A WhatsApp food delivery chatbot is an AI agent that shows the menu, builds the order, confirms address and payment, and announces each stage of the delivery. In 2026, the safe path is the WhatsApp Business API, with approved templates, handoff to human agents, and a cost per delivered message.
This guide shows how AI for food delivery works in restaurants, quick-service spots, and delivery-only kitchens: catalog, WhatsApp Flows, order status, payments, POS integration, peak hours, post-delivery follow-up, Meta’s costs, and data protection.
Key Takeaways
- Catalogs, WhatsApp Flows, and order messages let a restaurant sell on WhatsApp without making customers download another app.
- Order confirmations and status updates are utility templates; adding a coupon turns the message into marketing, which in Brazil costs about nine times more.
- From October 1, 2026, the agent’s replies become billable, with 1,000 free service messages per phone number per month.
- Every automated flow needs a fast way out to a human agent, a requirement of the WhatsApp Business Messaging Policy.
What Is a WhatsApp Food Delivery Chatbot?
The term WhatsApp food delivery chatbot covers everything from numbered-menu bots to agents that converse in natural language and query systems. Understanding the difference helps a restaurant choose what it really needs.
From Menu Bot to AI Agent
A menu bot responds to fixed options, such as “1 for the menu, 2 for opening hours.” It works for predictable questions but stalls when the customer writes “I want a large half pepperoni, half margherita pizza, no onions, and a Coke.”
An AI agent interprets free-form sentences, asks for whatever is missing, and records the order in the system. It is the natural evolution of the WhatsApp chatbot, and the step-by-step for building one is in the guide on how to create AI agents for WhatsApp.
Why Use the Official API for Delivery
The WhatsApp Business Platform was built for businesses to talk to customers at scale through APIs, with text, media, interactive messages, and templates. The app is still useful at the counter, but on its own it cannot connect an AI agent in an authorized way.
Tools that imitate WhatsApp Web violate the terms and can get the number banned. In food delivery, losing the number on a Saturday night means losing orders, so the lower risk of the official API outweighs the apparent savings.
How AI for Food Delivery Works on the Official API
Behind the conversation are three pieces: the Cloud API, which receives and sends messages; the AI agent, which decides the reply; and the restaurant’s systems, which hold the menu, orders, and customers.
Webhooks, the 24-Hour Window, and Templates
Every incoming message and every delivery status reaches the server by webhook, as the WhatsApp Cloud API guide explains. If the server does not acknowledge receipt, Meta resends the event for up to seven days, so the system needs to ignore duplicates.
When the customer writes, a 24-hour customer service window opens in which the agent replies freely. Outside it, only approved templates can be sent, which applies to order notices, surveys, and campaigns.
Meta’s Rules for AI Agents
The WhatsApp Business Messaging Policy allows automation but requires prompt, clear, and direct escalation paths to a human. The same policy prohibits asking for full payment card numbers, bank account numbers, or national ID numbers.
In practice, a delivery agent should have a closed scope: menu, order, delivery, payment, and questions about the restaurant. Anything outside that, serious complaints, and special requests go to the team, a model detailed in the guide to human and AI customer service.
The Menu on WhatsApp: Catalog and Product Messages
The menu is the starting point of any food delivery chatbot. On the official API, it can be sent as a browsable catalog instead of a PDF or an image the customer has to zoom in on.
How to Build the Restaurant’s Catalog
According to Meta’s catalog documentation, each item has a name, description, price, images, and availability, and the catalog is created in Commerce Manager or uploaded through the API. Products go through automated review under the Commerce Policies.
Formats include the full catalog, a single product, a carousel, and a multi-product message that shows up to 30 items organized in sections. For delivery, sections can follow the menu: appetizers, mains, drinks, and desserts.
- Availability: marking a sold-out dish as unavailable prevents orders the kitchen cannot fulfill.
- Lightweight photos: images sent through the API are limited to 5 MB, so standardize the size.
