Open the inbox of any shop in Cairo or Jeddah on a Ramadan evening and you will find a mix that looks nothing like a catalogue: "does this dress come in large?" typed in Egyptian Arabic, "how much is the oud oil in the post?" in Gulf Arabic, a long voice note from a customer describing a dress she saw in a story, and an English message from someone in Dubai asking about Eid cakes. Voice notes and dialect are simply how people talk to shops, and any automated reply that ignores them sounds odd from the first line. This article covers how to keep your shop's tone in AI replies, how to answer each customer in their dialect or language, when a voice note needs a human, and how to write a brand voice guide that keeps the bot away from stiff, formal answers.
Why customers write in dialect and send voice notes
Customers talk to a shop the way they talk to the person behind the counter. Nobody messages a clothes shop in Nasr City with "Kindly confirm whether this item is available in a larger size." They write the local equivalent of "got this in large?" and get on with their day. In Jeddah the same question arrives in a different dialect, with different words for "I want" and "is it available".
Voice notes have their own reasons:
- Speed: the customer is on the bus or at work, and talking is easier than typing.
- Detail: describing a colour, a cut or the look of a birthday cake is easier out loud.
- Habit: many customers, especially older ones, send every message to family and friends as audio and see no reason to change that for a shop.
- Trust: some people want to feel they are dealing with a human before they confirm an order.
Then there are customers who write Arabic in Latin letters, often called Franco or Arabizi, people who mix Arabic and English in the same sentence, and residents in the Gulf who are not Arab and write in English. They are all real customers, and your replies should meet them where they are.
Customers will not change how they talk because a bot is answering. The bot has to learn how your customers talk.
Voice notes and dialect: where automated replies fit
Start with the easy part: language. In Kwamle Media it takes one line in the instructions you write for the AI: reply in the language the customer writes in. An English message gets an English reply, an Arabic one gets Arabic, and your tone and the words you avoid still apply in every language. We go deeper into this in replying to customers in their own language, and you can see how to set it up on the multilingual AI customer replies page.
Dialect is a finer step. You can ask in your instructions for the bot to answer in a dialect close to the customer's, or in a simple, neutral Arabic everyone understands, with the everyday shop words people use. But do not assume every reply will land in a perfect dialect. Test with real messages from your customers before you switch it on for everyone, and adjust the instructions whenever a sentence does not sound like your shop.
Treat a voice note as a signal. Do not build your plan on the bot understanding every audio message the way someone on your team would, especially with background noise, fast speech and different dialects. It works better to write in your instructions what the bot says when a message arrives with no text, and to have a person who listens and takes over. More on that below.
Reply in the customer's dialect, not in official-letter language
What gives away an automated reply is rarely a mistake. It is stiffness. Look at three situations that happen every day (the Arabic ones are shown here in English):
| Situation | Stiff reply | Reply in the shop's tone |
|---|---|---|
| Clothes shop in Cairo, a customer asks in Egyptian Arabic if the dress comes in large | Yes, the product is available in the requested size. Please complete your purchase via the link. | Yes, we have it in large. Want me to send you the link so you can order it? |
| Perfume store in Jeddah, a customer asks in Gulf Arabic what to buy his father as a gift | We would like to inform you that a wide range of men's fragrances is available. | Welcome! We have a few options that make a good gift. Does he like a heavy scent or a light one? |
| Bakery in Dubai, a customer writes in English about an Eid cake | Dear customer, kindly be informed that orders are subject to availability. | Hi! Yes, we take Eid cake orders. Which day would you like to pick it up? |
Notice three things in the natural replies: they are short, they answer the question first, and they end with one clear question that moves the order forward. That is exactly what you write into the instructions.
One dialect, or the customer's dialect?
If your shop is in Cairo and most customers are Egyptian, ask for Egyptian Arabic and keep it simple. If you sell to Egypt and the Gulf at the same time, you have two options:
- Match the customer: the bot answers in a dialect close to the message. It feels warmer, but it needs more testing, because broken Gulf Arabic in a reply to a customer in Riyadh is worse than plain, correct Arabic.
- Neutral Arabic: simple Arabic close to everyday speech, without the weight of formal Arabic or heavily local words. It is the safer choice and suits most shops that sell to more than one country.
Pick one and write it down plainly. A bot that does not know what you want may switch between styles inside the same conversation, which confuses customers more than any dialect.
Franco and mixed languages
If a customer writes "3andak el size da?" ("do you have this size?" in Latin letters), you decide: should the bot reply in Arabic script, in Franco or in English? Most shops prefer Arabic script because prices, addresses and delivery details read more clearly. What matters is that the decision is written in the instructions, not left to chance.
When a voice note needs a person
Automated replies are great for repeated questions: is it in stock, what sizes, do you deliver to my area. But some messages, audio or written, should go to a human:
- A complaint or clear frustration: the tone of voice alone says the customer is upset, and a ready-made sentence will not fix that.
- Long order details: a complicated address, several items in different sizes, a gift with a written card.
- A special request: a cake with specific writing, special wrapping for a perfume bottle, a change to a design.
