Industries
Restaurants & Food
Take restaurant orders, table bookings and payments in one WhatsApp chat. Auto-reply to common questions and cut support costs with ChatMitra.
Real Estate
Use WhatsApp automation to capture property leads, qualify them automatically, and follow up fast. See how real estate agents close more with ChatMitra.
Travel & Tourism
Use the WhatsApp Business API to capture travel leads, send itineraries, confirm bookings, and keep travellers updated. See how agencies grow with ChatMitra.
Banking & Finance
Run secure KYC, send account alerts and payment notifications on WhatsApp. Cut support load and stay within Meta's rules with ChatMitra.
Beauty & Cosmetics
Sell your catalog, recover carts and win repeat orders on WhatsApp. Auto-reply to product questions and send launches to opted-in customers with ChatMitra.
Education
Answer admission enquiries, fee and timetable questions, and keep parents updated — all on WhatsApp. Cut admin load and costs with ChatMitra.
EduTech
Capture course leads, qualify them and onboard learners on WhatsApp. Automate replies and cut lead-qualification cost with ChatMitra's no-code chatbot.
Events & Webinars
Handle event registrations, reminders and RSVPs in one WhatsApp chat. Cut no-shows, automate FAQs and keep attendees updated with ChatMitra.
Freelancers & Consultants
Capture leads, send proposals and chase invoices in one chat. WhatsApp automation for freelancers and consultants with ChatMitra — from ₹0.20 a chat.
Health & Wellness
Book appointments, send reminders and answer client questions in one WhatsApp chat. Auto-reply to routine queries and cut support costs with ChatMitra.
Home Decor
Share your home decor catalog, take orders and send delivery updates on WhatsApp. Automate FAQs and cut support costs with ChatMitra.
IT Services
Use the WhatsApp Business API to qualify leads, run support over chat, and wire WhatsApp into your stack with webhooks. How IT companies scale with ChatMitra.
Marketing Agencies
Use WhatsApp automation to capture leads, run campaigns and manage multiple client brands from one account. How agencies scale on WhatsApp with ChatMitra.
Offline & Retail
Turn shop footfall into a WhatsApp list. Share catalogs, add a QR-to-chat at the counter and bring buyers back — with ChatMitra from ₹0.20 per chat.
Spas & Salons
Use the WhatsApp Business API to take spa and salon bookings, cut no-shows with automated reminders, and rebook regulars. How salons grow with ChatMitra.
Automotive
Use the WhatsApp Business API for automotive dealerships to qualify buyers, book test drives and send service reminders — automated, on one chat, with ChatMitra.
All Industries
Browse every industry guide
Automation & Chatbots

Multilingual WhatsApp Chatbot: A Practical Setup Guide

How to build a multilingual WhatsApp chatbot: language menus vs auto-detect, Hinglish, one template per language code, testing, handover and costs in ₹.

Janki Gohil
Written byJanki Gohil
Read time 19 min
Posted on
Multilingual WhatsApp Chatbot: A Practical Setup Guide
Summarise this post with:

Short on time? Read this first

  • A multilingual WhatsApp chatbot has three layers to translate: the bot's replies inside the 24-hour window, the approved templates you send outside it, and the human who takes over. Most guides only cover the first.
  • There are three ways for a bot to know a customer's language: ask with a button menu, detect it with an AI model, or remember a choice saved on the contact. Ask once, remember forever, and let AI handle the rest.
  • Meta treats each language of a template as a separate template with its own approval. There is no automatic fallback: sending a language that isn't approved fails with error 132001.
  • Meta's template list covers 10 of India's 22 scheduled languages plus English. Odia and Assamese are the largest gaps.
  • Language doesn't change Meta's price. A Hindi reply costs the same as an English one. The real costs are translation, review, approvals and an extra menu message per conversation.
On this page 0

    A multilingual WhatsApp chatbot is an automated assistant that talks to each customer in the language they prefer, on the same WhatsApp number. It greets a Tamil speaker in Tamil, answers a Hindi speaker in Hindi, and sends a Gujarati speaker their order update as a Gujarati template. Building one well is less about translation software and more about three decisions: how the bot learns a customer’s language, where that choice is stored, and who writes and checks every version of every message.

