WhatsApp AI Auto Reply: How to Set Up One That Actually Sells

A WhatsApp AI auto reply is an assistant connected to your WhatsApp Business number that reads each customer message and answers it from your own information — prices, products, policies and FAQs — instead of sending a fixed “thanks, we’ll get back to you”. To make one that sells, feed it real facts, write clear “never guess” rules, let it ask a few qualifying questions, and hand every serious buyer to a person.

Most WhatsApp enquiries are the same dozen questions: how much is it, do you have it, where are you, do you deliver. A good auto reply answers them and moves the buyer forward; a bad one sends a canned line and lets the lead go cold.

What are the three kinds of WhatsApp auto reply?

“Auto reply” covers three very different tools. The free WhatsApp Business app has greeting and away messages plus quick replies: one fixed text, whatever the customer asked. Menu and keyword chatbots, built on the WhatsApp Business Platform (the API), send “reply 1 for prices” or tappable buttons and follow a decision tree. AI agents read each message and answer from the knowledge you gave them. Meta now promotes its own AI agent on its Business app features page, and independent platforms run AI agents on the API.

An away message suits a one-person shop. When enquiries outpace you or arrive after hours, you need more people or the third kind. Choosing between the app and the platform? Read WhatsApp Business app vs API.

Does Meta allow AI chatbots on WhatsApp Business?

Yes — when the AI serves your own business. Since 15 January 2026, the Meta terms for the WhatsApp Business Platform bar “AI Providers” — developers of large language models, generative AI platforms and general-purpose AI assistants — from using the platform when that technology is the primary functionality rather than an incidental one, as Meta decides. The target is the “ask me anything” assistant, not a shop answering its own customers.

The same terms let a business retain an AI provider as its solution provider, as long as WhatsApp data is not used to train or improve AI models beyond one fine-tuned for that business’s exclusive use. Meta’s AI Providers page states that the change does not alter how Meta charges other businesses, and lists the few markets where Meta must legally allow general-purpose assistants under separate rules.

Two rules from the WhatsApp Business Messaging Policy also apply:

• Automation needs a way out. You may use automation when replying inside the 24-hour customer service window, but you must also offer prompt, clear and direct escalation paths, such as an in-chat transfer to a human agent, a phone line or email.
• The 24-hour window. The window opens and resets with each message the customer sends. Inside it you can reply without a template; outside it, only approved message templates are allowed.

What should you feed an AI chatbot for WhatsApp Business?

The model already knows how to write; it does not know your prices, delivery areas or refund rules. Give it facts:

• What you sell — products, services, packages and variants, and what each includes.
• Prices — real numbers and what changes them: deposits, delivery fees, installments, extras.
• Policies — payment, cancellation, returns, exchanges and warranty, worded as you would tell a customer.
• Real questions and objections — the most common questions from a month of chats with your answers, including “why are you more expensive than the shop next door?”
• Hours, branches and coverage — addresses, delivery areas and when a person is available.
• Files you already have — price lists, brochures, menus and catalogues as PDF, Word, Excel or CSV.

Write facts, not slogans:

Weak: “We deliver everywhere, fast, at the best prices.”
Useful: “Delivery inside Cairo and Giza costs EGP 60 and takes 2 to 3 working days; Alexandria costs EGP 90. We do not ship outside Egypt.”

The second answers three questions and leaves nothing to invent. When a price changes, change it in the knowledge base too, or the AI keeps quoting the old one.

How do you write “never guess” rules for an AI auto reply?

The most expensive AI mistake is a confident wrong answer: a discount you never offered, a delivery date you cannot meet, a medical opinion. “Never guess” rules name the topics where the AI must not answer from its own reasoning; instead it says it needs to check and passes the question to a person. Start with these:

• Discounts and special prices. “Never offer or accept a discount, a free extra or a price match. Say a colleague will confirm.”
• Medical, legal and financial questions. “Never say whether a treatment suits someone, a contract is safe or an investment will pay off. Offer a consultation.”
• Delivery dates. “Give only the standard delivery times in the knowledge base. Never promise a specific day for a specific order.”
• Stock you cannot see. “If a product or size is not listed, do not say it is available.”
• Policy exceptions — refunds outside the policy, custom orders, bulk prices, payment plans.

Write each rule as an instruction, not information: “we sometimes give returning customers a discount” is a fact the AI may repeat to everyone; “never mention discounts” is a rule. Then decide what happens when a question is not covered at all: ask one focused question, say the detail is not available, or pass the chat to the team — never guess.

Which qualification questions should the AI ask?

An auto reply that sells also learns who it is talking to, so your team spends its time on people ready to buy. Pick two to four questions that decide whether and how you can serve this person, and let the AI ask them one at a time.

Answer before you ask — a price question gets the price, then a question — and save every answer on the lead, so the salesperson never asks twice.

When must the AI hand over to a person?

Hand-over is how an AI auto reply turns conversations into sales, and Meta’s policy requires an escalation path whenever you automate. Hand over when:

• the customer asks for a person, in any words;
• the question falls under a “never guess” rule;
• the customer is upset, complaining or repeating themselves;
• the lead is qualified and ready for the next step — a quote, a visit, a payment, a negotiation;
• the request is something the AI cannot do, such as cancelling or changing an order.

A good hand-over tells the customer honestly that a team member will continue, stops the AI in that chat so two voices never answer at once, and gives the person taking over a short summary: what the customer wants, their answers and anything already said about price. Without it, the customer repeats everything and you risk the sale. The chat should land with a named person, not a shared pile; our guide to a WhatsApp CRM for sales teams covers assignment, stages and follow-up.

