There is one thing the past two years have taught people who work with WhatsApp, and taught them incorrectly: the idea that replying to a customer costs nothing. Since November 2024, service messages, the free-form text replies you send within the 24-hour window, have been free. Since July 2025, utility templates sent within that same window have also been free. Within a few months, a whole generation of automated flows, chatbots, abandoned-cart recovery sequences, and customer support teams was built around a marginal cost of zero. The problem is that zero was not an inherent feature of the channel. It was a promotion.
Starting October 1, 2026, Meta will once again charge for both service messages and utility templates sent within an open customer service window. The calculation is no longer based on the old conversation-based model: before 2024, you paid once for the 24-hour window regardless of how many messages you sent. Now, you pay per message, so one reply costs what one message costs, while five replies cost five times as much. It sounds like a technical difference, but it is actually the part that rewrites the economics, because everything we have designed around conversational flows over the past two years, the bot that asks three qualifying questions, the confirmation message, the gentle follow-up six hours later, now multiplies the cost by the number of messages, not by the number of customers.
Before panicking, it is worth defining the scope. The change applies exclusively to the WhatsApp Business Platform, meaning the API. The regular WhatsApp app and the free WhatsApp Business app are not affected. If you manage conversations from your phone, nothing changes. If you manage them through ManyChat, a CRM, a shared inbox, or any official provider, quite a lot changes.
What Exactly Becomes Paid
A service message is the non-template reply you send while the window is open. The window opens whenever the customer messages you first and resets whenever they send another message. Starting in October, each of those replies will be a billed message, at the same per-message rate Meta already applies to utility and authentication templates in the recipient’s country. Rates therefore remain tied to geography: what you pay depends on where your customers live, not where your company is based.
There is a buffer, and it is the only genuinely good news in the package. Meta is introducing an allowance of 1,000 free service messages per month for each business phone number, regardless of the destination country. You start paying from message 1,001 onward. The threshold resets every calendar month, and unused messages do not roll over. The allowance is calculated separately for each number because the 1,000 free messages are not shared at the business account level. A company operating ten locations with ten separate numbers therefore gets 10,000 free messages per month, but they are distributed rigidly: the location that uses 300 does not give its remaining 700 to the location that uses 3,000.
There is no introductory discount for utility templates. They are paid from the first message of the month and are billed even when sent within an open customer service window. This is also more significant than it might seem because it is exactly the opposite of the habit many businesses have developed: using an approved template inside the window to play it safe with moderation. Starting in October, that cautious choice becomes the more expensive of the two, since the service message has a free allowance and the utility template does not.
The Detail Almost Nobody Is Looking At
And here comes the part that should matter most to anyone running advertising. A service message remains free within the special 72-hour free messaging window, which opens when a customer messages you through a Click-to-WhatsApp ad or a call-to-action button on a Facebook Page. That window has not been affected by the update, meaning conversations generated by an ad still get a cost-free starting period.
Do you realize what that means? Starting October 1, two identical conversations with the same customer about the same question can cost different amounts depending on where the first message originated. If the person found you on Google, saved your number, and messaged you organically, each of your replies consumes your allowance and then your budget. If the same person arrived by clicking an ad, you get 72 hours of free conversation. Meta does not put it quite this way, but the implied message is fairly clear: organic WhatsApp becomes a cost center, while paid acquisition becomes the cheaper entry point. This is not the first time we have seen this pattern, and it will not be the last.
Then there is the AI chapter, which operates on a separate track and is worth keeping clearly distinct. Since August 2026, the Meta Business Agent, Meta’s AI assistant, has been billed by token, at two dollars per million tokens, which in practice means around four or five cents per response, including delivery. If your automations run through a third-party provider, that calculation does not apply to you and you pay for service messages instead. If they run through Meta’s native agent, the economics are different. It is worth checking this now because these are two different pricing models, and during the months when they overlap, the bill can become difficult to interpret after the fact.
What to Do Over the Next Few Weeks
The operational question is not “how much will it increase?” It is “how many service messages do I send each month, per number?” Until you have that figure, any estimate is just guesswork. The Pricing Analytics API returns the volume of service messages over a selected period by querying pricing_category: SERVICE, while delivery web hooks show them with category: service. Anyone working with a BSP can request the same data without touching the API. The goal is to reach October 1 with a number in hand, not a feeling.
Everything else follows from that number. If you are below 1,000 per month per number, nothing happens and the issue is essentially closed, with the only precaution being to watch for seasonal spikes that could push you over the limit in an isolated month. Above 1,000, the exercise becomes more architectural than pricing-related: how many messages does the bot really need to qualify a lead, how many courtesy messages exist out of habit rather than necessity, and how many utility templates sent within the window could become free-form replies that consume the allowance instead of budget? Every unnecessary message that has gone unquestioned over the past two years simply because it cost nothing now has a price that multiplies with volume.
Finally, there is the boring but decisive part: the wallet. If you have accounts with automatic top-ups calibrated to old volumes, that threshold was based on a world in which replies did not carry a cost. Raising it before October 1 takes five minutes. Discovering halfway through the month that a flow has stopped because a conversation is still open costs much more, and not just in euros.
Anyone who works with WhatsApp professionally has already seen this transition once, when the platform moved from the conversation-based model to the per-message model in 2024. Anyone who joined the channel during the past two years is about to discover that the economics on which they built their flows were, in effect, a 23-month-long trial period.