    This guide walks through those decisions for the WhatsApp Business Platform (the API), with examples from India, where a single business can easily serve customers in four or five languages. It covers language menus versus detection, the Hinglish problem, Meta’s rules for templates in multiple languages, testing, human handover, and what it all costs.

    Key takeaways

    • Translate three layers, not one: bot replies, approved templates and the human handover.
    • Ask for the language once with a button menu, save the answer on the contact, and use AI only where customers type freely.
    • Every language of a template is a separate template with its own approval. Missing languages fail; they never fall back.
    • Meta's template list covers 10 of India's 22 scheduled languages plus English. Check yours before you promise it.
    • Language doesn't change Meta's price. Translation, review and an extra menu message do.

    What a multilingual WhatsApp chatbot has to translate

    Most articles on this topic describe a bot that detects a language and translates its answer. That’s one layer. On WhatsApp, a business talks to a customer through three different kinds of message, and each needs its own language plan.

    LayerWhen it's usedWho writes the textLanguage rule
    Bot repliesInside the 24-hour customer service window, after the customer messages youYour rules or your AI model, liveFree-form: any language and script WhatsApp can show
    TemplatesOutside the window: reminders, order updates, offers, follow-upsYou, in advance, approved by MetaOnly Meta's supported language codes, one approval per language
    Human handoverWhen the bot can't help or the customer asks for a personYour staffDepends on who is on shift and what they speak

    The 24-hour rule is what splits the first two layers. Meta’s documentation says that when a customer messages you, a customer service window opens for 24 hours and resets each time they message again. Inside it, you can send ordinary messages: text, buttons, lists and media. Once it closes, you can only send approved templates. So a bot that speaks perfect Gujarati in the chat but sends the next day’s delivery reminder in English hasn’t really gone multilingual.

    The third layer is the one that breaks most often. A customer who chose Tamil in your menu and then asks for a person expects a person who reads Tamil, or at least an honest message about when one will be available.

    Which languages should your bot support?

    Start with your own data, not India’s language map. Count the languages customers used in your last few hundred chats, or in call-centre and store data if WhatsApp is new for you. Two or three languages usually cover a regional business; a national brand might need five or six.

    The wider picture helps you sanity-check that count. In the 2011 Census, Hindi was the mother tongue of 43.63% of India’s population, followed by Bengali (8.03%), Marathi (6.86%), Telugu (6.70%) and Tamil (5.70%). The Census groups about 50 mother tongues, including Bhojpuri, under Hindi, so the real picture is more varied than one big number suggests. Online behaviour points the same way: the KPMG–Google study Indian Languages – Defining India’s Internet (2017) estimated that nine out of every ten new internet users in India were likely to be Indian-language users, and found that over 90% of Indian-language internet users used chat apps.

    Then check what WhatsApp templates allow. Meta’s supported-languages list (updated 21 May 2026) covers ten of India’s 22 scheduled languages, plus English.

    LanguageMeta template codeMother tongue share, Census 2011
    Hindihi43.63%
    Bengalibn, bn_IN8.03%
    Marathimr6.86%
    Telugute6.70%
    Tamilta5.70%
    Gujaratigu4.58%
    Urduur4.19%
    Kannadakn3.61%
    Malayalamml2.88%
    Punjabipa2.74%
    Englishen, en_US, en_GB, en_IN(widely used as a second language)

    Together, those ten languages are the mother tongue of roughly 89% of the population in the 2011 figures. The largest language missing from the template list is Odia (3.10%), followed by Assamese, Maithili, Santali, Kashmiri, Nepali, Sindhi, Dogri, Konkani, Manipuri, Bodo and Sanskrit.

    Missing from the list doesn’t mean you can’t serve those customers. Inside the 24-hour window, a bot can still reply in Odia, because free-form messages aren’t tied to a language code. You just can’t send an approved Odia template, so reminders outside the window must go in another language the customer reads. Ask them which; don’t assume.

    Two details trip people up here. Punjabi’s code is pa, not pa_IN. And the code you send must match the approved template exactly: a template approved as en_US won’t go out if your system asks for en.

    Three ways a bot can know the customer’s language

    Every multilingual bot needs an answer to “which language do I use with this person right now?” There are three ways to get it, and the best setups combine them.