- Clear descriptions: ingredients, portion size, and allergens reduce repeated questions to the agent.
From Cart to Order in the System
The customer can add up to 99 units of each item to the cart, and the order reaches the business by webhook. On the Cloud API, the cart icon does not appear at the top of the conversation, so the agent needs to remind the customer to complete the purchase.
The same catalog, cart, and order flow is used in retail e-commerce. The difference in food delivery is timing: the order must reach the kitchen within seconds, with complete notes.
Orders and Addresses with WhatsApp Flows
Not every order fits in a cart. Building a half-and-half pizza, choosing how a steak is cooked, or entering the delivery address works better in a guided form inside the conversation.
Screens for Order, Address, and Notes
WhatsApp Flows create screens with text fields, lists, options, and dates. A delivery Flow can have one screen to build the item, another for the address and landmarks, and a final summary screen before confirmation.
According to the official Address Messages page, the API’s native address message is only available to businesses and customers in India. Elsewhere, the address is collected through a Flow, through free text interpreted by the agent, or by sharing a location.
Flow Requirements and Best Practices
The Flows documentation requires a verified business and high message quality to send Flows. They do not open on linked devices such as WhatsApp Web, so the agent needs a text-based fallback.
When a Flow queries live data, such as the delivery fee by neighborhood, it uses an endpoint that must respond within 10 seconds. The official best practices recommend one task per screen, completion within five minutes, and a summary screen.
Order Confirmation and Status with Utility Templates
After ordering, the customer wants to know whether the order was accepted, when it leaves, and when it arrives. These messages can go out automatically, triggered by the order system at every status change.
Utility vs. Marketing in Food Delivery
Under Meta’s categorization rules, order confirmations, shipping updates, and order status are utility templates. Utility cannot promote, recommend, or offer products, and a coupon in the “out for delivery” notice turns the template into marketing.
The official library already includes ready-made templates, such as order confirmation and delivery confirmation, and the approval decision comes within 24 hours. Details on variables and buttons are in the guide to WhatsApp message templates.
The Order Message Sequence
A lean sequence keeps customers from asking “where’s my order?” and reduces service volume during peak hours:
- Order received: a summary with items, total, and estimated time.
- Being prepared: a short notice when the kitchen accepts the order.
- Out for delivery: the courier’s name or a tracking link, if the system has one.
- Delivered: a confirmation with a button to report a problem.
If the customer is still chatting, the 24-hour window is open and the notice can be a free-form message. Once the window has closed, only the approved template gets through.
Payments on WhatsApp: Pix and Payment Links in Brazil
Closing the payment in the same conversation shortens the order and cuts down on calls to confirm amounts. Meta has a payments API specific to Brazil, with its own rules; in other markets, the usual path is a payment link generated by the restaurant’s payment provider.
How an Order with Pix Works
In the Brazil payments API, the business sends an order_details message with items, total, and a unique reference code. Options include dynamic Pix (Brazil’s instant payment system), payment links, boleto, and card payments outside the app.
With Pix, the customer copies the code and pays in their banking app; with a link, payment happens in the browser. The order status can be updated as pending, processing, shipped, completed, or canceled.
To request payment after the 24-hour window has closed, there is an order details template, which can be approved as utility or marketing depending on the message content.
Reconciliation and Sensitive Data
The documentation is clear: WhatsApp does not reconcile payments. The restaurant checks each transaction with its payment provider using the order’s reference code, preferably automatically.
The agent should never ask for a card number in the conversation, which the policy prohibits. For charges after delivery, such as invoiced corporate orders, payment reminders can go out as utility templates, as long as they contain no promotional content.
Integration with the Order System, POS, and CRM
A food delivery chatbot is only useful when it talks to the system the kitchen already uses. Without integration, someone has to retype every order, and the time savings disappear.
What to Integrate First
The most common order is to start with whatever prevents rework and expand later:
- Menu and prices: read from the system so the agent does not offer sold-out items.
- Orders: the order is created in the POS or delivery system, with notes.
