Reply to WhatsApp customers in their own language automatically
An AI reply can answer WhatsApp customers in the language they wrote in, because the model reads the message and can be told to match it. A fixed text cannot detect a language, so split by language with an AI-picks-the-path step, a keyword automation per language, or one bilingual text. Approved templates exist per language code, such as pt_BR or es_MX, and an automation step sends one code.
easyCoexistence runs these automations on the WhatsApp Business number you keep in the app, through Meta's Cloud API: AI replies on your own key that follow the customer's language, keyword sets per language, and templates by language code.
A rule added under the Instructions of each AI step tells the model to write in the customer's language.
Keywords per automation. Matching ignores capital letters but not accents, so preço and preco are different words.
Spanish alone has several template language codes in Meta's list; Portuguese (Brazil) is pt_BR, English (US) en_US.
Meta's error when a template does not exist in the requested language or is not approved.
The automations editor shows its own labels in English, Portuguese and Spanish; other locales see English.
How can an automatic reply know the customer's language?
Only from what the customer writes. An incoming WhatsApp message carries the text, the sender's profile name and number, not a language setting, and a country code is a weak hint: +1 covers many Spanish speakers, and +41 covers four languages. That leaves several ways to answer in the right language, which can sit side by side on one number:
How do AI replies match the customer's language?
By default. Under your Instructions, easyCoexistence adds a short block of rules to each AI reply, and one of them tells the model to write in the customer's language. So a message in Portuguese gets a Portuguese answer and the next customer's Spanish gets Spanish, from one automation. Write the Instructions in the language you write most precisely, and give the local words customers use, such as Pix, boleto or CEP in Brazil and transferencia or factura in Spanish-speaking countries. Two lines help in multilingual chats:
Answer in the language of the customer's latest message.
If you cannot tell the language, answer in Portuguese and Spanish,
one short line each.
Prices are in BRL for Brazil and in USD everywhere else.How do you set fixed replies per language?
With one keyword automation per language, each holding that language's words and a Reply with text in that language. Use Contains ("Anywhere in the message") and list accented and unaccented spellings, because matching ignores capitals but not accents. Words shared by two languages, such as "menu" or "info", should live in one automation with a bilingual reply: when two keyword automations match, the oldest answers and the other never does.
What about welcome, away and catch-all texts?
First message, After hours and the catch-all fire whatever the text says, so a fixed reply on them cannot pick a language by itself. Use one of three fixes. Swap the Reply with text for Reply with AI, which answers in the customer's language and can still say the same thing; or add an AI picks the path step with one path per language, each ending in its own Reply with text, so the wording stays exact; or write a bilingual text, the language of most customers first, kept to two short lines so it reads well on a lock screen:
- Welcome: "Olá {{
first_name}}! Obrigado pela mensagem. / ¡Hola {{first_name}}! Gracias por escribirnos." - Away: "We're closed and open at 9:00. / Estamos cerrados, abrimos a las 9:00."
- A catch-all hand-off: "A person will reply here soon. / Uma pessoa vai responder aqui em breve."
How do templates work across languages?
A template is a name plus a language, and each translation is submitted and reviewed by Meta on its own, so a welcome_coupon template may exist in pt_BR and es_MX but not en_US. A Send a template step takes one Language code and sends exactly that one: there is no fallback to another language. If that translation is missing or not approved, Meta refuses the send with error 132001, the run ends Failed and you get an email. Because First message cannot see the language, pick the code most new customers read, or send templates from somewhere that knows the customer: ask your AI to call send_template with the right code, or have your system send through the Cloud API with the number's credentials.
When should your own system pick the language?
When you already know your customers, or need more than the AI's judgment. Call my webhook POSTs the message text, the customer's number and profile name and the chat's last 10 messages to your URL, signed in X-EC-Signature. Your system can look the number up in your CRM and use the language stored there, or run a language detector on the text, and return {"reply": "..."} in that language within 20 seconds; easyCoexistence sends it on the chat. The same payload lets n8n route Spanish chats to a Spanish-speaking teammate. A stored preference beats detection for short messages such as "ok" or a single emoji, which no detector can place.
Common mistakes
- Listing only accented keywords. "cuánto" misses customers who type "cuanto", and "preço" misses "preco"; add both spellings to the same automation.
- Putting "menu" in an English and a Portuguese keyword automation. The older one answers it for everyone, in its language; move shared words into one bilingual reply.
- Choosing a template language code that has no approved translation. Meta refuses the send with 132001 and the run fails; check the code with get_template_status first.
- Writing "Answer only in English" in the Instructions while expecting Spanish answers. The model then receives two conflicting rules; name the languages you want instead.
Build them in the editor, in English, Portuguese or Spanish, or ask ChatGPT or Claude through the connector. US$ 9 down to US$ 0.75 per number per month, falling as you add numbers, with the first 3 days free.
easyCoexistence is a multichannel API and MCP server that connects AI agents such as Claude and ChatGPT to WhatsApp, Instagram, TikTok, Facebook and more. On WhatsApp it uses Coexistence: the business keeps using its number in the WhatsApp Business app while the AI works through Meta's official API.
Frequently asked questions
Does the AI translate my fixed texts?
No. Reply with text sends exactly what you wrote, and the AI step is separate. If a fixed answer must reach several languages, write a version per keyword automation, or one path per language after an AI picks the path step, or replace it with Reply with AI and put the facts in the Instructions.
Can I test the replies in each language before going live?
Yes. Run test with a sample in each language: "¿Cuánto cuesta?", "Quanto custa?", "How much is it?". The AI step really runs and shows its reply; nothing is sent. A keyword automation's Test it box shows whether each sample matches.
Can customers send voice notes in their language?
They can. With an OpenAI or OpenRouter key saved, easyCoexistence writes the voice note out and the AI reads "[voice message]" followed by the words, in any language the transcription model understands. Without such a key the AI sees only "[voice message]", so tell it in the Instructions to ask for the question in writing, in any language.
Does replying in several languages cost more?
Not on easyCoexistence's side: the price is per number, whatever the language. Your AI provider bills tokens, which vary with the language and length of the text, and Meta bills messages by category and the customer's country, to your own account.
Keep reading
Ready to get started?
Set up WhatsApp Coexistence in minutes, not months. The app keeps working on the phone.
Start Free TrialVerified on