    Three stacked speech bubbles labelled Ask, Detect and Remember, each a different shade of green, joined into one conversation thread
    Ask once, remember the answer, and let AI cover what customers type freely.
    MethodHow it worksStrengthsWeak spots
    AskFirst reply is a button or list menu: English, हिंदी, தமிழ்…Accurate, cheap, works with simple rule-based botsAdds a step; a customer who ignores the menu stays unidentified
    DetectAn AI model reads the message and answers in the same languageNo extra step; handles Hinglish and mixed messagesGuesses on short messages like "ok", "hi" or "1"; needs an AI model
    RememberThe choice is saved on the contact (a tag or field) and reusedNext conversation starts in the right language; lets you route templates and campaignsOnly works after one of the other two has produced an answer

    Asking: the language menu

    A menu is the most reliable method, and it’s easy to build because WhatsApp has two native formats for it. A reply-button message can carry up to three buttons, each with a label of up to 20 characters. A list message can hold up to ten rows in total, each with a title of up to 24 characters. So three languages fit on buttons; four to ten need a list.

    Write each option in its own script: “English”, “हिंदी”, “தமிழ்”, “ગુજરાતી”. A Tamil speaker scanning a menu recognises “தமிழ்” faster than “Tamil”, and it signals that the bot really speaks the language. Meta doesn’t say how it counts characters in Indic scripts, where one visible letter can be several Unicode code points, so test long labels on a real phone before launch.

    Ask only once. If the customer has a saved language, skip the menu and add a small “Change language” option to the main menu instead.

    Detecting: letting AI read the message

    Large language models are good at answering in the language they were addressed in, including Hinglish, without any special setup. That makes detection the natural choice for an AI agent that handles open questions.

    Detection has two weak spots. Short messages like “ok”, “hi” or “1” look the same in every language, so a model may default to English. And a customer who writes in English to be safe may prefer Marathi but never be offered it. Combine methods: reply in the language of the message, but use the saved preference when the message is too short to tell.

    Remembering: saving the choice

    Whichever method produces the answer, save it on the contact. A tag such as lang_hi or lang_ta is enough. It starts the next conversation in the right language, tells your team what to expect, and lets you send each campaign or reminder in the right language.

    Hinglish and romanised input: the part most guides skip

    Many Indian customers don’t type Hindi in Devanagari. They write it in Latin letters, often mixed with English: “mera order kab aayega”, “price kya hai”, “haan, confirm karo”. The same habit exists for Tamil, Marathi, Gujarati and other languages. A bot that only understands Devanagari Hindi and formal English will miss a large share of real messages.

    One long tangled speech bubble trail looping between a customer's phone and a shop counter, labelled with the words haan, han, haa and हाँ along the loops
    The same "yes" can arrive in four spellings. Keyword bots need all of them; AI models usually cope.

    How you handle it depends on the kind of bot.

    Keyword and rule-based bots match text exactly, or by the start of the message, or by a pattern. “haan” won’t match “han”, “haa” or “हाँ”. Either list every common spelling as a keyword or use a pattern (a regular expression) such as ^(haan|han|haa|ha)$ alongside the Devanagari form. Better still, avoid typed answers where you can: a button labelled “हाँ” or “Yes” sends the same text every time, however the customer spells things.

    AI agents usually understand Hinglish well, because their training data is full of it. The question is how they should reply. A good default is to mirror the customer’s script: answer romanised Hindi in romanised Hindi, Devanagari in Devanagari. If the customer chose “हिंदी” in your menu, use Devanagari for the bot’s fixed texts and templates, and let the AI mirror what they type.

    Templates are the exception. Meta has no separate code for Hinglish. A Hindi template is approved under hi, and you decide whether its text is in Devanagari or Latin script. Some businesses write their hi templates in Devanagari and keep their English templates conversational; test what your customers respond to.

    Templates in several languages: one name, many approvals

    Templates are where multilingual WhatsApp gets administrative. Meta’s rules, from its template documentation:

    1. Same name, different languages. “Names are not unique,” Meta says, so you can create order_update in Hindi, Gujarati and English. Each is a separate template with its own language code.
    2. Every language is reviewed separately. One version can be approved while another is rejected or still pending. Plan for that before a launch date.
    3. Meta doesn’t translate anything. Its documentation says template strings and variables are not translated, so you supply every version, including the example values for each variable.
    4. Each language counts against your limit. A business portfolio that isn’t verified can hold 250 templates per WhatsApp Business Account. Ten templates in four languages is already 40.
    5. There is no fallback. If you send order_update with the code ta and no Tamil version is approved, the message fails with error 132001: “The template does not exist in the specified language or the template has not been approved.” Meta’s current Cloud API documentation has no language fallback option.

