A WhatsApp bot that answers in Saudi and Gulf Arabic
Our WhatsApp front desk is an AI chatbot for small businesses. Saudi and Gulf Arabic is the dialect it is set up around, and it is part of our regular test suite. This page shows the real customer messages in that suite, what the assistant does with them, and what we have not tested yet.
The six checks that use Gulf customer messages in our regular test suite passed in the run of 7 October 2026. We have no 30-message score for Gulf yet.
Real messages and replies
These are the Gulf customer messages in our regular test suite, in the order a customer might meet them. Where the suite recorded the assistant’s reply, we show it exactly as written; for the others we describe what the assistant does.
The tests run through the website chat of a sample restaurant and a sample clinic that we made up in Kuala Lumpur, which is why some replies mention ringgit prices or ask for a WhatsApp number. On WhatsApp the assistant already has the customer’s number. Replies are shown exactly as the assistant wrote them. Test phone numbers are shortened.
Regular test suiteSample restaurantPassed
Customer: هلا، ابي احجز طاولة لأربعة الليلة
In English: Hello, I want to book a table for four tonight. (ابي means I want.)
Front desk: هلا والله! حياكم، الليلة الأربعاء نفتح من 12:00 الظهر إلى 11:30 بالليل. الساعة كم تبون؟ وعلى أي اسم؟ وإذا أمكن رقم واتساب عشان نتواصل معكم.
In English: Welcome! Tonight, Wednesday, we are open from 12:00 noon to 11:30 at night. What time would you like? And under what name? And if possible a WhatsApp number so we can contact you.
Regular test suiteSample restaurantPassed
Customer: hala, abi a7jez tawla l 4 ashkhas bokra
In English: Hello, I want to book a table for 4 people tomorrow. (Arabic in Latin letters: 7 stands for ح.)
What the assistant doesIt answers in Latin letters too, in Gulf spelling («hala wallah! 7ayakum»), asks for the time and the name, and keeps «4» and the clock times clear of the words. The test fails if the reply contains any Arabic script.
Regular test suiteSample restaurantPassed
Voice note, as text
Customer: أبغى أحجز طاولة لأربعة بكرة الساعة ثمانية، الاسم خالد، رقمي +601234567xx
In English: I want to book a table for four tomorrow at eight, the name is Khaled, my number is … (A voice note, shown as the text the machine wrote.)
Front desk: أبشر! فهمت إنك تبغى طاولة لأربعة بكرة الخميس 8 أكتوبر الساعة 8:00 مساءً، على اسم خالد ورقمك +601234567xx، صح؟ إذا أكدت لي أحجزها لك.
In English: Gladly! I understood that you want a table for four tomorrow, Thursday 8 October at 8:00 pm, under the name Khaled, and your number is …, right? If you confirm, I’ll book it for you.
Nothing is booked at this point. The test fails if the assistant books before the customer says yes.
Regular test suiteSample restaurantPassed
Voice note, as text
Customer: أبغى أحجز طاولة لأربعة بكرة الساعة ثمانية، الاسم خالد، رقمي +601234567xx
Before this, the assistant says back what it understood and waits, as in the example above.
Customer: أيوه صح
In English: The same voice note, then the customer answers: yes, that’s right.
Front desk: أبشروا، حجزكم تأكد! ✅
الخميس 8 أكتوبر، الساعة 8:00 مساءً، لأربعة أشخاص، على اسم خالد.
الطاولة تنحفظ لكم 15 دقيقة بعد الموعد.
وهذا مطعم تجريبي (ديمو)، يعني ما انحجز شي فعلي.
In English: Consider it done, your booking is confirmed! Thursday 8 October, 8:00 pm, for four people, under the name Khaled. The table is held for you for 15 minutes after the time. And this is a demo restaurant, so nothing was actually booked.
Regular test suiteSample restaurantPassed
Voice note, as text
Customer: أبغى أحجز طاولة لأربعة بكرة الساعة ثمانية، الاسم خالد، رقمي +601234567xx
Customer: لا، لخمسة مو لأربعة
In English: The same voice note, then the customer corrects it: no, for five, not four.
What the assistant doesIt takes the correction and books for five. It never goes ahead with four. The test reads the reply to check this.
Regular test suiteSample restaurantPassed
Voice note, as text
Customer: وين موقعكم وفيه مواقف قريبة؟
In English: Where are you, and is there parking nearby? (A voice note, shown as text.)
What the assistant doesIt answers straight away: Jalan Alor in Bukit Bintang, a five-minute walk from the MRT and Monorail stations, and the nearest car parks (Pavilion KL and Lot 10). It does not repeat the question back. Only bookings and orders are repeated.
What works, and where it still slips
Works well
A Gulf customer gets Gulf Arabic back: «هلا والله», «حياكم», «تبون», the way a person at the counter would say it.