- Status: each change triggers the right template.
- Customers: saved address, history, and preferences, as with an AI agent integrated with the CRM.
The integration usually relies on the order system’s API or on webhooks. When the system has no API, a spreadsheet or an intermediate database can work at first, with human review.
Securing the Agent’s Actions
An agent that creates orders and changes customer records needs limits. Recommended practices include least privilege, authorization in the backend, and human approval for high-impact actions such as refunds or cancellations.
Malicious messages may try to convince the AI to apply discounts or reveal other customers’ data. This kind of attack, known as prompt injection, is best contained with rules enforced outside the model, such as prices and discounts that only the order system can set.
Peak Hours and Human Handoff
Food delivery concentrates orders in a few hours of the day. That is when the agent helps most, and also when a failure costs the most, so test the peak-hour flow before Friday night.
Platform Capacity at Peak Times
The Cloud API’s default throughput is 80 messages per second per phone number, and the documentation also cites a pair rate limit of one message every 6 seconds to the same user. Combining the reply into a single message is better than sending three in a row.
At peak times, the agent should prioritize the essentials: confirm the order, give the real wait time, and announce when the kitchen hits capacity. Promising 30 minutes when the queue is at an hour leads to complaints after delivery.
When to Hand Off to an Agent
Some triggers call for immediate handoff: a wrong or incomplete order, a delay beyond the estimate, allergies, event orders, and an angry customer. The agent receives the full history and does not need to ask everything again.
With several agents on the same number, conversations sit in queues by topic, as shown in the guide to WhatsApp for multiple agents. The AI agent handles the repetitive work and the team takes care of the exceptions.
Post-Delivery Feedback and Customer Win-Back
The order does not end at delivery. What happens next determines whether the customer orders through WhatsApp again or looks for another channel.
A Satisfaction Survey After Delivery
A survey tied to the transaction, sent shortly after delivery, is classified as utility. A 1-to-5 rating with buttons and an optional comment field is enough.
Low scores should open human follow-up right away, before they turn into a public review. High scores can get a simple thank-you, without a coupon, so the template stays in the utility category.
Win-Back with Opt-In and Ads
Customers who have not ordered in weeks can receive offers through marketing templates, as long as they have opted in. Segmentation and frequency matter: sending only to people with a relevant order history keeps the campaign useful rather than intrusive.
Meta may limit how many marketing templates each person receives from all businesses combined, and error 131049 indicates that the limit was reached. Poorly targeted sends lower the number’s WhatsApp quality rating, which affects the whole operation.
Another entry point is Click-to-WhatsApp ads: a conversation opened by an ad unlocks 72 hours of free messages, templates included. A QR code on the packaging, with a pre-filled message, also brings the customer back to the channel.
WhatsApp Food Delivery Chatbot Costs from October 2026
Meta changes API billing on October 1, 2026, and food delivery is affected because it runs on customer service replies. The details are in the guide to WhatsApp API pricing.
Meta’s Rates for Food Delivery
Meta charges per delivered message, by message category and by the recipient’s country, according to the official pricing page. From October 1, 2026, in Brazil, Meta charges R$ 0.3217 per marketing message and R$ 0.035 per delivered utility, authentication, or service message.
| Category | Use in Food Delivery | Rate per Delivered Message (Brazil) |
|---|---|---|
| Marketing | Promotions, daily menu, win-back | R$ 0.3217 |
| Utility | Confirmation, status, post-delivery survey | R$ 0.035 |
| Service | Agent replies within the 24-hour window | R$ 0.035 (1,000 free per number/month) |
| Authentication | Login codes for an app or website | R$ 0.035 |
Each phone number gets 1,000 free service messages delivered per month. Conversations started from Click-to-WhatsApp ads stay free for 72 hours, and messages sent from the app under coexistence are not charged.
How to Lower the Cost per Order
The first step is to count messages per order. A flow that uses eight replies to close a pizza order costs twice as much as one that uses four, and the agent can combine questions into a single message.
