WhatsApp Service Messages Become Billable on 1 October 2026: What It Costs and What to Do Before Then

Since November 2024, replies your business sent inside the 24-hour customer service window have been free. That ends on 30 September 2026. From 1 October, every service message you send is billed individually, at the same per-message rate your country already pays for utility and authentication templates - and that includes replies typed by human agents and generated by third-party AI assistants.

Last verified: 6 August 2026 against Meta's published pricing notices and partner announcements. Meta has said it will publish the exact per-country service-message rate card before 1 September 2026. Treat every figure in this guide as indicative until you have checked your own rate card, and re-run your numbers when it lands.

Two things are worth separating immediately, because they are frequently conflated:

• This is not a rate rise. The per-message price of utility and authentication templates is not what changed. What changed is that a category of message which used to cost nothing now falls inside the meter.
• Inbound messages are still free. You are not charged when a customer messages you. You are charged for what you send back.

The practical effect is that WhatsApp support - which for nearly two years was effectively free once a customer started the conversation - becomes a variable cost that scales with how chatty your bot and your agents are. For a team handling a few hundred conversations a month, that is a rounding error. For a support operation handling tens of thousands, it is a line item that deserves a forecast before October rather than a surprise in the November invoice, and the metric that forecast should be built on is cost per conversation rather than a flat per-message estimate, because the two diverge sharply once you account for how many messages a real resolution takes.

Understanding the change is much easier if you know which model you are currently budgeting against, because many teams are still working from assumptions set two models ago.

The July 2025 shift is the one that still catches people out. Under the old conversation model, a support exchange of fifteen messages cost the same as an exchange of two. Under per-message billing, message count is the cost driver - which is precisely why the October change matters so much more than the headline suggests. Support conversations are long. A resolved support thread might run twenty outbound messages where a marketing blast is one.

The October change is actually two exemptions being withdrawn at once, and most coverage of it only mentions the first. Service messages - free-form replies from a human agent, a bot, or a third-party AI assistant - lose their free status, which is the headline everyone reports. Less widely reported: utility templates sent inside an already-open 24-hour window also lose the free pass they picked up in the July 2025 change. Before October, a business that replied to an open conversation using a utility template rather than a free-form message paid nothing either way. After October, both routes are billed, which removes what had become a minor optimisation for cost-conscious teams. One exception survives untouched: the 72-hour free entry-point window that opens when a customer starts a conversation from a Click-to-WhatsApp ad or a Facebook or Instagram call-to-action button stays free for its full 72 hours regardless of message type - if a meaningful share of your conversations start from paid social, that is worth knowing before you rebuild your cost model from scratch.

From October, your WhatsApp bill is determined by how each outbound message is categorised - a free-form service message, or one of the three message template categories. Getting the template categorisation wrong in either direction is expensive: misclassify a marketing message as utility and Meta will recategorise it, misclassify a genuine utility message as marketing and you pay several times more than you need to.

Two details in that table carry most of the financial weight.

Marketing has no volume discount by design. Meta has been consistent that promotional blasting should stay expensive regardless of scale. If your WhatsApp costs are already uncomfortable and most of your volume is marketing, no amount of growth will earn you a better rate.

Service messages are reported to carry no volume discount either, despite being charged at the utility rate. That combination is worth modelling carefully: a high-volume support operation pays the same per-message price on message one and message one million. This is the detail most likely to be clarified when the September rate card lands, and the one worth checking first.

Indicative per-message rates for 2026 sit roughly in these bands. Verify against your own rate card before budgeting - rates vary by country, are revised periodically, and Meta's service-message card is not published until September.

Now apply those to a realistic support thread. A customer asks where their order is. Your bot greets, asks for the order number, confirms it, returns the tracking status, offers to help further, and closes politely. That is six outbound messages - a conservative count for a bot with any conversational warmth.

All of that is currently zero.

The number that should concentrate the mind is the one in the middle column of the second row: a mid-sized support operation goes from nothing to roughly USD 18,000 a year, purely from a policy change, with no growth in volume. And because the cost is per message rather than per conversation, it scales with your bot's verbosity - a design choice nobody has previously had a financial reason to examine.

