Multilingual Telegram Bot: Uzbek, Russian and English Setup
A multilingual Telegram bot is not a toggle you bolt on in an hour. It means separate texts for every language, storing each user's choice, testing every scenario and getting number and date formats right. Done carelessly, multilingual support annoys people more than having no translation at all. Below is how to set it up properly for an audience in Uzbekistan, where Uzbek, Russian and English easily meet in a single chat.
Why a business in Uzbekistan needs several languages
Customers choose their language differently: some are most comfortable in Uzbek, some in Russian, and younger or technical audiences often read in English. If a bot speaks only one language, part of your audience simply closes the chat before submitting a request. This is not a statistic but funnel logic: the harder each step is to understand, the fewer people complete it.
Before building three languages, decide which ones you really need. For most local services, such as shops, clinics, schools and delivery, Uzbek and Russian are enough. English is worth adding if you work with tourists, foreign clients or a technical audience. Every extra language means separate texts that you must update each time the bot changes.
The best guide is your own data: which language customers use when they write or call you, and which language your website and ads use. That is more reliable than any guess.
How a multilingual Telegram bot works: choosing and storing the language
The language should be chosen explicitly at first launch with buttons such as O‘zbekcha, Русский and English, not inferred from Telegram's interface language. Many users keep their interface in English or Russian while writing in Uzbek, so auto-detection often gets it wrong. The interface language is fine as a hint, for example to put the likely option first in the list.
The choice is saved in the user's profile in your database next to their Telegram ID. On every message the bot reads the stored language and takes texts from the matching dictionary. Do not keep the language only in process memory: after a restart everyone would forget their choice and see the language screen again.
Users must be able to switch at any time through a /language command and a main-menu item. Write that item's name in all languages at once, as in Til / Язык / Language, so someone who landed in the wrong branch can find the way out. Switching must not wipe a cart, a form in progress or an order being placed.
Technically, all strings live in separate files per language with identical keys, and the code contains only keys. No text is hard-coded into logic, and the check that every key exists in every language can run automatically.
Why auto-translation fails
Machine translation of formal phrasing and short button labels often sounds unnatural in Uzbek, especially when the source was written with Russian grammar in mind. Uzbek has a different word order, its own polite forms of address and its own habitual phrasing for requests, payment and delivery. Users notice the unnaturalness instantly and lose trust in the bot.
A workable process: a native or fluent speaker writes the text, then a second person reads it through the customer's eyes. Machine translation is acceptable only as a draft that is always edited by hand. Keep a single glossary of terms, so that order, cart, delivery and payment are called the same thing in every message.
Decide in advance which script Uzbek uses in the bot, Latin or Cyrillic, and do not mix them within one scenario. Also account for string length: translations can be noticeably longer than the source, and buttons that fit in a row may break apart on a phone.
What is easy to forget to localize
Error messages, order-status notifications and out-of-hours auto-replies often stay in one language because they are written last and in a hurry. The same goes for input hints, captions under payment buttons, and confirmation and cancellation texts.
Telegram lets you set the bot description, short description and command list separately per language, and it is worth using: a person sees the bot in their language before pressing Start. Broadcasts must reach each user in their own language, which means a template per language and segmentation by the stored choice. Notifications coming from external systems such as a CRM, payments or delivery also pass through your dictionaries.
Check formats: amounts with spaces between digit groups and a so‘m or сум label, dates in day, month, year order, time in 24-hour format. A mismatch reads as sloppiness even when the text is translated correctly.
Finally, user input: the bot should understand names, addresses and search queries in both Cyrillic and Latin. If the catalog can be searched in only one spelling, some customers will find nothing.
How to test the bot before launch
Walk through every scenario in every language from /start to the request or payment, and do it on a real phone, not only in the desktop app. Buttons that do not fit and strings that get cut off show up exactly on a phone.
A minimal checklist: every text key exists in every language; the language survives a bot restart; switching mid-scenario does not break an order; input errors such as an empty field or an invalid phone number appear in the right language; broadcasts and notifications arrive in the chosen language; a user returning after a pause sees the same language.
Separately, give the bot to people who were not involved in development, one native speaker per language, and collect the fixes in a single list. Typical mistakes: leftover Russian in the Uzbek branch, two translations of one term, mixed languages in one message, and a broken layout from long strings.
After launch, watch which language people choose and at which step they abandon the scenario. Texts and step order should be corrected using that data.
What it costs and how long it takes
The price is made up of four parts: the logic for storing and switching language, writing the texts for each language, localizing formats, broadcasts and notifications, and testing each language. The more scenarios the bot has, the more strings there are and the more each extra language costs.
Market guide: adding a second language to an existing bot usually adds 15–25% to the project cost, and a native speaker's work on the texts of a small bot with 10–20 screens may run roughly 300,000–1,500,000 UZS per language. These are indicative figures; the exact price depends on volume. For a small bot, expect roughly 3 to 7 working days per extra language, including testing. Planning languages in from the start is noticeably cheaper than reworking a finished architecture.
4 Digital builds Telegram bots from 999,000 UZS, including bots in Uzbek, Russian and English. If you already have a bot or only an idea, send us your scenarios and the languages you need, and we will estimate the scope and name a price.
FAQ
Can the language be detected automatically? You can suggest an option based on Telegram's interface language, but it is better to leave the final choice to the user through buttons and always let them change it later.
How many languages does a bot need? For most businesses in Uzbekistan, Uzbek and Russian are enough. Add English if your customers include tourists, foreign partners or an English-speaking audience.
Can I translate a bot with a machine translator? Only as a draft. A native speaker should check and correct the final texts, otherwise buttons and messages will sound unnatural.
Can a language be added to a bot that is already running? Yes, if texts are moved into separate files with keys and the language is stored in the user profile. If texts are hard-coded in the source, they must first be extracted, which makes the work more expensive.