It also helps to separate utility from marketing carefully, use Click-to-WhatsApp ads for campaigns, and, for Brazilian numbers, simulate monthly volume in the cost calculator (Brazil, in Portuguese). The chosen platform’s fee is separate from Meta’s rates.
Coexistence, Data Protection, and Opt-In in Food Delivery
Many restaurants already have a number customers know on the WhatsApp Business app. Changing numbers or losing the history is usually the biggest concern when migrating.
Keeping the App at the Counter with Coexistence
With WhatsApp coexistence, the same number works in the app and on the API, with the last 180 days of history synced. The agent serves on the API while the team keeps replying from the phone at the counter.
There are points to watch: according to Meta, the app’s catalog, orders, and labels stay in the app and do not carry over to the Cloud API. The primary phone also needs to be opened regularly, because about 14 days without use drops the connection.
Consent and Data Protection
Replying to someone who started the conversation is part of customer service; sending promotions afterward requires opt-in. Meta’s opt-in guidance asks that the person agree to receive messages and know the business’s name, as the guide to WhatsApp opt-in explains.
Brazil’s data protection law (LGPD) gives individuals the right to withdraw consent through a free and easy procedure and places on the business the burden of proving consent was given; similar rules apply under GDPR in the EU. Addresses and order history should be kept only as long as they are needed.
How to Launch and Measure a Food Delivery Chatbot
Rolling out a WhatsApp food delivery chatbot works best in short stages, starting with the flow that takes up most of the team’s time and expanding as the numbers show results.
Step-by-Step Rollout
- Map the conversations: list the most frequent questions and the points where orders get stuck.
- Connect the number: activate the official API, with coexistence if the app is already in use.
- Organize the menu: a catalog with up-to-date prices, photos, and availability.
- Get templates approved: confirmation, status, delivery, and survey as utility.
- Integrate the order system: orders created in the POS and status changes triggering notices.
- Test and adjust: run during a quiet period before switching it on at peak times.
When choosing a platform, compare integration, handoff, reporting, and total cost, criteria gathered in the guide to WhatsApp AI tools.
Metrics That Show Results
The most useful customer service metrics in food delivery are time to first reply, orders completed without an agent, handoff rate, mid-order abandonment, messages per order, and the post-delivery survey score.
Also track order errors, Meta’s cost per order, and repeat purchases within 30 days. These numbers show where to adjust the prompt, the menu, or the integration, without relying on the team’s impressions.
ConverZap: AI Agents for Food Delivery on the Official API
ConverZap offers an AI agent platform for WhatsApp on the official API, with coexistence, template-based messaging with opt-in, and agents that take actions in CRMs and databases. The agent can be configured with each operation’s menu and rules.
Restaurants and chains that want to automate WhatsApp orders can talk to the ConverZap team and design the delivery flow around their volume and the systems they already use.
FAQ: WhatsApp Food Delivery Chatbot
The WhatsApp Business app is free, but it cannot connect an AI agent in an authorized way. On the official API, customer service replies include 1,000 free messages per phone number per month; above that, Meta charges per message, on top of the platform’s cost.
No. Catalogs, carts, Flows, and order messages work inside the conversation. Payment happens through a link that opens in the browser or, in Brazil, through Pix completed in the customer’s banking app.
The Cloud API receives audio messages normally. Transcribing and interpreting the audio depends on the agent’s platform, and it is wise to confirm the order in text before sending it to the kitchen.
Text messages arrive normally, but Flows do not open on linked devices, and templates with four or more buttons do not appear on WhatsApp desktop. The agent should offer the option to place the order by text.
Meta states that a template review is completed within 24 hours. Using templates from the official library, such as order confirmation and delivery confirmation, reduces formatting errors that lead to rejection.
Yes. Inside the 24-hour window opened by the customer, the agent can suggest add-ons in a free-form message. What it cannot do is include offers in utility templates, such as the confirmation or the out-for-delivery notice.