The cost impact of this change is not evenly distributed. It tracks directly with how many outbound messages a typical resolved conversation takes in your vertical, which is why the same policy change is a rounding error for one business and a real line item for another.

This table is reasoning, not a rate card - it is a starting hypothesis for where to look first, not a substitute for measuring your own median outbound-messages-per-conversation figure per decision one below. If your business spans several of these categories, budget for the highest-exposure segment and treat the rest as upside.

Separately from the October change, Meta began charging for its own AI assistant - Meta Business Agent, the conversational AI Meta launched for WhatsApp, Instagram and Messenger in mid-2026 - on a per-token basis from 1 August 2026. The free evaluation period is already over as of this writing.

Multiple industry reports converge on a rate of roughly USD 2.00 per million tokens, which they translate to something in the region of 4 to 5 cents per typical conversation. Meta has not published that per-token figure on its own pricing pages at the time of writing, so treat it as a widely reported estimate rather than an official rate card entry, and confirm the number in your own Meta Business Agent billing dashboard before budgeting against it.

This matters because the two changes together alter a decision many teams made during the free period. Per-token pricing means your cost varies with the length of the conversation and the size of the context the model processes, which is a fundamentally different and less predictable shape from per-message billing. A single complex support conversation - one where the agent pulls in a longer knowledge base article, or carries a lot of prior conversation history as context - can consume many times the tokens of a simple one, and unlike per-message billing, you cannot estimate that cost just by counting message bubbles.

If you adopted Meta Business Agent while it was free, this is the moment to price it against the alternative: running your own AI layer and sending the output as ordinary service messages. Under that model you pay your LLM provider for tokens and Meta for the messages, rather than Meta for both. Depending on your provider, your volume and your average conversation length, the difference can go either way - but it is now a real comparison rather than a free-versus-paid one, and it deserves a spreadsheet rather than a guess.

Worth noting for context: Conferbot does not run on Meta Business Agent, and does not offer a retrieval-augmented or freeform AI-generation layer at all - flows are built explicitly in a visual editor, so the cost of a Conferbot-built conversation on WhatsApp is the per-message service rate described above, with no separate per-token meter running underneath it.

These are ordered by how much they save relative to the effort involved.

1. Count your outbound messages per resolved conversation

This is the single most valuable number you can produce, and most teams do not have it. Pull last month's logs and compute the median count of outbound messages per conversation. Multiply by your country rate and your monthly conversation volume. That is your October bill, near enough.

Do this before optimising anything. Teams routinely discover their real figure is nine or twelve messages rather than the four they assumed, because greetings, typing acknowledgements, menu re-prompts and sign-offs were never counted. If you do not already log every outbound send with a timestamp, your webhook delivery-status events are the fastest way to reconstruct last month's numbers without waiting on a BSP export - each delivered message fires a status event you can aggregate directly.

2. Merge multi-bubble replies

Under conversation billing, splitting a reply into three short bubbles was free and felt more human. From October each bubble is a separate charge. A bot that says "Got it!" then "Let me check that for you" then "Your order is out for delivery" now costs three times what one well-written message costs, for the same information.

This is the highest-yield change available and it is a flow edit, not an engineering project. Audit your flows for consecutive sends and merge anything that does not need a user response between the parts. Cutting six outbound messages to four is a 33% reduction in your entire WhatsApp support bill.

3. Stop sending messages that carry no information

Standalone acknowledgements - "Thanks!", "One moment", "Sure thing" - were free social lubricant and are now billable. Fold the acknowledgement into the message that follows it. Typing indicators, which are not billable, do the reassurance job that a standalone "one moment" used to do.

The same applies to closing messages. A "Is there anything else I can help with?" followed by "Thanks for contacting us!" is two charges to end a conversation that was already finished.

4. Move deflectable volume to channels you do not pay per message for

WhatsApp is now a metered channel for support, and it is the only one of your channels that is. A website widget, a Telegram bot, an in-app SDK and a help centre all cost nothing per message.

