Meta starts charging for the reply on 1 October. Here's who actually pays.

On 1 October 2026 something small changes in Meta's billing, and something large changes in the economics of every WhatsApp tool you've been pitched this year. The reply stops being free.

Since November 2024, when a customer messaged you first, you could answer them free-form for the next 24 hours and Meta charged nothing. That window is where almost all real customer conversation happens — the questions, the haggling, the "is this still available." From 1 October every message you send inside it is billed individually, at the same per-message rate Meta already charges for a utility or authentication template in your country. Utility templates sent inside that window, free since July 2025, get billed too.

Meta announced this in July and said it will publish the per-market rates by 1 September. So at the time of writing nobody has the final numbers — including the vendors already emailing you about it.

Before you do anything else, check whether it applies to you at all. For most of the businesses I talk to, it doesn't.

Who actually pays

This change only touches the WhatsApp Business Platform, the API. It does not touch the app.

Most small businesses in Indonesia and Malaysia are in row two. They have one number, one or two admins, a phone, and no Meta invoice at all. Nothing in this announcement costs them a rupiah.

Expect the next few weeks of marketing to blur that line. "WhatsApp pricing is changing on 1 October" is true and useless — the sentence that matters is which WhatsApp. If an email about this arrives with a demo booking link at the bottom, that's the sentence it will be missing.

What the bill looks like, before and after

Take an ordinary conversation. A customer messages you asking if something is in stock; you answer; they ask the price; you answer; you offer to book them in; you send a booking confirmation template. Five messages from you and the customer, all inside the 24-hour window.

To put a number on it: utility and authentication messages in Indonesia currently run about Rp357 each before 11% VAT. If service messages land at the same rate, as Meta says they will, that conversation goes from free to roughly Rp1,400 plus tax.

Trivial once. Run three thousand conversations a month and it's several million rupiah for traffic that cost you nothing in September. And unlike templates, service messages get no volume discount — templates step down by up to 25% as you scale, service messages never do. The bill grows in a straight line forever.

Why this lands hardest on AI agents

Here's the part the AI chatbot vendors won't put in the subject line.

Meta bills the reply regardless of who wrote it. Your admin, or a third-party AI agent — same message, same charge. Which means chattiness is now a line item.

An admin who reads the question and answers it properly costs you one billed message. An AI agent that greets, asks a clarifying question, offers a menu, misreads the answer, apologises, and escalates costs you six — and then a human still has to answer the original question. Every failure mode a bot has is now metered, and its most common failure mode is talking too much to say too little.

Then there's the asymmetry underneath it. Meta's own AI is on a different meter entirely: Meta Business Agent replies are billed per token, at $2 per million tokens, since 1 August. A third-party AI's replies are billed per message, as service messages, from 1 October. Whatever the intent, the effect is that the company setting the price of your AI vendor's replies also sells a competing AI priced its own way.

That's not a reason to panic. It's a reason not to build your customer relationship on a meter someone else controls.

The cost-control playbook

1. Work out which row you're in. Log in and look. If you're on the API only because a vendor sold you a dashboard you use twice a week, price out what the free app plus a bit of discipline actually costs you. For a lot of two-admin operations the honest answer is that the API was never the bottleneck.

2. Count turns, not conversations. Your bill is now rate × number of messages you send. The single biggest lever isn't negotiating with your provider, it's answering the actual question in the first reply. Rewriting your five most common answers so each one is complete — price, availability, what happens next — takes an afternoon and cuts turns for years.

3. Audit your bot's transcripts for filler. "Hi! 👋" as its own message. "Let me check that for you." "Is there anything else I can help with?" That padding was free. From October each one is a billed message that answered nothing. Merge them or delete them.

4. Use the free entry point on purpose. Conversations that start from a click-to-WhatsApp ad or a Page call-to-action still open a 72-hour window where everything you send is free. If you're already buying ads, routing enquiries through that entry point is the one discount left on the table.

5. Don't move follow-up onto the API by reflex. Chasing quiet leads — our whole thing — is low-volume, high-value, and needs to sound like a person. Forty personal messages a month, sent by your admin from the app, cost nothing. Putting them through the API to get "automation" is paying per message for the privilege of sounding automated.

6. Recheck on 1 September. That's when the real rates land. Don't sign an annual contract off anyone's estimate before then, including this one.

Why we don't build an AI that replies for you

This is a good moment to say plainly why Dokwise doesn't do the thing people expect an AI WhatsApp product to do.

We argued about it internally long before this pricing change, and we decided against auto-reply for reasons that had nothing to do with cost. An AI answering your customer is confidently wrong in front of the one person you can't afford to be wrong in front of, and you find out about it days later. Customers can tell, and the ones worth having tell each other. And the person who actually knows your business — which patient is nervous, which buyer is stalling because of their spouse, which price you'd bend — is your admin. Standing an AI in front of that person doesn't scale them, it hides them.

So Dokwise reads the conversations you already have and tells the human what to do next. Who went quiet, who's warm, who asked something you never answered, what to say to each. Your admin reads the list, decides, and sends it from their own phone in their own words.

That costs nothing per message. Not because we found a loophole, but because a person sending a WhatsApp message from the WhatsApp Business app has never cost anything, and still won't on 2 October.

The pricing change didn't create that difference — it just put a price tag on it. AI that replaces your admin is now metered per message by Meta. AI that helps your admin isn't. We didn't pick that side because we saw this coming. We picked it because a small business's real advantage over a big one is that a human being remembers the customer, and automating that human away trades away the only thing you were better at. The October bill just makes being right about it cheaper.

What to do this month

• Check whether you're on the app or the API. Most people are wrong about this.
• If you're on the API, pull last month's conversations and count messages sent, not conversations opened. That's your new bill.
• Read your bot's transcripts with a red pen. Every message that didn't answer anything is now money.
• Wait for 1 September before signing anything annual.
• If you're on the app: nothing changes for you. Ignore the emails.