WhatsApp AI Chatbots, What They Can Do and How to Get Started

WhatsApp is the largest messaging platform on earth, with more than 3.3 billion monthly active users as of early 2026 and over 150 billion messages a day. More than 200 million people message a business account on it daily. It is also the most heavily governed channel available to a business, with a verification process, a template approval queue, a message taxonomy, and a pricing model that changed materially in July 2025.

Those two facts explain the shape of most WhatsApp chatbot projects: the potential is enormous and the setup is the longest of any channel. This guide covers what an AI chatbot can actually do on WhatsApp, the rules that determine what it costs, and how to work out whether it belongs on your roadmap or below it.

Where WhatsApp Is Not Optional

The decision to build here depends almost entirely on geography.

In India, Brazil, Indonesia, Mexico, and much of the Middle East and Southern Europe, WhatsApp is not one channel among several. It is the default way people contact a business, ahead of phone and far ahead of email. A business serving those markets without a WhatsApp presence is not choosing a channel mix, it is choosing to be harder to reach.

In the United States and the United Kingdom the calculus is different. WhatsApp has around 100 million users in the US, which is substantial in absolute terms and a minority of the market, and consumer habit still routes toward web chat, email, and SMS. For a US-only business, WhatsApp is usually the third or fourth channel to build, not the first.

The Split That Confuses Everyone: App Versus Platform

There are two entirely different WhatsApp products for businesses, and most confusion in this area comes from conflating them.

The WhatsApp Business App is a free mobile app for small businesses. It gives you a business profile, a catalogue, labels, and simple canned replies described as automation. It is genuinely useful for a one-person operation. It does not have an API, so you cannot connect an AI chatbot to it. No integration, no exceptions.

The WhatsApp Business Platform, formerly called the Business API, is the programmatic product. It has no app of its own; it is an interface your systems talk to. This is the only path to an AI chatbot on WhatsApp, and it is where all the verification, template, and pricing rules live.

If a vendor offers to connect an AI bot to your existing WhatsApp Business App, they are either using an unofficial automation method that risks the number being banned, or they are describing a migration to the Platform without saying so. Both are worth clarifying before you sign anything.

The 24 Hour Service Window, and Why Support Is Nearly Free

This is the single most important rule on WhatsApp, and it happens to be extremely favourable to support use cases.

When a customer messages you, a 24 hour customer service window opens. Inside that window you can send free-form messages, and since the July 2025 pricing change, service messages inside an open window are free. Not discounted. Free.

Outside the window, you cannot send a free-form message at all. You can only send a template, which is a pre-written message approved in advance by Meta, and templates are billed per message according to their category.

That produces a clean rule of thumb: reactive support on WhatsApp costs nothing at the messaging layer, and proactive messaging costs money. An AI chatbot answering inbound questions lives almost entirely in the free half.

The Four Message Categories

Since 1 July 2025, Meta bills per message rather than per conversation, priced by template category and by the recipient's country code. The old allowance of 1,000 free conversations no longer exists.

Two details that catch people out. Marketing template rates have moved in both directions by market since the change, falling in some countries and rising in others, so a cost model built on 2024 numbers is unreliable. And your actual bill includes your provider's markup on top of Meta's rate, typically a fraction of a cent per message for larger providers and more for smaller ones.

The design implication is direct. Keep conversations inside the service window and your AI messaging costs are dominated by your model inference, not by WhatsApp. Build flows that push outside the window and you have created a per-message cost line that grows with your success.

What an AI Chatbot Can Actually Do Here

WhatsApp supports a richer message vocabulary than most channels, and the useful subset for a support bot is smaller than the full list.

Genuinely valuable:

• Free-form text answers from your knowledge base, inside the service window. The core case.
• Interactive reply buttons, up to three per message, which are excellent for disambiguation and far better than asking someone to type a number.
• List messages, a menu of up to ten options, useful for topic routing at the start of a conversation.
• Media in and out. Customers send photos of damaged items and screenshots of errors constantly on this channel. A bot that cannot at least acknowledge and route an image is missing a large share of real traffic.
• Location sharing, which matters for delivery, service areas, and store lookups.

Worth being careful with:

• Voice notes. Widely used by customers in several of the largest WhatsApp markets. Transcription is possible and adds cost and latency; decide deliberately rather than discovering the volume later.
• Catalogue and cart messages. Powerful for commerce, and they turn a support bot into a sales system with a much larger surface to get wrong.
• Templates. Every one needs approval, and rejections are common for anything that reads as promotional in a utility category.

Getting Started, Realistically

The path has more gates than any other channel, and the waiting is front-loaded.

