Why is a bilingual chatbot harder than a translated one?
Translating an English chatbot into Arabic produces an assistant that is grammatical and stiff. Real customers write differently. They use Modern Standard Arabic in formal settings, Gulf dialect in everyday messages, English in the same sentence, and Arabic typed in Latin letters with numbers standing in for letters, often called Arabizi.
A good system understands all of these, replies in the register the customer uses, and keeps a tone that matches your brand. That takes design and testing, not just a translation step.
Which Arabic should the assistant speak?
| Style | Where it appears | Design implication |
|---|---|---|
| Modern Standard Arabic | Formal notices, contracts and official correspondence | Use for documents, policies and formal replies |
| Gulf dialect | Everyday customer messages and voice notes | Understand it reliably and reply in a clear, neutral register |
| English | Business, expatriate customers and technical terms | Treat as a first-class language, not a fallback |
| Mixed Arabic and English | Common inside a single message | Test explicitly and reply in the language last used |
| Arabic in Latin letters | Chat and social channels | Understand it and usually reply in Arabic or English script |
Dialects from outside the Gulf matter too, since customers and staff come from across the region.
How should WhatsApp fit in?
Many customers would rather message a business than phone it or fill in a form, so WhatsApp is often the first channel to build. Use the official WhatsApp Business Platform, which means approved message templates for business-initiated messages, clear opt-in and a human handover. Our guide to building a WhatsApp AI chatbot for customer support covers the steps.
Voice notes are common in messaging, so plan early for speech-to-text in both Arabic and English, and decide what the assistant does when a transcript is uncertain.
What does right-to-left design involve?
- Layout mirroring: navigation, progress indicators and directional icons flip, while some elements, such as phone numbers and logos, do not.
- Mixed text: Arabic sentences containing English names, numbers or links need correct bidirectional handling, or they display scrambled.
- Typography: choose a typeface with strong Arabic support and test sizes and line height, because Arabic script needs more vertical room than Latin.
- Numerals and dates: agree whether to show Western or Eastern Arabic numerals and which date format suits each audience.
- Real devices: many display faults appear only on phones, so test there.
Test the whole flow in Arabic from the first prototype, not at the end. Layout faults found late are slow to fix, and they are the first thing a native speaker notices.
Which documents can document AI read in Arabic?
Document AI can read trade licences, tenancy contracts, invoices, ID copies and delivery notes in Arabic and English, including bilingual forms. Accuracy depends on scan quality, stamps and handwriting, so each extracted field carries a confidence score and uncertain ones go to a reviewer. You can read more on our document AI and automation page.
Which UAE data rules should you plan for?
This is general information, so ask your counsel. The federal baseline is Federal Decree-Law No. 45 of 2021 on the protection of personal data. DIFC and ADGM, the financial free zones, have their own data protection laws and regulators. Some sectors, including health and financial services, have additional rules that can affect where data may be stored and who may process it. Settle these questions early.
- Which regime applies to your entity and to the people whose data you handle?
- Where will prompts, outputs and logs be processed and stored, and does any rule require them to stay in the UAE?
- Which model provider and region will be used, and what does that provider retain?
- What notice do customers see when they message the assistant?
- How can a customer reach a person, and how are requests about their data handled?
How do you test quality in two languages?
- Collect real messages and documents in every language mix your customers use.
- Build a test set with expected answers, grouped by language, dialect and topic.
- Use native reviewers, from your team or nominated by you, to grade tone and politeness as well as facts.
- Report results per language, so Arabic quality is never hidden inside an average.
- Route uncertainty to people, with a summary in the language of the conversation.
- Review failed conversations weekly and add them to the test set.
What should the assistant do when it is not sure?
Uncertainty handling matters more in a bilingual setting, because an assistant that guesses in Arabic can sound confident and be wrong. Set a confidence threshold per language, and let the assistant ask a short clarifying question before it commits to an answer.
For anything involving money, contracts, health or legal consequences, the assistant should state what it can do and hand over to a person with a summary, in the language of the conversation, so the customer never has to repeat the story. Keep handover hours explicit: if staff are offline, say when someone will reply and capture the details.
Which sectors benefit first?
- Real estate: lead qualification, viewing bookings and tenancy document checks.
- Logistics: shipment status, customs paperwork and driver messaging.
- Retail and e-commerce: order help, returns and product questions.
- Hospitality: booking enquiries and concierge requests in the guest's language.
Where to start
Start with one journey, such as booking enquiries or order status, in both languages and on WhatsApp. Our page on AI development for UAE businesses explains the team and the working rhythm, and you can try the live support agent demo or contact us.