How should you set up a chatbot for mixed-language customers?
- Write your main flow once. A good AI agent replies in the customer's language without three copies of every step.
- Add your own wording where it matters. Prices, promotions and policies are worth writing yourself in Malay and Chinese, so the meaning never shifts.
- Keep product names the same everywhere. Customers search for the name they saw in your ad.
- Test with real past chats. Paste in messages your team actually received last month, not invented ones.
- Review a few chats every week. Look for the places where customers repeated themselves or gave up.
How does Mampu AI handle Manglish and bahasa rojak?
Mampu AI flow agents read Manglish, bahasa rojak and Chinese-English mixed messages and reply in the customer's language, even when your flow is written in one language. In a 3 November 2026 test on our demo workspace, the agent understood every Manglish, Malay-English and Chinese-English message we sent, including "parking got anot" and "可以包 parking 吗？".
- The agent detects the customer's language and switches when the customer switches.
- You can write your own version of any message in Malay or Chinese; where you don't, the agent translates your main message.
- Voice notes are transcribed and answered like typed messages; photos and PDFs are read too.
- You decide in the flow when a chat goes to a person.
- One inbox covers WhatsApp, Instagram, Facebook Messenger, Telegram and TikTok.
To see how it reads your own customers' messages, send our team a few real ones; current plans are on our pricing page.
Frequently asked questions
Sources
Every rule and figure on this page is dated and linked to its source. Mampu AI figures are anonymised percentages across customers on Mampu AI.
See your own customers' messages answered
Message our team with a few real messages your customers sent. We'll show you how an AI agent reads and answers them.