    That last point shapes your sending logic. Before you send a template in a customer’s saved language, check that the language version is approved, and fall back yourself, usually to English or Hindi, if it isn’t. Don’t let a missing translation silently drop a delivery update.

    A simple naming plan keeps this manageable:

    Template nameCategoryLanguagesOwner of each translation
    order_confirmedUtilityen_US, hi, guOps lead writes English; two named staff review Hindi and Gujarati
    delivery_todayUtilityen_US, hi, guSame reviewers
    festive_offer_diwaliMarketingen_US, hi, guMarketing writes all three, reviewer signs off

    Keep the meaning, variables and buttons identical across languages. If the English version has {{1}} for the name and {{2}} for the order number, every translation must use the same variables in a natural position for that language. Hindi and Tamil word order differs from English, so the order number may need to move within the sentence. The variable number stays the same.

    Category doesn’t change with language. A delivery update is utility in every language, and an offer is marketing in every language. If a translation drifts promotional (“Order again and save!”), Meta can treat it as marketing.

    For the codes themselves, our WhatsApp template language codes reference lists the common ones and the matching rules.

    How to build a multilingual WhatsApp chatbot, step by step

    This sequence works for most businesses, whatever platform they use.

    1. Choose your languages from your data. Two to four usually cover most customers. Note which are on Meta’s template list.
    2. Write the master version in one language. Finalise the welcome, menus, answers and templates before translating anything. Translating a moving draft multiplies the edits.
    3. Build a glossary. List your product names, brand terms, units and the words you never translate, such as order IDs, “COD” or your app’s name. Decide how formal to be: in Hindi, a business usually writes “आप”, not “तुम”.
    4. Translate, then have a native speaker review. Machine translation is a fine first draft. A person who speaks the language and knows your business checks every line, especially prices, dates, money and health.
    5. Build the language menu. Buttons for up to three languages, a list for more, each label in its own script.
    6. Save the choice on the contact. A tag per language is enough. Make “Change language” reachable from the main menu.
    7. Branch the conversation. Each language gets its own main menu and answers, with an identical structure so reports compare like with like.
    8. Set the AI’s language rule if you use an AI agent. Write it into the instructions explicitly (see the next section).
    9. Create every template in every language and wait for all approvals before switching on reminders.
    10. Plan the handover by language. Decide who answers Hindi, Tamil or Gujarati chats, during which hours, and what the bot says when nobody who speaks the language is available.

    A worked example: a diagnostics lab in Pune

    Picture a hypothetical lab with three collection centres. It gets WhatsApp messages in Marathi, Hindi and English. Its bot does four jobs: test prices, home-collection requests, report status and centre timings.

    • The first reply is a three-button menu: “English”, “मराठी”, “हिंदी”. The customer’s tap is saved as a tag.
    • Each language has the same four-item main menu, translated by two staff members who speak Marathi and Hindi at home.
    • Report-ready alerts go out as a utility template, report_ready, approved in en_US, mr and hi. The sending system checks the customer’s tag and uses English if the tag is missing.
    • Home-collection requests go to a person. The bot says, in the customer’s language, that a colleague will confirm the slot, and staff reply in the same language.

    One detail is easy to miss. Marathi and Hindi both use Devanagari script, and some words look the same in both. If both menus have a button labelled the same way, a rule-based bot can’t tell which language the tap came from. The lab words its Marathi and Hindi buttons differently, so every label is unique.

    Writing a language rule for an AI agent

    An AI agent’s instructions (its system prompt) should say exactly how to choose a language. Left alone, most models do reasonably well but drift: they switch to English mid-conversation, translate brand names, or answer a Hinglish message in formal Hindi.

    A language block you can adapt:

    Reply in the language and script the customer used in their latest message. If their message is too short to tell (for example “ok”, “hi” or a number), use their saved language: Marathi, Hindi or English. Never translate product names, test names, order IDs or prices. Write prices with digits and the ₹ sign. If you aren’t sure what the customer means, ask a short question in their language rather than guessing. If the customer asks for a person, or the question is about a medical result, tell them a colleague will reply and stop answering.

