Prompts That Sell: Writing AI Prompts for Uzbek & Russian Buyers
Learn to write prompts that sell for Uzbek and Russian audiences — practical frameworks for Telegram bots, sales scripts, and AI agents that convert.
Most businesses in Tashkent that add an AI chatbot or sales assistant make the same mistake: they copy a generic English prompt from a tutorial, translate it once, and wonder why conversions barely move. Writing prompts that sell for Uzbek and Russian audiences is a different discipline — it means encoding local payment habits, trust signals, and conversational tone directly into the instructions you give the model, not just swapping vocabulary.
This matters more than most founders assume. An AI sales agent, Telegram bot, or CRM assistant is only as good as the prompt steering it. Get the prompt right and the same model that failed to close a single lead in English starts booking consultations and pushing customers toward Payme checkout in Uzbek. Get it wrong and you’ve built an expensive, fluent-sounding assistant that nobody trusts enough to buy from.
Why Direct Translation Kills Conversion
A prompt written for a US audience assumes things that don’t hold in Uzbekistan or the broader Russian-speaking market: that the buyer already trusts online payment, that “book a call” is a low-friction ask, that formal address is unnecessary. None of that transfers cleanly.
In Uzbek and Russian business culture, a slightly more formal, respectful tone (especially in Russian, using “Vy” rather than “ty”) builds trust faster than the casual, first-name-basis tone common in English sales prompts. Meanwhile, buyers expect concrete numbers — so’m prices, delivery windows, response times — rather than vague enthusiasm. A prompt that tells the model to be “friendly and persuasive” without specifying register, currency, and channel will default to a generic, translated-sounding voice that reads as foreign and untrustworthy.
The Core Framework: Four Layers of a Selling Prompt
A prompt that actually sells needs four layers, in this order:
- Role and register — who the AI is, and exactly how formal/informal it should sound in that language.
- Local context injection — payment methods, currency, channel norms (Telegram is the default channel for ~76% of Uzbekistan’s internet users), and business hours.
- Objection-handling logic — the 3–5 objections real buyers raise, with scripted-but-natural responses.
- Conversion instruction — the specific next step (a Payme/Click link, a Telegram contact button, a booked slot) rather than an open-ended “let me know if you have questions.”
Layer 1 — Role and Register
Instead of “You are a helpful assistant,” a selling prompt should specify: “You are a sales consultant for [business], writing in formal Uzbek/Russian, addressing the customer respectfully. Keep sentences short. Never sound like a script.”
Layer 2 — Local Context
This is where most generic prompts fail entirely. Explicitly tell the model:
- Prices are quoted in so’m, never dollars, for local customers.
- Accepted payment methods are Payme, Click, and increasingly Uzum Bank — mention whichever the business actually supports.
- The customer is most likely messaging from Telegram, so responses should be short enough to read on a phone screen, with a clear single call-to-action per message.
Layer 3 — Objection Handling
Build a short table of real objections into the prompt’s context (not a hidden system rule the model has to guess) — see the table below for a starting set you can adapt to your business.
| Objection (typical) | Prompt instruction for the AI |
|---|---|
| ”Narxi qimmat” / “Дорого” (too expensive) | Reframe around value/ROI, offer a smaller starter package before discounting |
| ”O’ylab ko’raman” / “Подумаю” (I’ll think about it) | Ask one clarifying question instead of pushing; offer to send info via Telegram |
| Payment trust concerns | Name the specific payment provider (Payme/Click) and mention it’s widely used, not a new/unknown method |
| Delivery/timeline doubt | Give a concrete range, not “soon” — e.g., “2–3 ish kunda” / “2–3 рабочих дня” |
Layer 4 — Conversion Instruction
End every prompt with an explicit closing instruction: “After answering, always propose one concrete next step — send the Payme payment link, offer to schedule a call, or ask for the customer’s phone number for a callback.” Vague AI responses that trail off without a next step are the single biggest reason Telegram sales bots underperform.
A Worked Example
A typical Tashkent online furniture store might use a prompt shaped like this for its Telegram ordering bot:
“You are a sales assistant for [store] on Telegram. Respond in the language the customer writes in (Uzbek or Russian). Use a respectful, moderately formal tone. All prices are in so’m. Accepted payments: Payme, Click. If the customer hesitates on price, offer to explain the value (delivery included, warranty) before any discount. Always end your response with one clear next step: sending a payment link, confirming delivery address, or offering a callback.”
In a project shaped like this, businesses typically see the AI assistant handle 60–80% of routine questions (delivery time, price, stock) without human intervention, freeing staff for the harder negotiations — a pattern consistent with the broader trend of automation removing roughly 80% of routine manual work once it’s implemented correctly.
Prompts for Qualifying vs. Prompts for Closing
Not every prompt should try to close the sale. It helps to separate two jobs:
- Qualifying prompts — designed to gather budget, timeline, and intent before handing off to a human, ideal for higher-ticket B2B or service leads.
- Closing prompts — designed for lower-friction, higher-volume sales (retail, food delivery, simple bookings) where the AI can push straight to payment.
Mixing the two into one generic prompt is a common cause of underperformance — a qualifying-style bot that keeps asking questions instead of offering to close loses impatient Telegram users fast.
Checklist: Before You Ship a Selling Prompt
- Specifies the exact language/register (formal Uzbek, formal Russian, or both)
- States local currency (so’m) and accepted payment methods explicitly
- Includes at least 3 real objections with response guidance
- Ends with a clear, singular call-to-action per message
- Has been tested with actual customer messages, not just internal team tests
- Matches the channel — short, mobile-readable replies for Telegram
Where This Fits Into Your Wider Funnel
A well-written prompt is only one piece. It needs leads flowing in from somewhere — see our breakdown of lead generation channels that actually work in Uzbekistan — and it should sound consistent with the rest of your brand’s written content, which is where AI content generation without losing your brand voice becomes relevant.
Frequently Asked Questions
Do I need separate prompts for Uzbek and Russian customers? Yes, ideally. Even if one model handles both languages well, register and idiom differ enough that a single blended prompt tends to sound stiff in both. Detect the customer’s language from their first message and switch prompt variants accordingly.
How much does a prompt-driven sales bot typically cost to build? A simple Telegram ordering bot with a tuned selling prompt typically runs 1 500 000–5 000 000 so’m depending on payment integration and catalogue complexity. A fuller CRM-connected AI sales agent starts from around 5 000 000 so’m.
Can I just use ChatGPT’s default settings for this? You can start there, but default settings won’t encode your payment methods, pricing, or objection patterns. The prompt engineering described above is what turns a generic chatbot into one that actually books sales.
How do I know if my prompt is working? Track conversion rate from first message to payment link sent, and from payment link to completed transaction. If the AI is answering questions but conversions stay flat, the issue is usually a missing or weak Layer 4 (conversion instruction).
Writing prompts that sell isn’t a one-time task — it’s iterative work that improves as you see how real customers respond. If you’d like help designing, testing, and connecting a Telegram sales bot to your CRM and payment provider, see our AI and automation services or get in touch to discuss your project.
Building something like this?
Fera Tech ships iOS & full-stack apps end-to-end. Tell us about your project.
Start a project