An AI chatbot on WhatsApp Business is an assistant that answers your customers on WhatsApp using a language model connected to your information (catalog, policies, orders) and systems, with handoff to a person when needed. It runs on the WhatsApp Business Platform (Meta's Cloud API), which has its own rules on message templates, a 24-hour customer service window, per-message pricing and AI use. For US companies serving customers in Latin America or Spanish-speaking customers at home, this guide explains how it all works and what you need to launch it compliantly.
Why WhatsApp matters for Latin American and Hispanic customers
Across much of Latin America, WhatsApp is where customers already message businesses to ask about prices, coordinate deliveries and request support; email and SMS often go unanswered. Many Spanish-speaking customers in the US also prefer it. If your company sells into Mexico, Central or South America, or serves Hispanic customers, a WhatsApp channel meets them where they are.
The challenge is scale and language. A well-built AI chatbot answers repetitive questions instantly, in Spanish or English, around the clock, and leaves your team the cases that need judgment. It is not a rigid menu tree: it understands free text, answers from your real data and knows when to hand off.
How the WhatsApp Business Platform works
The WhatsApp Business Platform is the API version of WhatsApp for businesses: instead of replying from a phone app, your system sends and receives messages programmatically. Its main parts:
- Cloud API: Meta's API for sending messages (text, media, buttons, lists) and receiving customer messages via webhooks that deliver each inbound message to your server as JSON.
- WhatsApp Business Account (WABA) and phone number: a dedicated number tied to your company's business portfolio in Meta.
- Solution partners: Meta partners that handle onboarding, billing and tools such as shared inboxes. Optional; you can integrate with the Cloud API directly.
- Message templates: pre-approved messages, the only type you can send outside the customer service window.
Templates, the 24-hour window and pricing
The customer service window opens when a customer messages you and lasts 24 hours from their latest message. Inside it you can reply with free-form messages, including your chatbot's answers. Outside it, you can only start a conversation with an approved template.
Since July 1, 2025, Meta charges per delivered template message instead of per conversation. Rates depend on the template category and the recipient's country code, so a message to a Mexican number is billed at Mexico's rate, not the US rate:
| Message type | Typical use | Charge per Meta's documentation |
|---|---|---|
| Free-form replies inside the 24-hour window | The chatbot answering questions | Free |
| Utility template | Order confirmation, shipping update | Free inside the window; charged outside it |
| Authentication template | One-time passcodes | Charged |
| Marketing template | Promotions, re-engagement | Charged (highest category) |
When a customer arrives through a click-to-WhatsApp ad, Meta also opens a 72-hour free entry point window for all messages. Rates change quarterly, so check the official pricing page before budgeting. For a support bot, most of the work happens inside the 24-hour window with no Meta messaging fee; the main variable cost is the AI model, covered in our guide to the cost of implementing AI.
What Meta allows when it comes to AI on WhatsApp
This changed in 2026. Meta's terms for the WhatsApp Business Platform now include a section on AI providers: they prohibit using the platform when AI itself is the primary product offered, such as general-purpose "ask me anything" assistants from model providers. Per Meta's documentation, that restriction has applied since January 15, 2026.
What remains allowed is what a business actually needs: a chatbot that uses AI incidentally to serve its own customers, for support, sales, bookings, order status or product questions. In practice:
- Scope the assistant to your business topics and have it politely decline anything else.
- Always offer a clear path to a person: WhatsApp's Business Messaging Policy requires direct escalation paths when you use automation.
- Do not use conversation data to train third-party AI models; the terms restrict it.
- Re-check the current terms before launch, since Meta updates them often.
Architecture: language model, RAG and CRM
A business-grade WhatsApp chatbot follows this flow:
- The customer writes; the Cloud API sends the message to your backend via webhook.
- The backend looks up the customer in your CRM by phone number and loads context (orders, open tickets).
- An enterprise RAG layer retrieves relevant passages from your catalog, policies and FAQs, in Spanish and English.
- The language model drafts the reply and, where appropriate, calls tools: check stock, open a ticket, book an appointment.
- The reply goes out through the Cloud API and the conversation is logged in the CRM.