    Three habits make this work in practice:

    • Give facts in one language. Keep your price list, timings and policies in one master language in the agent’s instructions. Models translate facts well; four hand-kept copies of a price list drift apart.
    • Test with the messages people really send. Use short Hinglish, voice-to-text with odd spellings, emojis and English words dropped into Tamil sentences, not neat textbook sentences.
    • Keep it business-specific. Meta’s terms for the WhatsApp Business Platform prohibit AI providers whose general-purpose AI is the “primary (rather than incidental or ancillary)” functionality offered. A support or sales bot for your own business, in any language, is incidental to your business and isn’t what that rule targets.

    For prompt patterns beyond language, see our AI prompts for WhatsApp chatbots.

    Translation quality: where multilingual bots usually fail

    Customers forgive a bot that’s simple. They’re less forgiving of one that sounds like a bad translation, because it signals that the business doesn’t really speak their language.

    Two phone screens side by side: on the left a muddled chat where a translated reply misses the customer's question, on the right the same question answered clearly in the customer's language
    A literal translation can answer the wrong question. A native reviewer catches it before customers do.

    The usual problems, and their fixes:

    • Literal translation of idioms. “We’ll get back to you shortly” translated word for word can read oddly. Ask the reviewer to rewrite, not just correct.
    • The wrong register. Overly formal Hindi or Tamil feels like a government notice. Match how your staff speak to customers at the counter.
    • Numbers, dates and money. Decide whether you use Western digits (1, 2, 3) or native numerals, and stick to it. Most Indian businesses use Western digits in every language because customers type them that way. Write dates in a way that can’t be misread: “15 Oct” is safer than “10/15”.
    • Right-to-left scripts. Urdu and Arabic run right to left. A Latin order number or a URL inside an Urdu sentence can display in a confusing order. Preview every RTL message on a real phone.
    • Label length. A 20-character button that fits “Track my order” may not fit the Tamil or Malayalam equivalent. Shorten the wording, not the meaning.
    • Inconsistent terms. If the Hindi menu says “ऑर्डर” and the Hindi template says “आदेश”, customers wonder whether those are different things. The glossary prevents this.

    WhatsApp can now translate some messages itself: it introduced on-device translation in September 2025, with six languages on Android at launch (English, Spanish, Hindi, Portuguese, Russian and Arabic) and more on iPhone. It depends on the customer’s device, and WhatsApp’s Help Center says it isn’t available in the WhatsApp Business app. It’s a convenience, not a substitute for writing in the customer’s language.

    What a multilingual WhatsApp chatbot costs

    Language doesn’t change Meta’s price. Meta charges per delivered message by category (marketing, utility, authentication, service) and by the recipient’s country. A Hindi reply costs the same as an English one, and a Gujarati utility template costs the same as an English utility template.

    What does change:

    • One extra message per new conversation. The language menu is a service message, and it counts towards your free tier.
    • Translation and review time. This is the biggest real cost, and it recurs every time you change a template or answer.
    • Template admin. More versions to approve, track and keep in sync.
    • AI usage, if an agent writes the replies. Your model provider bills you for it, and the language you write in can change how many tokens a reply takes.

    Meta’s current rules for India, from its pricing page: since 1 October 2026, each business phone number gets 1,000 free service messages a month, and Meta charges from the 1,001st. After the free tier, service messages and utility templates are charged ₹0.115 each, and marketing templates ₹0.8631.

    A worked month for the Pune lab

    Assume 2,000 customer-started chats a month. The bot uses about four service messages per chat, one of which is the language menu, so it sends 8,000 service messages. The lab also sends 1,500 report-ready utility templates, a day or more after the test, when the window has closed. It’s on ChatMitra Pro.

