WhatsApp's business messages run inside a metered window

The conversation

Meta's WhatsApp Business Platform is documented here as it stood on 16 September 2026, rather than as it was at any one launch date. Meta's own developer documentation, read through an archived copy after the current live page did not load during this research, describes two ways a business connects: a hosted Cloud API, which Meta calls the preferred solution due to its ease of implementation and low maintenance, and an On-Premises API that a business must host itself. Either way, a business must also use the Business Management API, which the documentation says is used to manage the WhatsApp Business Account and message templates.

What the documents show

What the platform actually restricts, and charges for, is set out in Meta's own pricing rules. Meta's Conversation-Based Pricing documentation states that WhatsApp charges per conversation, not per individual message, and that conversations are 24-hour message threads between a business and its customers. A conversation opened by a business sending a pre-approved template message is billed under a category such as marketing or utility; a conversation opened because a customer messaged first, and the business replies with an ordinary message, instead falls into a service conversation tied to what the documentation calls a customer service window. Separately, Meta's opt-in documentation states that businesses are required to obtain opt-in before sending proactive business-initiated messages to customers outside the 24-hour window, and requires that any opt-in method clearly name the business and state that the person is agreeing to receive WhatsApp messages from it.

The system boundary

The 24-hour customer service window is the platform's central mechanical boundary: inside it, a business, or an automated system acting for it, can send free-form replies to a customer who has recently messaged; outside it, only a pre-approved message template can restart contact, and only if the recipient has opted in. An automated assistant built on this platform cannot simply message a customer whenever its own logic decides to; Meta's rules, not the bot's design, set when a proactive message is even possible, and templates must clear the platform's own review before they can be sent at all.

Where it fails

The documentation is unambiguous that opt-in and the service window are enforced by WhatsApp's own systems, including a business quality score Meta says it will use to rate limit businesses if quality is low for a sustained period. A builder treating WhatsApp like an open messaging channel, rather than a rate-limited, template-gated one, risks exactly that kind of throttling.

• Which conversation category, and which billing rate, applies to each type of automated message the assistant sends?
• Can the team show, for every proactive message template, when and how the recipient's opt-in was actually captured?
• What happens to an in-progress automated conversation if the 24-hour service window closes before the exchange is finished?

None of this describes an assistant's intelligence; it describes a metered, permissioned messaging layer that any automated system built on top of it has to respect.