Designing an Arabic Chatbot: Bilingual Conversations in Arabic and English

Designing Right-to-Left Chat Interfaces

Arabic is written right to left, but a bilingual chat window has to handle both directions, often in a single bubble. Getting it wrong produces the familiar bugs: a question mark at the wrong end of a sentence, an order number that looks scrambled, a send button on the wrong side.

For web widgets, the World Wide Web Consortium (W3C) guidance on structural markup and right-to-left text in HTML is the best place to start. In practice, check these points:

• Set the base direction from the conversation language with the dir attribute, not in the stylesheet, and mirror the layout: input box, send button, avatars and timestamps.
• Give each message bubble dir="auto", so an English reply inside an Arabic conversation still reads correctly.
• Isolate opposite-direction text such as order numbers, email addresses, links and product codes, as described in the W3C article on inline markup and bidirectional text.
• Mirror icons that point somewhere, such as back arrows, but not logos, media playback controls or digits. Numbers are still written left to right inside Arabic text.
• Use logical properties in your CSS (cascading style sheets), start and end rather than left and right, so one stylesheet serves both directions.
• Choose a font with clear Arabic letterforms at chat sizes, allow a little more line height, and avoid letter-spacing and capital-letter styling. Arabic has no capitals, and extra spacing can break how letters join.
• Let buttons and quick replies wrap, because Arabic and English labels for the same action rarely have the same length.

On WhatsApp and social inboxes the app handles direction, so the work moves to content: native Arabic templates and quick replies, short button labels, and a check on both iPhone and Android.

Tone, Politeness and Gender-Neutral Phrasing

Courtesy works differently in Arabic customer conversations, and a literal translation of an English script often misses it. A customer who opens with السلام عليكم expects وعليكم السلام in return, not a generic "Hello". Small touches like this tell the customer the bot was designed for them, not converted for them.

• Return greetings in kind. Map common greetings and courtesies to their proper replies in both languages.
• Do not assume gender. Arabic addresses men and women differently: "you want" is تريد to a man and تريدين to a woman. Rephrase around it, for example هل نؤكد الحجز؟ (Shall we confirm the booking?).
• Be warm, not over-familiar. Keep slang and terms of endearment out of replies unless your brand voice genuinely uses them.
• Confirm commitments with buttons. A courteous reply such as "in sha Allah" (God willing) is not a booking or a payment instruction, so end those flows with an explicit Confirm button.
• Plan for the calendar. Prepare Ramadan and Eid greetings, changed opening hours and holiday messages in both languages ahead of time, and have a native speaker review them.

Handover to Bilingual Staff

Even a well-trained bot hands some conversations to people: complaints, unusual requests, anything sensitive. Handover is where many bilingual designs quietly fail, because the bot speaks Arabic but the person who picks up the chat does not. Our chatbot buying guide covers handover triggers in general; for two languages, add these rules:

• Tag every conversation with its language, and route Arabic conversations to agents who read and write Arabic fluently.
• Show the agent the full transcript in the original language, with the details the bot has already captured, so the customer never repeats themselves.
• If no Arabic-speaking agent is on shift, say so honestly in Arabic and offer a call-back, rather than switching to English without asking.
• Give agents saved replies in both languages, written or reviewed by fluent writers, for the situations that come up most.

Testing an Arabic Chatbot with Real Customers

Internal testers tend to write tidy sentences. Customers do not. The most useful test material is your own history: past WhatsApp chats, website inquiries, emails and call notes, with personal details removed.

1. Build a test set from real messages. For each top intent, collect real phrasings in MSA, Gulf dialect, the other dialects your customers use, Arabizi, mixed Arabic and English, and common typos.
2. Include the awkward cases. Neutral replies, Arabic-Indic digits, dual forms such as شخصين (two people) and ليلتين (two nights), voice notes and photos. Decide what the bot does with each, even if the answer is a polite request to type or a handover.
3. Review with native speakers. Ask fluent reviewers, ideally from more than one Arabic-speaking background, to score answers for accuracy, tone and naturalness.
4. Pilot with a limited audience. Open one channel or one section of your website, and read transcripts daily in both languages.
5. Compare the languages. Track fallbacks (messages the bot could not handle), handovers and satisfaction separately for Arabic and English. A gap usually points to missing training phrases or weaker content in one language.
6. Fix, retest and widen. Add the missing phrases and answers, rerun the test set, then open the next channel.

A Worked Example: A Restaurant Booking in Two Languages

Here is how the pieces fit together for a hotel restaurant taking bookings on WhatsApp, a common pattern in hospitality. A guest writes مرحبا، ابي احجز table لشخصين باجر الساعة 8 (Hello, I want to book a table for two tomorrow at 8).

1. Language. The sentence is Arabic with one English word, so the conversation language is Arabic.
2. Details. لشخصين gives a party size of two through the dual form, باجر is Gulf dialect for tomorrow, and "8" still needs a time of day.
3. Clarify, neutrally. The bot replies in MSA, أهلاً بك! هل الحجز الساعة 8 مساءً؟ (Welcome! Is the booking for 8 in the evening?), with two buttons: yes, or another time. The wording suits any guest.
4. Switch on evidence. The guest answers in a full English sentence: "Yes, and can you arrange a birthday cake?" The bot continues in English.
5. Hand over the special request. The bot checks availability and confirms the table, and the cake request goes to the restaurant team with the transcript, booking details and language tag. A bilingual host replies in English.
6. Confirm clearly. The confirmation repeats the date, time and party size in English.

How Star Bit Solutions Can Help

StarBot, our conversational AI chatbot, is built for this mix. It detects Arabic and English in every message, replies in the customer's language and keeps context across turns through its dialogue manager. Its answers, guided flows and handover messages are maintained in both languages from one knowledge base. Intent detection is trained on your own phrases and customer vocabulary, so the way your customers really write becomes training data. Handover routes conversations by topic, language, channel and business hours, with the full transcript passed to the agent.

We start by building the knowledge base with you in Arabic and English, then pilot on one channel, such as your website or WhatsApp, and review transcripts together before scaling up. StarBot can also look up orders in StarBiz or in Odoo through OdooStar, and check calendars for bookings. To see it answer your own questions in both languages, book a free consultation, call +971 55 973 4524 or email [email protected].