Step 1. A Meta Business account, verified. Business verification requires documentation proving the business exists and that you represent it. This is the step that takes the longest and the one people underestimate.

Step 2. A dedicated phone number. It must not be active on the consumer WhatsApp app or the Business App. Migrating an existing number to the Platform is possible and irreversible in practice, so most businesses use a new number.

Step 3. A display name that survives review. Meta reviews the name customers will see, and names that do not match the verified business, or that read as generic keywords, get rejected.

Step 4. Choose your route. Either through a Business Solution Provider, which is the normal path and gets you a working integration in days, or directly against the Cloud API, which is free to call and means you own the integration, the retries, the media handling, and the template management.

Step 5. Submit templates and wait. Anything you want to send outside the service window needs approval before you can use it. Approval is not instant and rejections are normal, so build this into the schedule rather than discovering it at launch.

Step 6. Connect your chatbot to the inbound webhook and the send endpoint, and treat the whole thing as a channel on the same knowledge base rather than a separate bot.

Realistically, budget two to four weeks from decision to live traffic, with most of that spent waiting on other people. That compares with an afternoon for Telegram and a couple of weeks for Messenger, where the app review is the bottleneck.

Design Notes for WhatsApp Specifically

Several habits transfer badly from web chat.

The customer's phone number is your identity key, and it is genuinely useful. Unlike Instagram handles or page-scoped IDs, a verified WhatsApp number often matches a record you already hold. It is still not an authentication factor on its own, and confirming order details on the basis of a matching number is a decision to make consciously rather than by default.

Read receipts and typing indicators exist and set expectations. Marking a message read and then taking two minutes to reply reads worse than not marking it. If your pipeline is slow, do not mark read early.

Answers should be shorter than you think. WhatsApp renders in a narrow column on a phone held in one hand. Two short paragraphs is a long message. Use buttons rather than asking people to type choices.

Media arrives unannounced. A customer sends a photo with no text and expects you to understand. At minimum, acknowledge, ask one clarifying question, and route to a human if the answer depends on the image.

Groups are not a support surface. The Platform does not support group messaging, and that is a mercy rather than a limitation.

Compliance, Opt-In, and Disclosure

WhatsApp enforces consent more strictly than any other channel, and the enforcement is commercial as well as policy-based.

Opt-in is required for templates. You need demonstrable consent to message a customer outside a conversation they started, collected on any channel, with the business clearly identified. Sending templates to a list you bought is the fastest route to a quality rating collapse and eventually a blocked number.

Quality rating is a live score. Blocks and reports from recipients lower it, and a low rating reduces your messaging limits and can suspend the number. This is another reason a reactive support bot is the safest possible first use case: people who messaged you rarely report the reply.

Disclose that it is AI. Article 50 of the EU AI Act has been enforceable since 2 August 2026 and requires systems interacting with people to make clear they are machines. WhatsApp's intimacy makes this more important rather than less, because the register of the channel implies a person.

Message content is personal data in a high-trust context. Messages are end to end encrypted in transit between the user and the platform endpoint, but once they reach your systems they are ordinary data with ordinary obligations. Retention, access control, and a privacy notice that names the channel.

Should You Build It Yet?

A short decision framework, honestly applied.

That last row is the one worth dwelling on. Every channel article ends up at the same place: the channel is a distribution problem and the answer quality is a content problem, and adding distribution to a bot that answers badly simply distributes the bad answers more widely.

Where the Platform Fits

Worth being direct about scope. Paperchat currently supports Telegram, Facebook Messenger, and Instagram as channels on a chatbot, and does not yet offer WhatsApp; the Platform requires provider enrolment with Meta that sits ahead of it on the roadmap. What does transfer today is the part that determines whether any of this works: one knowledge base behind every channel, one inbox for the conversations, and one set of escalation rules, so that adding a channel later is a connection rather than a second bot with its own answers. If you are planning a WhatsApp deployment now, that is still the architecture to insist on from whichever provider you use, because the alternative, a separate WhatsApp bot with its own content, diverges from your website answers within a month.

The Bottom Line

WhatsApp is the largest channel available and the slowest to open. The Business App cannot be automated; the Business Platform can, and it carries business verification, a dedicated number, display name review, and template approval before you send anything.

Once you are through that, the economics favour exactly the thing an AI chatbot is good at. Replies inside the 24 hour service window are free, so a bot answering inbound questions costs you inference and nothing else, while templates outside the window are billed per message and demand consent and approval. Build for the free half: fast answers, buttons instead of typed choices, short messages, media acknowledged rather than ignored, and a clean escalation to a person well inside the window.

Then judge the whole thing on geography. If your customers already message businesses on WhatsApp, this is not a channel experiment, it is the front door.