How do you handle Arabic dialects and voice notes?

In Egypt and the Gulf, a WhatsApp AI chatbot that only answers in formal Arabic can sound like a government form. Customers write in Egyptian, Gulf or Levantine Arabic, in English, or in Arabic typed with Latin letters — sometimes in one message. Choose a voice:

• Match the customer — reply in the language and dialect they used; this suits most consumer businesses.
• One fixed dialect — Egyptian for a Cairo store, Gulf for a clinic in Riyadh or Dubai — for a consistent brand voice.
• Modern Standard Arabic — for law firms, financial services and B2B, where a formal tone builds trust.

Voice notes matter just as much: many customers would rather speak than type, and an auto reply that ignores them leaves those people waiting. Check that your tool transcribes the audio, test it with real voice notes in your customers’ dialects, and have the AI confirm critical details — a size, an address — before acting on them.

How do you test a WhatsApp AI auto reply before going live?

Test an AI agent like a new salesperson on probation, not with three friendly messages:

1. Use real chats. Send last month’s customer questions, awkward ones included, from a second phone exactly as they were written.
2. Try to break it. Ask for a discount, a product you do not sell, a medical opinion and something unrelated. Each should end in a correct refusal or a hand-over, never an invented answer.
3. Write the way customers do. Voice notes, Arabic in Latin letters, a product photo, English.
4. Test the hand-over. Ask for a person; check that the AI stops, the team sees the chat and the summary is accurate.
5. Fix the knowledge, not the reply. A wrong answer means a fact is missing or unclear.
6. Launch in stages. Start with after-hours chats, then new chats, then everything, reading conversations closely in the first weeks.

What mistakes make a WhatsApp auto reply lose sales?

• Letting it invent prices. If a price is not in the knowledge base, the AI must say so, not estimate. One invented price means an angry customer and a dispute.
• No hand-over. An AI that keeps answering a customer who wants a human loses the sale and ignores Meta’s escalation rule.
• No follow-up. Plenty of buyers ask a price and go quiet. One polite follow-up while the 24-hour window is open — “did you decide on the size?” — brings some back; after the window you need an approved template.
• Interrogating before answering. Four questions before the price feels like a trap.
• Walls of text. Short replies, one idea each, one question at the end.
• A stale knowledge base. New prices and sold-out products have to reach the AI.
• Pretending to be human. Meta’s policy tells businesses not to deceive or mislead people; let the assistant say it is the business’s virtual assistant.

How to set up a WhatsApp AI auto reply with Scale

Scale is an AI agent for WhatsApp, Instagram and Messenger with a built-in CRM. It connects your WhatsApp Business number through Meta’s official signup:

• Knowledge base. Guided sections you type in — business, offer, pricing, common questions, policies, objections, hours, locations — plus PDF, Word, Excel and CSV uploads. See how to train your AI employee.
• «Never Guess These». Topics the AI must not answer by itself; it says it needs to check and passes the question to your team. It also never invents information or prices and never answers outside your business.
• When it is not sure. You choose: ask one focused question, say it does not have that detail, or pass the chat to your team.
• Dialect and voice notes. Match the customer, or pick Egyptian, Gulf, Levantine, Jordanian, formal Arabic or English. Voice notes are transcribed and answered, and it can read a photo a customer sends. See the AI assistant settings.
• Qualification. You choose the questions; answers are saved on the lead, which moves through New Lead, Interested, Qualified, Appointment Booked and Customer, or Lost.
• One-tap takeover. The AI hands over when the customer asks for a person, and any team member can take over a chat with one tap; the AI stays quiet there until you give it back. Every lead carries an AI summary, an owner and its history.
• One follow-up. If a chat goes quiet while a question is open, the AI sends one follow-up inside the 24-hour window; after that, approved templates go out as campaigns.

What Scale does not do: it is not a general-purpose chatbot; it does not match a customer’s photo to your catalogue automatically; it does not decide discounts or exceptions; and it sends nothing free-form outside the 24-hour window. The Training screen has no test chat — test by messaging your number from another phone.

See how it works for an online store, a clinic or a real estate agency, or create your Scale account and connect your number.

Frequently asked questions

Can I use ChatGPT or another AI model as my WhatsApp auto reply?

Yes, to answer your own customers about your own business: Meta’s terms let a business work with an AI provider as its solution provider. What the terms have barred since 15 January 2026 is offering a general-purpose AI assistant as the main product over the WhatsApp Business Platform, outside the few markets where the law requires Meta to allow it.

Is the WhatsApp Business app’s away message enough?

For a small business that answers personally during working hours, often yes. An away message sends the same text to everyone, so it cannot quote a price or qualify a lead; when enquiries arrive after hours or faster than you can reply, an AI agent does that work.

Can a WhatsApp AI chatbot understand voice notes?

Only if the tool transcribes the audio before the AI reads it; away messages and most menu bots ignore voice notes. Test with real voice notes in your customers’ dialects before going live.

How many questions should the AI ask a new lead?

Two to four, asked one at a time and only after the AI has answered what the customer asked.

Should customers know they are talking to an AI?

We recommend it. Meta’s Business Messaging Policy tells businesses not to deceive or mislead people, and an assistant that says it is the business’s virtual assistant, with an easy way to reach a person, builds trust.

Can the AI message a customer after 24 hours?

Not with a free-form message. The 24-hour customer service window opens and resets each time the customer writes; outside it, WhatsApp allows only approved message templates.