This is the integration work our AI development services focus on: the model is one piece; the value comes from wiring it to your data and CRM. If the bot takes actions rather than just answering, apply the controls from our AI agents implementation guide: least privilege and human approval for sensitive steps. And because customer text goes straight to the model, defend it against prompt injection.
Human handoff without losing context
Handoff is what separates a helpful bot from a frustrating one. Define explicit triggers:
- The customer asks for a person.
- The model finds no answer in your documents or has low confidence.
- Sensitive topics: complaints, billing disputes, cancellations, health.
- Signs of frustration, or the same issue raised twice.
The human agent should receive a summary and the customer's details so nobody has to repeat themselves. Route Spanish conversations to Spanish-speaking agents, and after hours, tell the customer clearly when a person will reply. A nearshore team in Quito (UTC-5) can cover Spanish-language support during US business hours.
Opt-in, disclosure and privacy
Meta's rules come first: you may only start conversations with people who gave you their number and agreed to receive your messages, and you must honor opt-out requests on or off WhatsApp. Then layer the laws that apply to you and to your customers:
- Capture opt-in explicitly (web form, unchecked box or confirmation message) and keep timestamped evidence.
- Separate service consent from marketing consent.
- Disclose up front that the customer is talking to an automated assistant. California's bot disclosure law (Business and Professions Code section 17941) prohibits using an undisclosed bot to mislead people in order to incentivize a sale.
- US state privacy laws such as the CCPA/CPRA govern the personal data you collect; customers in Latin American countries are also protected by local laws, such as Ecuador's LOPDP or Brazil's LGPD.
- Confirm your model provider does not train on your data, collect only what you need and set a retention period for transcripts.
Metrics that show whether it works
- Automated resolution rate: conversations closed without a human and without the customer returning on the same issue.
- First response time and total resolution time.
- Handoff rate and reasons: they reveal gaps in your knowledge base.
- Accuracy: a weekly sample of conversations reviewed against your policies, in both languages.
- Conversion: quotes, bookings or orders started in chat.
- Cost per conversation: model tokens plus Meta template fees.
Frequently asked questions
Can I put ChatGPT on my company WhatsApp?
You can use models such as GPT or Claude inside a chatbot for your company, as long as it serves your business (support, sales, bookings) rather than being offered as a general-purpose assistant. Since January 2026, Meta's terms prohibit general-purpose AI chatbots on the WhatsApp Business Platform but allow AI that supports a business's own customer service. Scope the topics and offer human handoff.
How much does an AI chatbot for WhatsApp Business cost?
There are three costs: development, AI model usage and Meta template fees. Replies inside the 24-hour window carry no Meta fee; templates outside it do, by category and recipient country. At MisterProSoft, a prototype on your real data starts at USD 4,000 and takes 2 to 4 weeks; production cost depends on CRM and system integrations.
Do I need a solution partner to use the WhatsApp API?
No. You can integrate directly with Meta's Cloud API, which Meta hosts and bills for. A solution partner can help with onboarding, billing on a credit line and tools such as shared team inboxes. The choice depends on whether you have in-house engineering and which agent-facing features your support team needs.
Can a WhatsApp chatbot handle both Spanish and English?
Yes. Modern language models handle Spanish and English well, and a bot can reply in the language the customer uses. The work is in the knowledge base and testing: keep approved content in both languages, evaluate answers in each, route handoffs to agents who speak the customer's language and make sure templates are approved in every language you send.
Does a WhatsApp chatbot have to disclose that it is a bot?
It is best practice and, in some cases, legally required. California's bot disclosure law prohibits using an undisclosed bot to mislead people in order to drive a sale, and other jurisdictions are adding transparency rules. A short opening message saying the customer is chatting with an automated assistant, with an option to reach a person, builds trust and reduces complaints.
Conclusion
An AI chatbot on WhatsApp Business works when it handles concrete business topics with your real data, follows Meta's rules on templates, opt-in and AI use, respects privacy law on both sides of the border and hands off to a person at the right moment. Measured properly, it frees your team from repetitive work without hurting the customer relationship.
To bring it to your company, explore our AI development services or book a free consultation, and we will design scope, integrations and metrics with you, in Spanish and English.