- Anything involving money: returns, exchanges, a problem with cash on delivery or a card payment.
- A customer who prefers talking: someone who sends voice note after voice note usually wants to deal with a person.
Write a clear rule in your instructions: when a voice note or a photo arrives with no text, the bot replies kindly that the message has arrived and someone from the team will answer. For example: "We got your voice message, someone from our team will listen and get back to you shortly." Just as important, the bot should never pretend it understood a message it did not, and answer with something generic that has nothing to do with what the customer said.
In Kwamle Media every conversation lands in one inbox, and anyone on your team can listen to the voice note and take the chat over from the bot. If your team moves between the shop floor and the stockroom, they can reply from the mobile app for support teams. With a bigger team, assignment rules share out conversations so two people do not answer the same voice note and none gets lost between them.
Write a brand voice guide the bot can follow
The bot only knows your shop from what you write. That is why you need a short brand voice guide, which is your brand voice written down in a practical way. It does not need to be long; one clear page beats ten pages of generalities. Include:
- Who you are: the type of shop, the city and who your customers are, in two lines.
- Dialect: Egyptian, Gulf, neutral Arabic, or matching the customer.
- How you address people: formal or casual, which greeting you use. When the bot cannot tell whether the customer is a man or a woman, ask for neutral wording, which matters a lot in Arabic.
- Your words: product names exactly as you write them, and terms like "order", "pick up from the shop" and "cash on delivery".
- Phrases you never use: "Dear customer", "We apologise for any inconvenience", "Kindly be informed". Write them plainly in the instructions under a clear heading such as: phrases we never use.
- Sample replies: a few real replies you wrote and like. They teach the bot faster than any description.
- What it must never say: no promised delivery date, no unannounced discount, no order confirmed before the details are complete.
- When to hand over: complaints, returns, special requests and voice notes.
Here is what a guide for a Cairo clothes shop might look like:
We are a women's clothing shop in Cairo. We talk to customers in simple Egyptian Arabic, kindly and without exaggeration.
We say "order", "pick up from the shop" and "cash on delivery". We never say "Dear customer" or "Kindly".
We reply in two or three sentences and end with one question.
We do not promise a delivery date or a discount that is not announced.
If a customer complains, sends a voice note or asks about a return, we say someone from the team will reply.
The perfume store in Jeddah writes the same kind of guide in its own spirit: a warm Gulf welcome, perfume names exactly as they appear on the bottle, and a clear rule about gifts and wrapping during Eid.
How to avoid stiff, formal replies
Even with a good guide, a bot sometimes drifts towards official-letter language. These habits keep it away from that:
- Keep sentences short: the customer wrote one line, so do not answer with a paragraph.
- One question per reply: "Black or beige?" works better than three questions at once.
- Lead with the answer: "Yes, we have it" before any long greeting.
- Do not repeat the product description: the customer saw the post and only needs the missing detail.
- Use the customer's name sparingly: once early in the conversation is enough.
- Read conversations every week: open the inbox, and whenever you spot a stiff sentence, add a better sample reply or add the phrase to the list of things the bot never says in your instructions.
You will find more ideas in DM automation that sounds human.
Getting started in Kwamle Media
Write the language and dialect rule
Add to the AI's instructions: reply in the language the customer writes in, then state the dialect you chose and what to do with Franco.
Add your voice guide and the phrases you avoid
Paste your guide into the instructions, list the phrases you do not want clearly inside them, and add your sample replies.
Test with real messages
Take questions from past conversations in Egyptian, Gulf Arabic, English and Franco, and see how the bot answers before you turn it on for everyone.
Decide who listens to voice notes
Agree with your team who takes over audio messages and complaints, and at what times, so no customer waits for a reply that never comes.
Common questions
What happens when a customer sends a voice note?
Write in your instructions what the bot says when a message arrives with no text, for example that the message has arrived and a person will reply. Then someone on your team listens from the inbox or the mobile app and takes over the conversation. That way the customer is never left without an answer, and never gets a reply that ignores what they said.
Can the bot reply in Egyptian or Gulf Arabic?
You can ask for it in your instructions and add sample replies in the dialect you want. No reply is guaranteed to be perfect, so test with real messages first. If you are unsure, simple Arabic close to everyday speech is a safe choice.
Should I use one dialect for all customers?
If most of your customers come from one country, a single dialect is simpler and more consistent. If you sell to several countries, choose between matching the customer's dialect and neutral Arabic, and write your choice clearly in the instructions.
What if a customer writes in Franco or mixes Arabic and English?
Decide the reply language in advance and write it in the instructions. Most shops reply in Arabic script because prices and addresses read more clearly, while some reply in English to certain customers. What matters is that the choice stays the same from one message to the next.
Does the same tone apply to comments and messages?
Yes, the same instructions apply to comment replies and private messages. A comment is public, though, so ask for shorter replies there and move order details into private messages.
How do I know the bot needs adjusting?
Read a handful of conversations in the inbox every week. Repeated formal phrases, replies in a dialect you did not choose, or customers asking the same question twice are all signs to update the instructions or add a new sample reply.



