View All
ai

Adding an AI Chatbot to Your Website in Dari and Pashto

Jawad Hakimi6 min read

An AI chatbot website assistant is a chat window on your site, powered by a large language model, that answers visitor questions in their own language using your business information. For Afghan businesses, the practical setup is a chatbot that reads Dari, Pashto and English, answers only from your approved content (prices, services, hours, policies), and passes the conversation to a person on WhatsApp or by phone when it cannot help.

Key Takeaways

  • Modern language models understand Dari and Pashto reasonably well, but Pashto answers need more testing and tighter instructions than English.
  • The chatbot should answer from your own documents (a knowledge base), not from the model's general memory, or it will invent prices and policies.
  • The chat widget must support right-to-left text, Persian and Pashto fonts, and mixed-script messages such as Dari written in Latin letters.
  • A clear handoff to a human, usually WhatsApp or a phone call, matters more than clever answers.
  • Start with the 20 to 30 questions your staff already answer every week, then expand.

Can an AI Chatbot Really Understand Dari and Pashto?

Yes, the leading models in 2026 can read and write both languages, but quality is uneven and you must test it with real customer messages. Dari benefits from the large amount of Persian text the models were trained on, so grammar is usually good, although some answers drift toward Iranian vocabulary. Pashto has far less training data, so answers can be stiff, mix dialects, or occasionally misread a word.

Customers also write in ways that textbooks do not cover. Many people in Kabul and the diaspora type Dari or Pashto in Latin letters ("salam, qimat chand ast?"), switch between languages in one message, or send voice notes. A good setup handles the first two with instructions and examples, and either transcribes voice notes or politely asks for text.

How Does a Website Chatbot Know Your Business?

It answers from a knowledge base you control, using a method called retrieval-augmented generation. Instead of trusting the model's general knowledge, the system searches your documents for the relevant passage and asks the model to answer only from that passage. We explain the method in more depth in our guide to RAG chatbots that answer from your own documents.

Your knowledge base can be simple: a set of short pages covering services, prices, delivery areas, opening hours, admission requirements, refund rules and contact details. Write each fact once, in all three languages if you can. When a price changes, you update one document and the chatbot is correct the same day.

What Should the Chatbot Do and Not Do?

The chatbot should answer routine questions, collect leads and route people, and it should never make commitments your staff have not approved. Typical good jobs include:

  • Explaining services, packages and what is included.
  • Answering opening hours, location and delivery or service areas.
  • Collecting a name, phone number and request, then sending it to your team.
  • Booking or pre-qualifying appointments through your existing booking form.
  • Pointing people to the right page on your website.

Jobs to keep away from the bot: final price negotiation, medical or legal advice, refunds, anything involving payment details, and complaints. For these, the bot's job is to say clearly that a person will help and to hand over the conversation.

How to Add an AI Chatbot to Your Website: Step by Step

The process takes two to six weeks for most small businesses, and most of that time is writing and testing content rather than code.

  1. List real questions. Export a month of WhatsApp, Facebook and phone enquiries and group them. These become your test set.
  2. Write the knowledge base. Short, factual pages in Dari, Pashto and English. Remove anything outdated.
  3. Choose the model and platform. Either a hosted chatbot product or a custom build connected to a model API. Custom builds give you control over language, tone and data.
  4. Write the system instructions. Tell the bot to reply in the customer's language, use Afghan Dari terms, stay within the knowledge base, and say "I don't know, let me connect you" when unsure.
  5. Design the widget for RTL. Right-aligned bubbles for Dari and Pashto, proper fonts, and an input box that switches direction automatically.
  6. Set up the handoff. A button that opens WhatsApp with the conversation summary, or a form that sends the lead to your team.
  7. Test with real people. Ask Dari and Pashto speakers from different regions to try to break it. Fix the knowledge base, not just the prompt.
  8. Launch, then review weekly. Read conversations, note wrong or missing answers, and update content.

How Much Does an AI Chatbot Cost in Afghanistan?

Costs have two parts: the one-off build and the ongoing usage. The build covers the knowledge base, widget, integrations and testing. At Sheen, chatbot projects fall under our automation service, which starts from AFN 14,000 for simpler setups and AFN 28,000 for more involved ones, with custom quotes for complex integrations.

Usage costs come from the model provider and depend on how many conversations you handle and which model you use. For most small businesses these are modest, but pricing changes often, so check the provider's current rates and set a monthly spending limit. Hosted chatbot platforms usually charge a monthly subscription instead.

What About Speed, Privacy and Accuracy?

Keep the widget light, protect personal data and measure accuracy, because a slow or wrong chatbot does more harm than no chatbot. On slow mobile connections in Afghanistan, load the chat script only after the page has loaded or when the visitor taps the chat button, so it does not slow down your site.

On privacy, tell visitors that they are talking to an automated assistant, avoid asking for sensitive information in chat, and check where conversation data is stored. On accuracy, keep a simple log: each week, mark a sample of answers as correct, partly correct or wrong. If Pashto answers lag behind, add more Pashto examples and simplify the Pashto knowledge base wording.

If you want the chatbot to work beyond your website, the same knowledge base can power replies on WhatsApp. See our guide to WhatsApp Business API automation and our overview of AI customer support automation.

Which Businesses Benefit Most?

Businesses with many repeated questions and customers who message outside office hours benefit most. Education centers get the same admission and fee questions every term, clinics get appointment and doctor schedule questions, and online stores get delivery and product questions. Diaspora customers in Europe, North America and Australia often write when it is night in Kabul, so an assistant that can answer basic questions and capture the lead is genuinely useful.

Work With Sheen

Sheen builds multilingual AI chatbots that answer in Dari, Pashto and English from your own content and hand off cleanly to your team. See our automation service or book a call to discuss your website. You can also email info@sheen.af or call +93 784 966 018.


Frequently Asked Questions

Which AI model is best for Dari and Pashto chatbots?

There is no permanent winner, because models improve every few months. The leading commercial models generally handle Dari well and Pashto acceptably. The right approach is to test two or three models on your own customer questions in both languages, compare accuracy and tone, and choose the one that performs best for your content. Re-test whenever you consider switching models.

Will an AI chatbot slow down my website?

It should not if it is built carefully. The chat script can be loaded only after the main page content has appeared, or only when a visitor taps the chat button. The model runs on the provider's servers, not on the visitor's phone, so the main cost to the visitor is the small widget download and the text messages, which matters on slow mobile data in Afghanistan.

Can the chatbot answer customers who write Dari in English letters?

Usually yes. Current language models can often understand romanized Dari and Pashto, especially common phrases about prices, locations and timings. You improve results by adding examples of romanized messages to the system instructions and test set, and by deciding which script the bot should reply in. Many businesses reply in proper Dari or Pashto script, or in the same style the customer used.

What happens if the chatbot gives a wrong answer?

Wrong answers are reduced, not eliminated, by limiting the bot to your knowledge base and telling it to say when it does not know. Add a visible note that it is an automated assistant, keep a human handoff one tap away, and review conversations weekly. When you find an error, fix the source document so the same mistake does not repeat, and avoid letting the bot confirm prices or bookings on its own.


Keep reading

Read More