This does not mean abandoning WhatsApp - it is where your customers are, and that is worth paying for. It means being deliberate about which volume belongs there. Order-status lookups, FAQ answers and password resets are cheap to deflect to self-service and expensive to handle over WhatsApp at six messages each. Complex, high-value or emotionally charged conversations are exactly what the paid channel is for. In practice this looks different by vertical: an ecommerce business might keep order tracking on WhatsApp because customers expect it there but move general product FAQs to the website; a logistics operation might do the opposite, since shipment status is exactly the high-frequency, low-complexity traffic worth deflecting. There is no universal answer, only a universal method: measure, then move the cheapest-to-deflect volume first.

A bot that runs the same flow across WhatsApp, a website widget and Telegram makes this a routing decision rather than a rebuild. That is the practical argument for omnichannel right now: it turns channel cost into something you can steer.

5. Re-examine your template categorisation

Marketing rates are materially higher than utility rates and carry no volume discount. Templates that are genuinely transactional - a delivery notification, an appointment reminder, a receipt - should be categorised as utility. Templates that were written with a promotional line tacked on the end will be categorised as marketing, and that one sentence can multiply the cost of the whole send.

Review your highest-volume templates and strip promotional content from genuinely transactional ones. If a template needs to sell, make it a separate marketing template sent to people who opted in for that, rather than contaminating a utility send. If you are not sure why a specific template was categorised the way it was, our guide to WhatsApp template rejections covers the transaction test Meta actually applies, and running drafts through the free WhatsApp template checker before submission catches the mechanical mistakes that cost money later.

Do not try to keep conversations open to avoid template costs. The 24-hour window is no longer a cost boundary for service messages - it stopped being free on 1 October. Sending filler messages to hold a window open now costs money and achieves nothing. Any flow logic built around window-extension should be deleted rather than adjusted.

Do not switch to unofficial WhatsApp APIs to dodge the fees. Unofficial libraries and modified clients violate WhatsApp's terms, and the outcome is a permanently banned business number - usually the number printed on your packaging, your invoices and your storefront. The savings are real and the risk is existential. This is not a close call; see our guide to WhatsApp Business account restrictions for how hard that specific mistake is to reverse.

Do not assume your BSP's price equals Meta's price. Business Solution Providers frequently add a per-message markup on top of Meta's rate, or bundle messages into plan tiers with their own economics. When the October change lands, some providers will pass through the exact Meta rate and some will apply their markup to a category that previously had no cost to mark up. Ask your provider, in writing, what your service-message rate will be from 1 October and whether it includes a markup. Ask before September, while you still have time to move if the answer is bad - switching a BSP under deadline pressure, with a migration to run and contracts to review, is a worse negotiating position than switching two calm months early.

For nearly two years, WhatsApp support was effectively free once a customer initiated. That produced a generation of flows designed with no cost discipline at all - chatty bots, multi-bubble replies, generous re-prompts - because there was no reason to design them any other way.

From 1 October, every one of those design choices has a price. That is not a reason to leave WhatsApp. It is the channel with the highest open rates and the lowest friction in most of the world, and paying for support conversations there is defensible in a way that paying for ignored email is not.

What it does mean is that channel mix becomes a cost decision rather than purely a reach decision. The teams that handle this well over the next two months will be the ones that can answer three questions: how many messages does a resolved conversation take, which conversations genuinely need to happen on WhatsApp, and how quickly can we move the rest somewhere unmetered. Worth knowing while you build that forecast: billing is triggered on delivery, not on the send attempt, so a message that fails outright - see our WhatsApp API error codes reference for what a failure like 131026 or 130429 actually means - does not add to the bill. A retry loop that eventually delivers costs the same one message it always would; it just wastes engineering time getting there.

If your flows only exist on WhatsApp, that third question is a rebuild. If the same flow already runs on your website, Telegram, Instagram and your mobile app, it is a routing rule. Conferbot builds one flow and deploys it to all of them, with a shared knowledge base, shared agent inbox and shared analytics - so "which channel should absorb this volume" is a question you can actually act on before the invoice arrives.

Start free with Conferbot - the free plan includes 600 conversations a month and needs no credit card, so you can have the same flow running on a second channel this afternoon.

Start with the measurement. Everything else follows from knowing your real message count per conversation, and that number is available today.