WhatsApp service messages become paid on 1 October 2026: what an AI-agent inbox should expect

For almost two years the cheapest thing you could do on the WhatsApp Business Platform was answer a customer. Someone writes in, you reply inside 24 hours, nothing is billed. That is the assumption every 24/7 AI agent on WhatsApp was priced on, ours included.

On 1 October 2026 that assumption ends. This is what Meta’s own pricing page says, what other providers are reporting on top of it, and what it does to the bill of a business whose inbox is answered by an AI agent all day.

What Meta has actually published

Two sentences on Meta’s non-template message pricing page carry the whole change:

• Service messages are charged from 1 October 2026. They have been free since November 2024. Per market, the rate is the same as the utility and authentication rate.
• Utility templates inside an open window are charged too. Since July 2025, a utility template sent while the customer’s 24-hour window was open cost nothing. From 1 October it is billed per message like any other utility send.

Two things are not changing. Free Entry Point windows stay free: a customer who arrives from a Click to WhatsApp ad or a Facebook Page button, answered within 24 hours, opens a 72-hour window where any message type costs nothing. And none of this touches the WhatsApp Business app on a phone; it is the API only.

What is reported but not (yet) on Meta’s page

360dialog, Wati, SendPulse and YCloud have all published explainers that add two details: a free allowance of 1,000 service messages per business phone number per month, charged only from the 1,001st and reset monthly, and a requirement to have a payment method on file by 30 September 2026 or have service-message delivery stopped when charging begins. Both are plausible and consistent across providers. As of 14 September we could not find either on Meta’s pricing pages, so we are labelling them as reported. If you run through a BSP, your BSP’s notice is the one that governs your account; read it.

Why an AI agent feels this more than a campaign tool

A campaign tool sends templates. Templates were already paid, so 1 October changes nothing about a broadcast.

An AI agent does the opposite job. It sits on the inbound side and replies to everything, in free-form text, inside the customer’s window. Every one of those replies is a service message. A support inbox that used to cost only the AI plan now also costs a per-message Meta fee on each reply the agent sends, and a busy agent sends a lot of them: a clarifying question, the answer, a follow-up, a “anything else?”. Four service messages for one customer question is normal.

The bill is therefore driven by reply count, not conversation count. That is the number to start watching now, while it is still free to be wrong.

A worked example, with the rate left as a variable

We are not going to print an INR rate here, because it is set per market on Meta’s rate card and can move. Pull the India utility rate from the current rate card and call it R. Then, per phone number, per month:

• Small clinic, ~800 inbound questions, ~3 replies each: 2,400 service messages. If the reported 1,000 free allowance holds, 1,400 × R. If it does not, 2,400 × R.
• D2C brand, ~5,000 inbound (order status, returns, sizing), ~3 replies each: 15,000 service messages, so 14,000 × R under the reported allowance.
• Agency running 20 client numbers: the allowance, if it is per number, applies 20 times; the per-reply rate applies to the sum. Multi-number accounts are where the difference between “reported” and “confirmed” matters most.

Do the multiplication with your own inbound count from August. The answer is usually smaller than the panic and larger than zero, and it scales with exactly one thing you control: replies per question.

Five ways to keep the number down

1. Answer in one message, not four. Tune the agent to resolve the common question in a single reply. A clarifying question that could have been avoided is now a billed message. The single biggest lever on this list.
2. Stop the “anything else?” reflex. A courtesy closer is a service message. Let the customer close the thread.
3. Route ad traffic through the free entry point. A customer who arrives from a Click to WhatsApp ad, answered inside 24 hours, gets a 72-hour window where every message is free. That is the cheapest conversation on the platform and it is unchanged.
4. Watch the reply window. Once a customer’s 24-hour window closes, a free-form reply fails and you are into template territory anyway. Knowing how long is left on each thread keeps human agents from writing replies WhatsApp will refuse.
5. Hand off cleanly. An agent that hands a hard question to a person on the first miss sends one message; an agent that tries three more times sends four.

What this looks like in AI WhatsApp by Vfastrr

AI WhatsApp by Vfastrr already shows a live reply-window countdown on every thread in the team inbox, so an operator sees “1h left” before writing rather than an error after. The AI agent hands off to a human on its own terms, and our AI plans are priced per included reply, which is the same unit Meta is about to bill on, so the two numbers line up on one screen. We will update this page the day Meta’s rate card confirms or corrects the reported allowance.

Related

• WhatsApp daily messaging limit exceeded: what it means and how to get your tier back
• WhatsApp number quality rating dropped: why it happens and how to fix it