    ItemCalculationMonthly cost
    Meta: service messages8,000 − 1,000 free = 7,000 × ₹0.115₹805.00
    Meta: report-ready templates (utility)1,500 × ₹0.115₹172.50
    ChatMitra Pro planMonthly fee, includes 1,000 credits₹999.00
    ChatMitra conversation fee3,500 conversations − 1,000 included = 2,500 × ₹0.20₹500.00
    Total₹2,476.50

    The language menu accounts for 2,000 of those 8,000 service messages, or ₹230 of Meta’s charge. Remembering each customer’s language and skipping the menu for repeat customers recovers part of that. ChatMitra’s ₹0.20 is a conversation fee, charged once per customer per 24 hours however many messages are exchanged, so the menu doesn’t add to it. The 3,500 conversations are the 2,000 chats plus the 1,500 report alerts, each on a separate day. The cost the table doesn’t show is a native speaker’s review time, at launch and after every content change.

    To run your own numbers, use the WhatsApp pricing calculator or read our guide to WhatsApp Business cost per message.

    Handing over to a human who speaks the language

    WhatsApp’s Business Messaging Policy says you may use automation in the 24-hour window, but you must have “prompt, clear, and direct escalation paths”, such as a transfer to a human agent in the chat, a phone number, an email address, a support page, a store or a support form. For a multilingual bot, the escalation path has to work in the customer’s language too.

    Practical rules:

    • Know who speaks what. Keep a simple roster of which team members read and write each language, and during which hours.
    • Be honest when nobody is available. “Our Tamil-speaking team replies between 10 am and 6 pm. You can also write in English and we’ll answer now” is better than silence.
    • Pass the context. The person taking over should see the language tag and the bot conversation, so the customer doesn’t repeat themselves.
    • Don’t switch languages silently. If a person must reply in English after a Gujarati bot conversation, say why.

    Testing before you launch

    Run every check in every language, on real phones.

    • Every menu option leads where it should, in each language.
    • The language choice is saved, and the next conversation starts in that language.
    • “Change language” works from every menu.
    • Short replies (“ok”, “1”, “haan”, a thumbs-up) don’t break the flow.
    • Every template is approved in every language you’ll send, and your sender falls back correctly when one isn’t.
    • Button and list labels fit without being cut off.
    • RTL messages display in the right order.
    • A native speaker has read every message in context, inside the chat, not in a spreadsheet.
    • The handover reaches a person, or an honest message, in each language.

    How to tell whether it’s working

    Track a few numbers per language, not just overall:

    • Language mix: the share of chats in each language. A language under 5% may not be worth maintaining.
    • Menu drop-off: customers who receive the language menu but never tap it. A high rate suggests the menu is in the way. Try detection for first messages.
    • Fallback rate: how often the bot fails to understand and repeats its menu or hands over. Compare languages: a high rate in one usually means missing keywords or weak translations.
    • Handover rate and wait time by language. If Tamil chats wait twice as long for a person, the bot isn’t the problem.
    • Template failures by language, especially error 132001, which points to a missing or unapproved translation.

    Building it in ChatMitra

    ChatMitra can run each layer of a multilingual bot, with some manual setup. Here’s what the product does today, and what it doesn’t.

    Language menu and branching (Auto Reply, Pro plan). A rule can answer a greeting with a button or list message. A tapped option arrives as text the next rule matches, so you chain rules: “हिंदी” sends the Hindi main menu, each Hindi item sends its answer. Keywords match exactly, by how the message starts, or by a pattern, ignoring capitals and extra spaces. Devanagari, Tamil and Gujarati keywords work as written. There is no “contains” match, so cover Hinglish spellings with a pattern, and keep button titles unique across languages, as in the Pune example.

    Remembering the language. A rule can add a tag to the contact when it fires, so the “हिंदी” rule can tag the customer lang_hi. ChatMitra has no dedicated language field; tags do that job.

    AI Chatbot (Pro plan). You connect your own key for one of 18 AI providers and write the instructions, including a language rule like the one above. It sees recent conversation history, so it can stay in the customer’s language, but it doesn’t detect or force a language itself: your instructions decide. It reads typed text only, not button taps, voice notes or photos. With Auto Reply also on, a typed message that matches a rule gets two answers, so decide which one owns which messages.

    Templates and campaigns. Templates can be created in 71 languages, including all ten Indian languages on Meta’s list, with several language versions under one name. A campaign sends one template in one language, so a three-language reminder is three campaigns, each targeting one language tag, plus an English one for untagged customers. Through ChatMitra’s developer API, each request names its template language, so your system can choose per customer.

