We tested what AI writes for WhatsApp Business

2026-10-10 · 6 min de leitura · Bruno Vieira

The first text your customer receives on WhatsApp is already worth money. Since October 1, companies on the official WhatsApp Business API pay for service replies after 1,000 free messages per month; those using only the app on their phone still pay nothing. The lesson applies to every business: the fewer back-and-forths to solve the same question, the better.

The question here is direct: does AI already write a business's first contact well? To find out, we tested ZapToMe's automatic message generator, audited its entire library of ready-made texts and measured the results. Below is what we found and what to adjust before publishing.

The reply got more expensive, but only on the official API

Here is the context: the new charge hits the official API, the layer connecting WhatsApp to CRMs, chatbots and service systems. Before, any reply inside the 24-hour window opened by the customer was free; now, each service message delivered after the monthly allowance of 1,000 per number costs money, according to Forbes Brasil. In Brazil, the rate is about R$ 0.035 per delivered message. Messages sent by customers remain free, and businesses on the WhatsApp Business app pay nothing extra.

The heaviest detail is fragmentation. A "Hi", a "How are you?", a "Here's the link" and a "Any questions, I'm here" add up to four charges when they could be one. On the official API, every fragment sent counts as a separate charge. Answering everything in the first message became a saving.

How we tested

We audited the full library of ready-made messages in ZapToMe's generator: 8 languages, 4 tones of voice, 3 focuses per type (engagement, information and lead capture) and 3 variations of each text, with the same fixed input in every scenario. That is 576 messages measured one by one: length, emojis, formatting, questions and the presence of an action verb. We also captured an AI generation on the same day, with a restaurant scenario in Portuguese.

A method note: during the test, the public endpoint's AI mode was unstable and the tool fell back to the reserve library, the behaviour expected when the AI does not answer. The pricing facts were checked with Forbes Brasil and Infobae; the metrics carry a date.

What the 576 ready-made messages deliver

The library is solid on the basics and discreet on the rest. None of the 576 messages goes over 159 characters: quick to read on any phone. No placeholder is left empty, and 79% use at least one emoji, one per message on average. They are safe texts, ready to paste.

• Clarity first. Short and direct, easy to read between tasks.
• Explicit action in the minority. 27% carry a clear action verb; among lead-capture ones, 31%.
• Formatting almost absent. Only 6% use bold; none use lists, italics or strikethrough.
• Room to reply. One in four ends with a question.

The library delivers clarity and safety, but leaves you the two things that convert most: formatting and a clear call to action. That is not only its flaw: no generator knows your tone or what you want the customer to do next.

What changes when AI does the writing

The sample we captured showed the difference in temperament. The six AI-generated messages stayed between 170 and 199 characters, with bold on business hours and concrete calls in the lead ones, like inviting the customer to place an order. Instead of "how can we help?", the text pulls the conversation to the next step.

Always revise before publishing; a sample does not guarantee every generation. The two paths complete each other: AI drafts, the library guarantees the conversation never goes silent. It is the same fit as Meta's shift to automatic messages.

A good automatic message does not feel automatic. It feels like someone read your message and replied.

The 4 fixes that matter before publishing

1. Add a call to action. End by naming the next step: ask for the name, confirm the address, pick a time. It is missing in 7 of every 10 ready-made messages.
2. Put the essentials in one message. Hours, address and what you need from the customer in the same text. On the API, this also cuts charges.
3. Format with restraint. Bold on hours and prices guides the eye. The walkthrough is in the guide on formatting messages to sell more.
4. Test like a customer. Send it to yourself, read it on your phone and only then set it up in the app, under Automatic messages.

Answering everything you can in the first message is what separates a conversation that moves from one that drags. ZapToMe's generator creates greetings and away messages in seconds, with AI or the ready library, and shows the character count so you can adjust the length without guessing.

Frequently asked questions

Do businesses using the free WhatsApp Business app have to pay the new charge?

No. The charge applies to the official API, used by operations connected to CRMs, chatbots and service platforms. Those answering manually on the WhatsApp Business app still pay nothing, and people writing to a business are never charged.

AI or a ready-made message, which should you choose for the first contact?

Use AI for a contextual text and the library as a safety floor when it stumbles. In both cases, revise with the checklist. To move up a level, AI agents in customer service are the natural evolution.

How many automatic messages should you set up?

At least two: greeting and away. The generator delivers three variations of each, and you pick the one that fits the moment. Review after every change of hours, holiday or promotion.

Do automatic messages push customers away?

No, as long as they are useful. What pushes people away is delay and replies in pieces that do not solve anything. A greeting that already states the hours and how to get help keeps the conversation alive. To complete the path from contact to order, see WhatsApp sales links.