A voice note that asks for a booking is said back first. The assistant books only after the customer’s yes, and the check fails if it books early.
A correction wins. «لا، لخمسة مو لأربعة» (“no, for five, not four”) makes it book five.
After the yes, the confirmation gives the day, date, time, number of people and name, and says the table is held for 15 minutes.
Still slips
Gulf has six checks in our regular suite, not the 30-message set we wrote for Sudanese, Yemeni and Iraqi. So there is no score to show for it yet.
The Latin-letter check only confirms the reply is in Latin letters. Whether every spelling looks natural to a Gulf reader still needs a native speaker’s review.
“Gulf” is several dialects. The assistant is set up around Saudi wording and the words the Gulf states share. Local differences (Hejazi, Kuwaiti, Emirati, Omani) are not tested one by one.
A few words are shared with other dialects («شنو» with Iraqi, «ايش» with Yemeni). A message made only of shared words may not show the dialect, and the assistant can answer in simple neutral Arabic until the customer’s words do.
The words it listens for
The assistant works in two layers. A list of Gulf words gives it a hint about who it is talking to, and the AI model reads the whole message itself, so a word that is not on the list is still understood. These are the words on the list.
ابي · أبغى · ابغا · نبيI want, we want
تبي · تبين · تبون · تبغىyou want (a man, a woman, several people)
وش · وشو · ايشwhat
الحينnow
زينgood, fine
شلونhow
وايدa lot
حياك · حياكمwelcome (to one, to several)
مبnot
عطني · وديgive me · I’d like
هلا والله · أبشرwelcome · “consider it done”
طال عمرك · يبهpolite ways to address someone (“may you live long”, “dad”)
سالفة · باكرa matter, a story · tomorrow
شنوwhat (also Iraqi and Sudanese, so it only breaks ties)
In Latin letters (Arabizi)
abi · abghaابي · I want
weshوش · what
al7eenالحين · now
zainزين · good
shlonشلون · how
wayedوايد · a lot
abshirأبشر · consider it done
Words it uses back
هلا والله
حياك
أبشر
تبي
الحين
زين
تامر
Voice notes and Arabizi
A voice note is turned into text by a machine, and it can be wrong, especially with names, numbers and times. So when a Gulf customer’s voice note asks for a booking, the assistant first says back what it understood and waits for the customer’s yes, as in the examples above. The machine also tends to write formal Arabic (أريد, ماذا) even when the customer spoke Gulf, so the assistant does not treat formal wording in a voice note as a sign the customer wants formal Arabic.
Gulf customers who write in Latin letters (“hala, abi a7jez tawla”) get Latin letters back, in Gulf spelling. This is the Latin-letter case our regular suite checks; Arabizi in the other dialects is covered on their own pages.
Where this dialect is spoken
Gulf Arabic is spoken along the Arabian Peninsula’s Gulf coast and in the centre: in Saudi Arabia (Najdi speech in Riyadh and the centre, Hejazi in Jeddah, Makkah and Madinah, plus the Eastern Province and the south), and in Kuwait, Bahrain, Qatar, the UAE and Oman. Each has its own words and accent, and Saudi Arabia alone has several. The assistant is set up around Saudi wording and the words the Gulf states share.
Riyadh
Jeddah
Kuwait City
Doha
Dubai
Muscat
Questions about this dialect
Does it answer a Saudi customer differently from a Kuwaiti or an Emirati?
There is one Gulf setting, built around Saudi wording and the words the Gulf states share. Hejazi, Kuwaiti, Emirati and Omani speech are not separate settings and we have not tested them one by one. The AI model still reads each message itself, so local words are understood; what we cannot yet promise is that every reply sounds local.
Can a Saudi customer book with a voice note?
Yes. In our tests the assistant says back what it understood (the day, time, number of people and name), books only after the customer says yes, and uses the customer’s correction if they change a number. It is built this way because a machine can mishear a number.
What if my Gulf customers write «abi a7jez» in Latin letters?
The assistant understands it and answers in Latin letters too, in Gulf spelling («hala wallah, 7ayakum»). Our regular test checks that the reply stays in Latin letters; a native speaker has not yet reviewed every spelling it uses.
What if a customer switches between Arabic and English?
The assistant follows the customer’s latest message and switches when they do. In one of our checks a customer wrote in English and then asked to continue in Arabic, and the next reply came in Arabic.
We run them through the assistant and send you every reply, the good and the bad, before you decide anything. It is free. Or start the free 30-day pilot and watch every chat yourself.
Tell us about your business. We load your menu and rules, connect the front desk to your WhatsApp, and you watch every chat. 30 days free, no card. We build your website once you decide to keep it.
We take on 3 new pilots a month, so each one gets our full attention.
See it answer as your business
Send us your Instagram, Google Maps or website. We build a front desk that knows your menu, prices and hours, and send you a link to try it, usually within a day. Free, nothing to sign.