    What ChatMitra doesn’t do:

    • Detect a customer’s language automatically.
    • Translate messages in the inbox or translate your templates for you.
    • Provide a visual drag-and-drop flow builder for chatbot menus. Menus are built from chained Auto Reply rules.
    • Send Odia, Assamese or any other language Meta doesn’t support for templates.

    Pro is ₹999 a month and includes Auto Reply and the AI Chatbot; see pricing for current plans. For the build basics, our guide to WhatsApp interactive messages covers buttons and lists, and how to build a WhatsApp chatbot covers the rest.

    Summary

    A multilingual WhatsApp chatbot works when every layer speaks the customer’s language: the bot’s replies, the templates you send outside the 24-hour window, and the person who takes over. Pick your languages from your own chats, check them against Meta’s template list, ask once with a menu written in each script, save the answer on the contact, and let an AI agent handle free-form and Hinglish messages with a clear language rule. Create every template in every language and handle missing translations yourself, because WhatsApp won’t. Language doesn’t change Meta’s price, so the real investment is a native speaker’s review time, and that’s what makes customers feel the business really speaks their language.

    To see how an AI agent fits into this, read our guides to the WhatsApp AI agent and AI agents versus chatbots.

    Sources: Meta for Developers: Supported languages, Templates overview, Error codes, Send messages, Interactive reply buttons, Interactive list messages, Pricing. Meta: Meta Terms for WhatsApp Business Platform. WhatsApp: Business Messaging Policy, Introducing message translations, Help Center: translate messages. Office of the Registrar General, India: Census 2011, Language (C-16). KPMG and Google: Indian Languages – Defining India's Internet.

    Clear answers

    Frequently Asked Questions

    Yes. Replies inside the 24-hour customer service window are free-form, so a bot can answer in any language and script WhatsApp can display. Templates sent outside the window are limited to Meta's supported list, which includes Hindi, Bengali, Marathi, Telugu, Tamil, Gujarati, Urdu, Kannada, Malayalam and Punjabi, plus several English variants. Odia and Assamese are not on it.

    It can ask, detect or remember. Asking means a button or list menu on the first message. Detecting means an AI model reads the customer's words and answers in the same language, which is unreliable on short messages like 'ok' or 'hi'. Remembering means saving the choice as a tag or field on the contact so the bot and your campaigns use it next time. The most robust setup asks once and remembers the answer.

    Create the template once per language under the same name, each with its own language code, for example order_update in hi, gu and en_US. Meta reviews every language version separately and each one counts against your template limit. When you send, you pass the name plus the exact language code. If that language isn't approved, the message fails with error 132001.

    No. Meta's documentation says template strings and variables are not translated, so you must write every language version yourself, including the sample values for variables. There is also no automatic fallback to another language when a translation is missing.

    Treat romanised Hindi as its own input style. Keyword rules need every common spelling ('haan', 'han', 'haa') or a pattern that covers them, while an AI model usually understands Hinglish without extra work. For replies, follow the customer's script: if they write Hindi in Latin letters, answer the same way unless they've chosen Devanagari in your language menu.

    Not per message. Meta prices by message category and the recipient's country, not by language. In India, bot replies are service messages: each number gets 1,000 free a month, then each one costs ₹0.115. A language menu adds one extra service message per new conversation, and you pay for translation and review work. On ChatMitra, the ₹0.20 platform fee is charged per conversation (once per customer per 24 hours) on Starter and Pro, in any language.

    Sometimes. WhatsApp added on-device message translation in 2025, but it is available only on certain devices and languages (on Android, Hindi is the only Indian language listed at launch), and the feature is not available in the WhatsApp Business app. Don't rely on it: write your bot in the customer's language.

    Yes, with some manual setup. On the Pro plan, Auto Reply rules can send button and list menus, chain rules on the option a customer taps, and add a language tag to the contact. The AI Chatbot uses your own model key and follows a language instruction you write. Templates can be created in 71 languages, and you send one campaign per language using tags. ChatMitra does not detect language automatically or translate messages for you.

    Janki Gohil

    ChatMitra

    Janki Gohil writes for the ChatMitra blog about the WhatsApp Business API.

    Ready to try WhatsApp Business API?

    Start free on the Starter plan — free subscription. Pro is ₹999/month with a 15-day trial when you need automation.